The LeanScale Podcast · Episode 126

The Agents Are the Easy Part

Bastien Botella on the 55-agent Marketing OS that runs Fountain's marketing: collectors, orchestration and action layers, why the data layer is the hard part, sizing models to context, and why velocity is the real edge in AEO

Bastien Botella · Head of Growth & Marketing, Fountain · Fountain Hosted by Anthony Enrico
Published Updated 00:44:48 32 min read 6383 words
AI in GTM Demand Generation Revenue Operations Brand & Positioning
▸ Listen on Spotify Watch on YouTube Apple Podcasts
SummaryTakeawaysFrameworksQuotesAdviceAIEntitiesMetricsFAQTimelineTranscript
Executive Summary

The one-paragraph brief, extended

Why this conversation matters — and who should spend the hour.

Bastien Botella runs growth and marketing at Fountain, the platform behind high-volume frontline hiring, and picked up RevOps on top of it. He came to the job as an engineer. He co-founded a Paris startup, Clevy, sold it to Fountain three years ago, spent three or four months back in engineering, and was then pulled into marketing because the company needed systems built. What he has built since is Marketing OS: 55 agents that collect everything the company knows, analyze it, and hand his team scoped work every day. The system took about a month and a half to reach a solid MVP.

The episode opens with the result. One of the first tasks Marketing OS gave the team was to change a specific set of LinkedIn ads. The dashboards tracked CPC and CTR. The agents found something those metrics can't show: a correlation between ad format and messaging for one ICP and whether those leads went on to become pipeline. Acting on it saved more than $10,000 in the first month. "I never thought it would do that."

The architecture has three layers. Collector agents pull data from Salesforce, Gong, PostHog, Google Analytics, marketing spend and public market signals into a single data lake. Orchestration agents turn that data into recommendations. Action agents do the work, and whatever they ship flows back into the lake. Bastien's main point is that the action layer, the one people post about on LinkedIn, was the easiest part to build. The hard part was the data: listing every object and signal, documenting the join keys (some of which have to be constructed), and following a user from first website visit to revenue at a company that has been around for 14 years. Once that documentation existed, he says, the agents were exceptional at joining the data together.

Cost is managed by keeping context small. Only the collectors deal with large context and run on heavy models. Expert agents look at one account at one funnel stage, and QA agents check each number in a recommendation against the lake, so both run on leaner models. Anthony Enrico expects today's subsidized token pricing to end and says teams should plan for it. Bastien agrees, with a caveat: for now it is fine to use whatever works best, because velocity and speed to value matter most. Architecture decisions are the ones to think through early.

The second half covers what that speed buys. Fountain Watch, a signal engine the team built internally, picks up acquisitions, press releases and new regulations. Within a few hours of a relevant regulation going live, Marketing OS can recommend a page edit and a blog post, and the post can be drafted, approved by a human and published. That is Bastien's whole AEO strategy. He says nobody knows the exact recipe, but the faster he publishes on a signal, the higher Fountain ranks in ChatGPT. He also covers brand versus demand for a C-level audience, why a two-week sprint wastes opportunity, and the agent builder inside Marketing OS. His goal is an autonomous marketing organization where the only ceiling is how many outcomes the team can ask for.

Key Takeaways

14 things worth stealing

The load-bearing ideas, each with the business implication and who should care.

01

The agent that pays for itself looks past CPC and CTR to pipeline

Fountain's team already watched the usual LinkedIn ad KPIs: CPC, CTR and the other standard metrics. In the first weeks of Marketing OS, the agents found a correlation between the format and messaging of a specific set of ads for a specific ICP and whether the leads from those ads became pipeline. The resulting ad change saved more than $10,000 in the first month.

Why it matters: The value came from joining ad data to downstream pipeline outcomes, which a performance dashboard alone can't do. If your ad reporting stops at the click, agents built on the same data will stop there too.

Marketing LeadersRevOps LeadersRevenue Executives
02

Marketing OS has three layers: collect, orchestrate, act

Collector agents have one job: fetch data from Salesforce, Gong, PostHog, Google Analytics, marketing spend and external market signals, and put it all in one data lake. Orchestration agents interpret that data and turn it into actionable recommendations. Action agents take those recommendations and deliver the outcome.

Why it matters: It works as a feedback loop. Whatever the action agents put into production flows back into the data lake, and every layer is aimed at the same goal: building more pipeline.

Marketing LeadersRevOps LeadersFounders
03

The action layer gets the attention, and it was the easiest to build

Bastien calls the action agents very literal: they get a scope of work, take tickets and deliver them. Writing articles and coding landing pages were among the first things generative AI could do. Anthony notes that this is also the layer that gets the glory and the LinkedIn posts.

Why it matters: Teams that start with content-writing and page-building agents are starting with the easy part. The difficult work sits in the layers below.

Marketing LeadersFoundersRevenue Executives
04

The data layer is the hard part

Collecting data is only part of the job. It also has to be cleaned, correlated and put in one place so a user can be tracked from a website or landing-page visit down to becoming a customer in the revenue buckets. At a company that has been around for 14 years, reconciling marketing and sales data with market signals, Gong calls and marketing spend is where it gets hard.

Why it matters: Anthony repeats it for listeners: the outcome use cases are the easy part, and the unglamorous work happens in the data model layer. Budget time for this layer before promising results from the other two.

RevOps LeadersMarketing LeadersRevenue Executives
05

List and document every data source first, then let agents do the joins

The first step was to list every data source, object and signal, work out how each could be joined to the others, and document it all. Joining data by hand is hard because the shared ID isn't always obvious and sometimes has to be constructed. With that documentation in place as the source of truth, the agents turned out to be exceptional at joining data together.

Why it matters: Adding a new source is now routine: add it to the list of sources, types and objects, find a key to reconcile it with the rest, and the agents handle the rest. The documentation is what the system depends on.

RevOps LeadersMarketing Leaders
06

Expert agents are scoped to one account at one funnel stage

Inside the orchestrator, each expert agent has a single specialty. For ABM, that means an agent that is an expert on one account at one stage of the funnel, for example awareness. It combines that account's data with market signals to recommend a specific campaign, such as after a leadership change or a regulation the account is exposed to.

Why it matters: Narrow scope is what makes the recommendations specific enough to act on. It also keeps each agent's context small, which is what makes the system affordable.

Marketing LeadersSales LeadersRevOps Leaders
07

Size the model to the context, not the task's prestige

Of the 55 agents, only the collectors handle large context, and they run on heavy models. Expert agents run on mid-sized models, and QA agents run on lean models because their only job is to check that each number in a recommendation matches the data lake. Bastien says large context drives cost, latency and hallucination, so the architecture passes work between agents so that each one sees only what it needs.

Why it matters: Anthony calls this the clearest evidence of where the difficulty lies: expensive models on data collection and interpretation, cheap ones on the action and expert agents. Most teams do the reverse and run everything on max reasoning.

RevOps LeadersFoundersRevenue Executives
08

Use the best model now, but get the architecture right early

Anthony expects subsidized token pricing to end, possibly around an Anthropic IPO, and usage-based pricing to follow. Bastien partly disagrees: for now it is fine to use whatever is most effective, because velocity and speed to value are what matter and the goal is to be faster than competitors.

Why it matters: The exception is architecture. Marketing OS took about a month and a half to reach a solid MVP, and finding out afterwards that the architecture was wrong is expensive. Model choice can change later. The shape of the system is hard to change.

FoundersRevOps LeadersRevenue Executives
09

AEO has no recipe, so publish faster

After a year or two of meeting companies that specialize in AEO, Bastien concluded that nobody knows exactly how it works. What he does know is that the faster Fountain picks up a signal and publishes on it, the higher it ranks in ChatGPT. If one piece doesn't rank, the next one might.

Why it matters: He calls AEO a game of volume, of getting the right content out at the right time. Done relentlessly, it gets you cited both when buyers ask about your category and when anyone asks about the event you wrote about.

Marketing LeadersFounders
10

From signal to published page within a few hours

When Fountain Watch detects a new regulation affecting a customer segment, Marketing OS makes two recommendations: update the website page about regulations with a call to action, and write a blog post based only on facts, with no legal advice. Action agents draft the post, a human approves it, and it goes live within a few hours of the regulation.

Why it matters: This is the velocity argument in practice. A two-week or one-week sprint would mean publishing about a signal from a week earlier, which Bastien calls a cost of opportunity. Agents can also be proactive and flag the team when they spot an opportunity.

Marketing LeadersRevOps Leaders
11

Brand versus demand depends on who you sell to, and on what the data shows

Bastien won't give a universal split. Fountain sells to C-level buyers and the one or two levels below them, so being top of mind matters, and it invests a fair part of its budget in awareness. It does so because a large share of pipeline comes from inbound and pipeline grows when awareness spend grows. At his previous early-stage company, growth hacking drove 80 to 90 percent of pipeline and brand was not a game changer.

Why it matters: The brand budget follows what the data shows, not a belief that brand is good in principle. Stage and audience change the answer.

Marketing LeadersFoundersRevenue Executives
12

The goal is an autonomous marketing organization with no ceiling on output

The roadmap is more expert agents. Bastien wants to stop measuring the team by how much it can deliver and start measuring by how many outcomes and intents it can ask the system for. Because the work arrives well scoped, it can be done by an agent, a team member or a contractor brought in for a specific task.

Why it matters: He describes it as a change in mindset as well as operations: instead of needing a large workforce to carry out activities, you need people who choose the right outcomes to ask for and build the right agents to deliver them. The only ceiling is how much you ask for.

Marketing LeadersRevenue ExecutivesFounders
13

New agents come from an agent builder inside the system

Velocity also applies to the system itself. A team member with an idea for an agent describes it inside Marketing OS, for example an agent that watches for a certain signal and launches campaigns for a certain ICP, and the built-in agent builder creates it and puts it into production. This works because every agent shares the same taxonomy, prompt architecture and skills.

Why it matters: Anthony highlights what this removes: no collecting skills, data connectors and context by hand, and no handoff to rework something for production.

RevOps LeadersMarketing Leaders
14

An engineer's route into marketing

Bastien says he lacks creativity, relies on exceptional creative and copywriting people on his team, and is drawn to marketing because it runs on data. The more efficiently you find meaning in the data and act on it, the faster pipeline grows. For him, systems are a fast track to success, but quality still matters and the systems depend on high-quality content and visuals.

Why it matters: Anthony closes by predicting that many more engineers will move into marketing, since it has plenty of data to work with and plenty of systems to build.

Marketing LeadersFoundersRevenue Executives
Frameworks Discussed

9 named models

Every framework Jimmy names, defined and time-stamped.

Marketing OS: Collectors, Orchestration, Action

05:22

A three-layer agent architecture for marketing. Collector agents gather internal and external data into one data lake, orchestration agents turn it into actionable recommendations, and action agents deliver the outcomes. Results feed back into the lake.

Bastien designed it about six months before recording, after concluding that agents are powerful but need systems to feed them the right data. Collectors pull from Salesforce, Gong, PostHog, Google Analytics, marketing spend and market signals such as competitor and customer activity. The orchestration layer contains specialized expert agents. The action layer takes scoped tickets and delivers them. Everything is aimed at building more pipeline, and QA agents and an agent builder were added on top of the three core layers.

The Glamorous Layer Is the Easy One

07:52

In an agent system, the action layer (writing content, building landing pages) is the easiest to build. Collecting, cleaning and interpreting the data is the hard part.

Action agents are literal. They get a scope of work and deliver it, and content writing and page coding were among the first things generative AI could do. The hard part is collecting data and making sense of it: cleaning it, finding correlations, and following a user from first visit to revenue across 14 years of systems. Anthony states it for listeners: the thankless work happens in the data model layer.

List Every Source, Document the Keys, Let Agents Join

10:51

Before building agents, list every data source, object and signal, work out and document how each one joins to the rest, and treat that documentation as the source of truth the agents work from.

Joining data by hand is hard because the shared ID is often not obvious and sometimes has to be constructed. Bastien says the documentation was the key step: without it the team would still be trying to make sense of the data. Adding a source now means adding it to the list, finding a reconciliation key, and letting the agents, which are exceptional at joins, do the rest.

Expert Agents: One Account, One Funnel Stage

13:31

Orchestration agents each own a narrow specialty. For ABM, one agent is the expert on a single account at a single funnel stage and recommends campaigns for that account.

An agent responsible for one account in awareness uses Fountain Watch signals about that account, such as a leadership change or a regulation it is exposed to, to recommend a specific ABM campaign. Recommendations run daily and fall into three groups: web content (website and landing pages), lead generation (relaunch this campaign alongside this landing page) and nurturing (re-engage this segment now because this message will resonate today).

Size Models to Context

18:47

Keep most agents' context small and give heavy models only to the few agents that must handle large context, since large context drives cost, latency and hallucination.

In Marketing OS only the collector agents deal with large context and run on heavy models. Expert agents read a small part of the data lake and run on mid-sized models. QA agents only check that each figure in a recommendation matches the lake, so a lean model does the job with no noticeable drop in quality. Work passes from agent to agent so that each one sees only the context it needs to act on.

AEO Is a Velocity Game

37:07

Since nobody knows the exact recipe for ranking in AI answers, compete on speed: detect relevant signals, publish on them before anyone else, and keep doing it.

Fountain Watch detects a signal such as a new regulation. Marketing OS recommends a page update with a call to action and a factual blog post, action agents draft them, a human approves them, and they go live within hours. If one piece doesn't rank, the next might. Bastien doesn't know the right phrasing or title structure. He knows that publishing faster ranks Fountain higher in ChatGPT, and says speed also helps with SEO and with timing ABM and awareness campaigns ahead of competitors.

Autonomous Marketing: The Ceiling Is What You Ask For

23:44

Measure a marketing organization by how many outcomes and intents it can ask its system for, not by how much work it can deliver. With the work scoped well, delivery has no ceiling.

Because Marketing OS scopes each piece of work, it can go to an agent, a team member or a contractor hired for a specific task. The team's job becomes choosing the right outcomes to ask for and building the right agents to deliver them, rather than adding workforce to handle more activities.

Daily Sprints Instead of Two-Week Sprints

39:41

Act on a signal the day it appears. Waiting for the next one- or two-week sprint to act on something found a week ago is a cost of opportunity.

Marketing OS analyzes data and hands the team scoped work every day, and agents flag Bastien, his marketing manager or a team member as soon as they see a pipeline opportunity. He calls velocity central to how marketing works and says it is something Fountain lives by as a company.

Build Agents Inside the System

42:26

An agent builder inside the operating system creates new agents from a plain-language description. It works because every agent shares the same taxonomy, prompt architecture and skills.

Instead of building an agent separately and then fitting it into Marketing OS, a team member describes the use case, for example an agent that watches for a certain signal and launches campaigns for a certain ICP, and the system builds it and puts it into production. Anthony notes this removes the handoff and the rework usually needed to make an agent production-grade.

Best Quotes

19 lines worth clipping

Pulled verbatim. Copy or share any of them.

“So that correlation that agents found is a correlation that made us save over $10,000 on month one, which is crazy. I never thought it would do that.”
Bastien Botella 03:22
“About six months ago, we found that agents are super powerful, but we need systems to make sure that we can feed them the right data.”
Bastien Botella 04:49
“It's like a perfect feedback loop. Once the action agents deploy something to production, then we get the data again back into the data lake and everything starts over. And everything is really built just for one single goal, just build more pipeline.”
Bastien Botella 07:14
“I think the one that is oddly the easiest to build, I would say, is the action layer.”
Bastien Botella 07:52
“They get all the glory of everything that happened and that's what people are posting on LinkedIn or talking about.”
Anthony Enrico 08:15
“And that's actually the easy part. Those agents, they're just doing stuff all day long. They are just taking tickets and just delivering the tickets.”
Bastien Botella 08:24
“The agents are exceptional at joining data together.”
Bastien Botella 12:33
“All the outcome stuff, all the fun use cases you see is the easy part. The thankless, roll-up-your-sleeves hard work happens here in this data model layer.”
Anthony Enrico 12:55
“What really triggers the costs, and what also takes time, is the massive context.”
Bastien Botella 19:04
“I think we're enjoying some subsidized token usage right now.”
Anthony Enrico 21:34
“I'm going to contradict myself here, but I think it's perfectly fine to just roll with whatever is the most effective for now. Because velocity and speed to value is really the coin.”
Bastien Botella 22:14
“There is no ceiling, again, no ceiling in terms of activities we can deliver. The only ceiling is how much can we ask for?”
Bastien Botella 25:26
“We learn that the more we invest, the more pipeline we get.”
Bastien Botella 32:18
“The one learning that I got is that there is no recipe. Nobody knows exactly how it works. No magic words.”
Bastien Botella 33:53
“AEO is a game of volume. AEO is a game of getting the right contents at the right time.”
Bastien Botella 37:07
“What I know is that the faster I am picking up the signals and publishing, the higher I am in ChatGPT ranking. And that's just good enough for me.”
Bastien Botella 37:52
“That is the first time I have heard the importance of velocity and it's making a ton of sense.”
Anthony Enrico 38:08
“I don't want to have to wait two weeks until the next sprint, or a week, in order to consider making an article about a signal that was found a week ago. I think it's a cost of opportunity.”
Bastien Botella 39:47
“MarketingOS is actually built to build agents.”
Bastien Botella 42:26
Practical Advice

What should you actually do?

The playbook, split by the seat you sit in.

Marketing Leaders

  • Judge ad formats and messages by the pipeline they produce, not only by CPC and CTR. Join ad data to downstream opportunities so the correlation can be found.
  • Put the data layer first. Content-writing and page-building agents are the easy part and depend on the data underneath them.
  • Scope ABM agents narrowly, for example one account at one funnel stage, and feed them public signals such as leadership changes, press releases and regulations.
  • Treat AEO as a speed problem: detect a relevant signal, publish factual content and update the relevant page within hours, and keep doing it.
  • Keep a human approval step before agent-written content goes live.
  • Base the brand-versus-demand split on your own data. If pipeline rises with awareness spend, that is the case for the budget.

RevOps Leaders

  • Start with an inventory: list every data source, object and signal, and document the key that joins each one to the rest, building a key where none exists.
  • Make that documentation the source of truth agents work from. Adding a source means adding an entry and a reconciliation key.
  • Build the path from first website visit to closed revenue across marketing, sales, call-recording and spend data before asking agents for insights.
  • Tier models by context size: heavy models only where context is large, leaner models for expert and QA agents with narrow scope.
  • Add QA agents whose only job is to check every figure in a recommendation against the data lake.
  • Standardize taxonomy, prompt architecture and shared skills across agents so new ones can be generated rather than hand-built.

Founders

  • Plan the architecture of an agent system carefully even if you move fast on everything else. Rebuilding it after the MVP is costly.
  • Use the most effective models for now if speed to value is the priority, and expect token pricing to rise.
  • Consider putting an engineer in marketing. Systems thinking suits a function that runs on data, as long as creative quality is covered by others.
  • At an early stage, expect growth-hacking activities rather than brand to drive most pipeline. Revisit the mix as the audience moves up to the C-suite.

Revenue Executives

  • Measure the marketing team by the outcomes it can ask its system for, not the activities it can staff.
  • Replace two-week planning cycles with daily, scoped work where signals are time-sensitive.
  • Expect well-scoped work to go to agents, team members or specialist contractors, and plan headcount with that in mind.
  • Ask for evidence tied to pipeline, like the LinkedIn ad correction, before trusting agent output on performance claims.
AI Takeaways

How AI actually changes GTM

LeanScale's signature read on the AI-in-GTM question this episode wrestles with.

The thesis

The outputs of an agent-run marketing function, such as articles, landing pages and campaigns, are the easy part. What makes Fountain's Marketing OS work is a documented, joined data layer, agents with narrow scope and small context, models sized to that context, and a daily operating rhythm that turns speed into an advantage, especially in AEO.

Agent & automation ideas

  • A collector agent that inventories every data source and object and suggests the join key for each new source added to the documentation.
  • A correlation agent that ties ad format and messaging for each ICP to downstream pipeline rather than CPC and CTR.
  • Per-account, per-stage ABM expert agents that watch public signals such as leadership changes, acquisitions, press releases and regulations, and recommend a specific campaign.
  • A regulation-watch agent that drafts a factual blog post and a page update with a call to action for human approval within hours of a new rule.
  • A QA agent that rejects any recommendation whose figures don't match the data lake.
  • An agent builder that generates new agents from a plain-language use case using a shared taxonomy, prompt architecture and skill library.
Operations Takeaways

By function

The same conversation, filtered for RevOps, pipeline/marketing ops, and customer ops.

Revenue Operations

  • .
  • .
  • .
  • .
  • .

Pipeline & Marketing Ops

  • .
  • .
  • .
  • .
  • .
Metrics Mentioned

The numbers, with context

$10,000+ in month one
Savings from the LinkedIn ad correlation

Found by Marketing OS agents by correlating ad format and messaging for a specific ICP with downstream pipeline, which CPC and CTR tracking had not shown.

55
Agents in production

The number of agents running in Marketing OS at recording, grouped into roughly seven to eight categories including orchestration, collectors and QA.

7–8
Agent categories

The rough number of groups the 55 agents fall into.

About a month and a half
Time to a solid MVP

How long Marketing OS took to build to a solid MVP, which Bastien gives as the reason to get the architecture right from the start.

About six months before recording
When the Marketing OS work began

When the team concluded that agents need systems to feed them the right data and started designing the operating system.

14 years
Fountain's age

Bastien's reason why following a user from first visit to revenue across systems is hard.

3 years
Time since Fountain acquired Clevy

Bastien has worked in marketing at Fountain for about three years, after three to four months back in engineering following the acquisition.

80–90%
Pipeline from growth hacking at his previous startup

At Clevy's early stage, small growth activities rather than brand and awareness produced most pipeline. At Fountain, a much larger share comes from awareness.

Within a few hours
Signal to published content

From a relevant regulation going live to an approved blog post and an updated website page with a call to action.

Entities

Companies, people & tools mentioned

Auto-extracted and linked into the knowledge graph.

Companies

FountainHR Tech / Frontline Hiring

Bastien's employer, where he runs growth and marketing and picked up RevOps on top of it. The platform behind high-volume frontline hiring, which he describes as about 14 years old. It sells AI and agentic solutions to its customers and sells to C-level buyers and the one or two levels below them. It acquired Bastien's startup, Clevy, three years ago, and its marketing now runs on the 55-agent Marketing OS.

00:11 · 04:35 · 09:52 · 27:10 · 31:10 · 40:41Company →
ClevyConversational AI

The startup Bastien co-founded with two other co-founders, one as CEO and one as CTO, and sold to Fountain three years ago. The founders were as much engineers as business people, and moving fast led Bastien to apply engineering methods to marketing there. At that early stage, growth-hacking activities drove 80 to 90 percent of pipeline.

26:23 · 27:10 · 32:27Company →
AnthropicFrontier AI

Mentioned by Anthony, who guesses that today's subsidized token usage will end around an Anthropic IPO and give way to more usage-based pricing, with fewer flat-rate max accounts.

21:34 · 21:39Company →
LeanScaleGTM Operations

Anthony's company. He raises LeanScale's own brand-versus-demand debate when asking Bastien how to split investment between the two.

29:56Company →

People

Bastien Botellaguest

Head of Growth & Marketing, FountainProfile →
Anthony Enricohost

Co-founder of LeanScale and host of The LeanScale Podcast.

Co-FounderProfile →

Tools & software

LinkedInSocial Platform

Fountain runs many LinkedIn ads. The first task Marketing OS assigned was to change a specific set of them after agents correlated ad format and messaging for one ICP with downstream pipeline, which saved more than $10,000 in month one. Anthony also notes that LinkedIn is where action-layer agents get talked about.

Website ↗
SalesforceCRM

One of the core internal sources the collector agents pull into Marketing OS's data lake.

Website ↗
GongRevenue Intelligence

Call data collected into the data lake. Reconciling Gong calls with marketing, sales, spend and market-signal data is where Bastien says the data work gets hard.

Website ↗
PostHogProduct Analytics

Named among the marketing and product-analytics tools the collector agents pull data from.

Website ↗
Google AnalyticsWeb Analytics

Named among the marketing tools the collector agents pull data from, alongside PostHog and marketing spend.

Website ↗
ChatGPTAI Assistant

The AI answer engine Bastien uses to judge AEO results: the faster Fountain picks up signals and publishes, the higher it ranks in ChatGPT, and ChatGPT will cite the article when people ask about the regulation it covers.

Website ↗
ClaudeAI Assistant

Named alongside ChatGPT as an AI assistant where Fountain's fast-published regulation content can rank.

Website ↗
Methodologies referenced Data-source inventory and join-key documentationFountain Watch signal engineDaily recommendations in three groupsQA agents that check the numbersSignal-to-publish with a human in the loopMixed delivery of scoped work
Related LeanScale Services

Where this shows up in the work

AI for GTM

Agents and workflows that multiply operator throughput — the 'make humans superhuman' thesis, operationalized.

Marketing OS is an AI go-to-market build: collector, orchestration, action and QA agents, model tiering by context, and a signal-to-publish AEO workflow with a human approval step.

Fractional RevOps

Embedded, field-facing RevOps leadership and execution — the operator model Jimmy describes, delivered as a service.

The hard part Bastien describes is RevOps work: inventorying every data source, documenting join keys, and connecting first website touch to revenue across CRM, call and spend data.

Frequently Asked Questions

Straight answers

Generated from the conversation, marked up for search and AI extraction.

What is Fountain's Marketing OS?

Marketing OS is the agent system that runs Fountain's marketing, built by Head of Growth & Marketing Bastien Botella. It has 55 agents across three main layers. Collector agents pull data from Salesforce, Gong, PostHog, Google Analytics, marketing spend and public market signals into one data lake. Orchestration agents, each an expert in something narrow, turn that data into recommendations. Action agents deliver the work. Results flow back into the data lake, and everything is aimed at building more pipeline. QA agents and an agent builder sit on top. It took about a month and a half to reach a solid MVP and gives the team scoped work every day.

Which part of an AI agent system for marketing is hardest to build?

According to Bastien Botella, who built a 55-agent marketing system at Fountain, the data layer is the hardest and the action layer the easiest. Action agents that write articles or code landing pages are literal: they take a scoped ticket and deliver it, and generative AI has done that for a while. The hard work is collecting data and making sense of it: cleaning it, finding correlations, and following a user from a first website visit to revenue across marketing, sales, call-recording, spend and market-signal data. At a 14-year-old company that means listing every source and documenting how each one joins to the rest before the agents can do anything useful.

How do you keep a large multi-agent system affordable?

Keep each agent's context small and match the model to that context. In Fountain's 55-agent Marketing OS, only the collector agents handle large context, so only they run on heavy models. Expert agents read a small part of the data lake and use mid-sized models. QA agents only check that every number in a recommendation matches the data, so a lean model works with no noticeable loss of quality. Bastien Botella says large context is what drives cost, slows agents down and makes them hallucinate, so the system passes work from one agent to the next and each one sees only what it needs.

How can a B2B company improve its AEO ranking in ChatGPT?

Bastien Botella's answer after a year or two of meeting AEO specialists is that there is no known recipe, so speed is the advantage. Fountain uses an internal signal engine, Fountain Watch, to spot events such as a new regulation affecting a customer segment. Its agent system then recommends a website page update with a call to action and a factual blog post. Agents draft them, a human approves them, and they go live within a few hours. If one piece doesn't rank, the next might. He calls AEO a game of volume and of getting the right content out at the right time, and says the faster Fountain publishes on a signal, the higher it ranks in ChatGPT, both for category questions and for questions about the event itself.

How should a company split marketing budget between brand and demand?

There is no universal ratio, according to Fountain's Bastien Botella. It depends on company strategy and audience. Fountain sells to C-level executives and the one or two levels below them, so staying top of mind matters, and it puts a fair part of its budget into awareness. It does so because the data supports it: a large share of pipeline comes from inbound, and pipeline grows when awareness spend grows. At his previous early-stage startup, brand was not a game changer, and small growth-hacking activities produced 80 to 90 percent of pipeline.

Why run marketing on daily sprints instead of two-week sprints?

Because signals lose value quickly. Bastien Botella says that waiting one or two weeks for the next sprint to act on something found a week earlier is a cost of opportunity. Fountain's Marketing OS analyzes data and hands the team scoped work every day, and its agents proactively alert team members when they see something that could grow pipeline. The same speed applies to AEO, SEO and to timing ABM and awareness campaigns ahead of competitors.

What does an autonomous marketing organization mean?

For Bastien Botella, it means a marketing team whose output has no ceiling, because delivery no longer depends on headcount. Fountain's Marketing OS scopes each piece of work well enough that it can go to an agent, a team member or a contractor hired for a specific task. The team's job becomes choosing the right outcomes and intents to ask the system for and building the right agents to deliver them. The only remaining ceiling, he says, is how much the team can ask for.

Timeline

Jump through the conversation

Click any chapter to open it in the transcript.

00:00Cold open + intro 01:35The receipt: a LinkedIn ad find worth $10K in month one 04:07Collectors, orchestration, action: the three layers of Marketing OS 07:33Which layer was hardest to build? 10:25List every source, document the keys 13:21Expert agents and Fountain Watch: mining signals from the public internet 16:5955 agents, and why context size drives everything 23:25What's next: an autonomous marketing organization 25:50From engineering to founding Clevy to Fountain 29:56Brand versus demand 32:59AEO: there is no recipe and nobody has one 38:08Why velocity beats the two-week sprint 41:32The agent builder inside Marketing OS 43:28Wrap
Full Transcript

The whole conversation

Broken into chapters, searchable, verbatim from the audio. Speakers inferred (not diarized).

Transcript as markdown ↓ Whole episode as markdown ↓
00:00Cold open + intro

0:00 And we get the data again, back into the data lake, and everything starts on ARR.

0:06 And everything is really built just for one single goal, just build more pipeline.

0:11 My guest today is Bastian Botella, who runs growth and marketing at Fountain, the platform

0:16 behind high volume frontline hiring for companies like El Pollo Loco, Bojangles, and UPS.

0:23 And so listing all the data, trying to understand how we can join all this data together, and

0:29 just documenting, has been the first step that we took.

0:35 And that has been the key step.

0:36 Without that, we would be used to trying to make sense of all this data.

0:39 My hunch or assumption is we're enjoying a lot of these models that are being subsidized.

0:48 A lot of teams are using lots and lots of tokens.

0:52 And I'm going to contradict myself here, but I think it's perfectly fine to just roll with

0:56 whatever is the most effective, because velocity and speed to value is really the coin, right?

1:07 That is the first time I have heard the importance of velocity, and it's making a ton of sense.

1:15 Everyone's talking about volume.

1:17 Everyone's talking about the hacks.

1:18 I don't want to have to wait two weeks until the next sprints.

1:21 I think it's a cost of opportunity.

01:35The receipt: a LinkedIn ad find worth $10K in month one

1:35 Bastian, let's start with a receipt.

1:37 Your agents went through all of your LinkedIn ad data and came back with something about

1:42 images versus carousels that your team hadn't caught.

1:46 And it moved real money in the first month.

1:50 Take me through everything that happened there.

1:52 Hey, Han, from the first of all, thank you so much for having me.

1:56 Super excited to be here.

1:59 And also just excited to talk about marketing OS.

2:02 And I think what you just mentioned here, these ads that we changed is really what shows

2:10 the power of marketing OS.

2:12 So let me just focus through what happened, actually.

2:17 So when we launched marketing OS, and I'm sure we'll talk more about the actual operating

2:22 system in a minute, but when we launched it, one of the very first tasks that was assigned

2:29 to my team by the operating system was to just tweak some of our LinkedIn ads.

2:38 We run a bunch of LinkedIn ads.

2:40 We run quite a lot of them, and obviously my team is checking CPC, CTR, you know, like

2:47 all the YouTube suspects, all the KPIs that we want to track in order to understand performance.

2:53 What's the agents that's leading to inside marketing OS form was not so much just like

3:00 a pure performance indicator.

3:03 It was a correlation between the format and messaging of our ads, or actually a very specific

3:10 set of ads for a very specific ACP, and how those leads that we get from those ads will

3:19 eventually correlate into pipeline.

3:22 So that correlation that agents found is a correlation that's made us save over $10,000

3:32 on month one, which is crazy.

3:34 I never thought it would do that.

3:37 So we were actually the first one to be surprised by the flow of the system.

3:41 Yeah, and I think some of the capabilities are really starting to get mind blowing, but

3:47 it takes a lot of orchestration.

3:50 So I think some people assume, hey, these models are just getting so powerful, they can start

3:55 to do things on their own.

3:56 Yes, to an extent, they're getting much better at reasoning.

4:02 But you also have to build quite a bit of infrastructure to get the right context in

4:06 play.

04:07Collectors, orchestration, action: the three layers of Marketing OS

4:07 And I would love if you could maybe just walk through what some of the biggest pieces are

4:13 and maybe some things that people who are trying to build out AI programs in the marketing

4:19 space are probably missing.

4:22 Yeah, absolutely.

4:24 So we've been working with AI and agents for quite a while, for months and months and months

4:32 in marketing, but also with our own products.

4:35 We sell AI and we sell agentic solution to our customers.

4:41 And so we've been really at the forefront of this agentic journey from the start.

4:49 And about six months ago, we found that agents are super powerful, but we need some sort

4:57 of garbage and we need systems to make sure that we can feed them the right data.

5:05 And so based on this sort of insight that we got, we started thinking about what would

5:16 the perfect operating system, what would it look like?

5:22 And we came to the conclusion that there were like three big pieces to the system.

5:28 And we built the system from those big three pieces and then we added more on top.

5:33 But the big three pieces were a set of what we call collector agents and those agents

5:40 have only one job, it's to go and fetch data, just collect data everywhere from Salesforce,

5:47 from Gong, from all our marketing tools.

5:52 So post-hoc, Google Analytics, all our marketing spans, so all our internal tools, but also

6:02 external data, market signals, for instance, what our competitors do or what our customers

6:09 do, right?

6:10 So all that is put into one data lake and that's done by our collector agents, the set

6:17 of agents that only job is to collect data and make sense of it.

6:23 The second piece of the framework of the operating system was the orchestration, now that we

6:29 have the data, we need to process it.

6:31 So we have a very wide range of agents that only job is to get the data, make sense of

6:39 it, understand how they can transform it, translate that into actionable items that

6:46 will eventually lead to the outcome that we want.

6:51 And then we have a third layer, that is the actionable agents that will take those insights

6:58 from the orchestration layer and just deliver the outcome.

7:03 And so those three layers, the signal correction, the orchestration that we come up with, all

7:09 the recommendations, and the action agents are the three big layers, that's for marketing

7:13 OS.

7:14 And it's like a perfect feedback loop, when once the action agents deploy something production,

7:22 then we get the data again back into the data lake and everything starts over.

7:28 And everything is really built just for one single goal, just build more pipeline.

07:33Which layer was hardest to build?

7:33 Which of those three components was the most difficult to build and may relay into most

7:42 difficult to maintain as well, or they might be different?

7:48 They are three very different layers, very different layers.

7:52 I think the one that is oddly the easiest to build, I would say, is the action layer.

8:03 Those are very literal agents.

8:07 They get a scope of work and they just deliver the work.

8:13 They get all of the attention though, right?

8:15 They get all the glory of everything that happened and that's what people are posting

8:19 on LinkedIn or talking about.

8:22 Exactly.

8:24 And that's actually the easy part.

8:27 Those guys, those agents, they're just doing stuff all day long.

8:31 They are just taking tickets and just delivering the tickets.

8:35 And these, I think people have been, we've been building that for quite a while now.

8:39 It was like the first really use case that came out of like the chat and the cheese and

8:43 the clothes, you know, like building content, for instance, like writing articles.

8:48 Those were the first use cases that were found.

8:51 Same goes for coding.

8:52 Coding a landing page has been something that we've been able to do for quite a while now.

8:57 So that piece was actually pretty easy.

9:00 I think that the data collection piece was actually pretty hard because it's not only

9:06 about collecting the data, it's also about making sense of it.

9:12 Like making sense of the data, cleaning the data, finding correlation in this data, right?

9:20 Putting all this data into one single place so agents can make sense of the data and see

9:28 that the user and just track a user from like the awareness level, like the user visited

9:34 the website or landing page down to the user assigned and is now into the revenue buckets,

9:40 right?

9:41 And it's a customer.

9:43 And there are like a bunch of steps in between.

9:46 And that's actually not always very easy to track, at least not for a company that have

9:52 been around for like 10 to 15 years, you know, it's a fountain has been around for 14 years.

9:57 And, and, and it's, and it's hard to, to, to track everything like we are doing a pretty

10:03 good job at tracking like anything from marketing and sales, the reconciliation, all these data

10:10 with market signals and with like all the gun calls.

10:13 And that's when things start to get, to, to get, to get hard and marketing spend as well.

10:20 Agents can do that.

10:22 And and I think that's key.

10:25List every source, document the keys

10:25 It's been your approach to help the agents make the most sense out of the data.

10:30 Cause I agree.

10:31 I think that's, that's the hairiest part.

10:32 It's like, what's the context of this data that the agent's looking at?

10:38 What approach have you taken both?

10:40 Maybe get as practical as like tools that you've had some of this context in or, or processes

10:45 that you follow on a daily, weekly basis.

10:48 How do you even approach it?

10:51 The first thing that we've done when we started that was just released everything once.

10:57 We just started listing every single data that we have, every single object, every single

11:02 signal, everything.

11:04 And then when we, when we worked out doing that, we try to understand like, what are

11:09 the ideas?

11:10 What, what things can we feed into the agents who agents can join data together?

11:17 Joining data is really hard.

11:19 It's really hard to do it manually.

11:21 It's really hard to build reports with like several sources of data that are joined across

11:26 one single AD because sometimes the AD is not like super obvious.

11:31 And sometimes the AD needs to be sort of built.

11:33 It's not like just a very obvious AD.

11:37 And so, and so listing all the data, trying to understand how we can join all these data

11:43 together and just documenting has been the first step that we took.

11:50 And that has been the key step.

11:52 Without that, we would be still trying to make sense of all this data, but being able

11:56 to just list everything and have a source of truth where we have all the types of data,

12:03 all the keys, and just having agents just making sense of all these data based, based

12:10 on these documentation is what led us to success.

12:13 And so today, when we, when we add more data, what we do is we do the same, we just add

12:19 the entry to the list of data sources, data types, data objects, try to understand how

12:25 we can find a key that we can use to reconciliation with and join with the rest of the data, and

12:32 then the agents to the rest.

12:33 The agents are exceptional at joining data together.

12:40 So now that you have the hard part in place, you built the infrastructure, you have clean

12:44 data, it's joined properly, you have context between the data, you have stuff that these

12:48 outcome agents can actually operate on.

12:50 I'm going to highlight again, for anyone listening, this is the hard part.

12:55 All the outcome stuff, all the fun use cases you see is the easy part, the thankless roll

13:01 up your sleeves, hard work happens here in this data model layer.

13:05 But once you have that in place, we'd love to hear some of the things that your agents

13:11 were able to do.

13:12 I know we talked about the LinkedIn ad example, but I know there's so many things you're deploying

13:17 these against.

13:18 What are a few ideas that highlight the art of the possible?

13:21Expert agents and Fountain Watch: mining signals from the public internet

13:21 Yeah, we have so many use cases of so many recommendations that come out of marketing

13:29 OS.

13:31 The way that we build marketing OS, and the way that the orchestrator works, is we build

13:43 what we call expert agents within the orchestrator.

13:47 The orchestrator just runs a bunch of agents that each have a specialty.

13:52 They each are like super experts into something very specific.

13:59 For instance, for ABM agents, we will have agents that are experts in one account.

14:08 They're not only one account, they are experts in one account, and one accounts with a very

14:15 specific state of the funnel.

14:17 For instance, they would be experts of one single account for awareness, and they will

14:23 just focus on finding recommendations for this account in awareness.

14:27 Today, we'll use the market signals that we source.

14:31 By the way, we have this tool at Fonten called FontenWatch.

14:36 It's something we built internally, and it's like a signal engine that will just go and

14:43 fetch all sorts of signals from the web, from the public internet.

14:48 It will fetch the acquisitions that have been made lately.

14:51 It will fetch all the press releases.

14:53 It will fetch all the regulations that came out.

14:57 All this stuff is also gathered into our data lake, into marketing OS with FontenWatch,

15:05 and so those experts agents will use all of the data that sometimes talk about the very

15:11 specific accounts, and just so happen that it sometimes is the account that they are

15:16 in charge of, and they can leverage this data to tell us, "Hey, you actually need to launch

15:21 this very specific ABM campaign," because we know that there was a change of leadership

15:27 in that company, or we know that such regulation came out, and they are very exposed to this

15:34 regulation.

15:36 We launch our campaigns based on the recommendations that we get from those expert agents, based

15:42 on all the data that we have in our data lake, and that just runs every day.

15:52 We get all sorts of recommendations, but they are usually around, I would say, three big

15:57 topics.

15:58 The first one is website contents.

16:01 I want to say website, I'm talking about website as well as landing pages, just web contents.

16:09 Web content is one, the land generation is another one, so agents are able to give us

16:20 recommendations and tell us, "Hey, you should re-launch that campaign alongside that landing

16:25 page," and then the third is nurturing.

16:30 We are very good at telling us, "Hey, you should re-launch that specific segments,"

16:36 because that messaging that you can share with them is going to resonate today.

16:43 Maybe not tomorrow, but at least today it will.

16:49 I think it's really impressive that the orchestration can lead all the way to those outcomes, and

16:54 then just produce something that's so relevant, context-driven.

16:5955 agents, and why context size drives everything

16:59 One thing I want to get into is, my hunch or assumption is we're enjoying a lot of these

17:10 models that are being subsidized.

17:13 A lot of teams are using lots and lots of tokens, putting lots of model power behind

17:19 some of the tasks they're trying to complete or get done.

17:24 We've built a really interesting model that I think is ahead of how most teams are using.

17:30 I think most teams pick max account, max reasoning, just run everything through this.

17:36 How have you set up the infrastructure that is really leveraging a lot of AI power to

17:43 also operate as lean as possible?

17:46 That's a great question.

17:49 That's a question that we ask ourselves right in the start, because we know that setting

17:56 up the wrong architecture and running with the wrong models can be very costly.

18:03 We made a few tests.

18:05 It took us some time to get it right.

18:07 Even today, we keep iterating.

18:10 We keep moving from one model to the next, also because models are evolving.

18:15 We keep moving from one model to another, and we are still trying to fine-tune every

18:20 single agent, because today we have 55 agents running in marketing hours.

18:29 Those 55 agents, they can be classified into around seven to eight buckets.

18:37 The orchestration, the collector agents, the QA agents, we have agents that do QA for us,

18:45 so they're like loads of different agents.

18:47 I would say that the key for us in the way that we build the architecture is that we

18:56 didn't want an agent to have to look into a huge context.

19:04 That wasn't really the plan, because what really triggers the costs and what also takes

19:12 time and what also woke agents into a health and aging is the massive context.

19:21 We wanted to make sure that only certain agents deal with big contexts, but most agents would

19:29 deal with very, very, very specific data.

19:34 Today only the collector agents deal with big contexts, and those ones have some pretty

19:41 heavy models, but all the expert agents, all the action agents, actually most agents, even

19:49 the QA agents, they actually run on pretty lean model, and mostly because the context

19:54 they have is not as big.

19:57 The difference in quality between Fable 5 and Sony 3 for QA, for instance, is not something

20:06 ... It's not huge.

20:09 It's not something that we would feel.

20:12 The QA is still kind of on the same, because the QA agents has a very, very, very small

20:19 scope.

20:20 It just looks at the recommendation, and we check that every single number, every single

20:25 figure in that recommendation checks out with the data lake, and that's it.

20:31 For the expert agents, the expert agents will only fetch some of the data in the very small

20:38 portion of the data, of the data lake.

20:40 Same again.

20:41 They have a very small scope.

20:42 They have a bigger model than the QA agents, but still smaller than the data collector.

20:51 The key for us was the context, and we worked really hard to actually build several layers

20:56 and have a process where agents just hand off work to the next one.

21:02 Every single agent has the scope, the context for which he only needs to take an action

21:07 on.

21:10 If anything highlights which part of the problem is more difficult more than that, I think

21:14 that's the best example of using the lowest cost model on the action outcome expert agents,

21:21 using the high cost models on getting the data collection and actually making sense

21:26 of it.

21:28 It makes a ton of sense.

21:29 I just think a lot of teams, maybe we have another three months, six months, whatever

21:34 that timeline is, I think we're enjoying some subsidized token usage right now.

21:39 I'm assuming it's going to line up maybe around when Anthropic does their IPO, that they're

21:44 going to pull the rug, and then all of a sudden, it's going to be more usage-based pricing.

21:48 We're going to get rid of these Cloudmax accounts and things like that.

21:51 But I think it's just important for teams to be thoughtful and mindful of how they're

21:58 organizing which model to the agents that they're out there performing.

22:03 Yeah, for sure.

22:06 And I think it's good to at least think about that quite early.

22:14 I think it's perfectly fine to-- and I'm going to contradict myself here, but I think it's

22:19 perfectly fine to just roll with whatever is the most effective for now.

22:27 Because velocity and speed to value is really the coin.

22:34 That's what we all want.

22:37 We want to be faster than our competitors.

22:41 And so I think speed is what matters most.

22:47 But I think it's useful, and it's very helpful to think about when you build a large architecture,

22:57 to think about all those topics right on the stars.

23:00 Because when you build a large architecture, it doesn't cost you too much to actually think

23:04 it through and think about the future.

23:07 Because that operating system, for instance, took us like a month and a half to have a

23:14 solid MVP.

23:16 So by that time, if you figure out that you messed up the architecture, it's unfortunate.

23:24 Totally agree.

23:25What's next: an autonomous marketing organization

23:25 What's next for MarketingOS?

23:27 What are things that you want to add to the system?

23:30 Things that maybe you're hoping you can accomplish in the near future?

23:35 What's on the roadmap?

23:38 Just more agents, more expert agents.

23:43 Think about it.

23:44 Now we are in a place where-- I want marketing to be autonomous.

23:53 I want to have an autonomous marketing organization.

23:58 What does that mean?

23:59 It means that we have no ceiling on how much we can deliver.

24:08 So I want to be able to move the performance from how much we can deliver to how many outcomes,

24:18 how many intents we can ask the system.

24:23 So my goal today is just to create more agents that are super great at finding activities,

24:30 finding recommendations that will move toward that goal of expanding pipeline.

24:35 Because each agent is going to work toward that goal.

24:38 How we get there, how we actually get the work done, that's something that we can figure

24:43 out pretty easily.

24:45 Because agents can do part of the work.

24:47 Team members can do part of the work.

24:49 We can also use-- we can also hire contractors to do some very specific tasks, because everything

24:56 is so well-scoped.

24:58 So it's really like a shift, not only in operation.

25:02 It's also a shift of mindsets, where we shift from thinking, OK, I need loads of work force

25:08 to work on a bunch of activities to deliver to a mindset where you need people to just

25:18 find the right intents, find the right outcome to ask, and build the right agents that we

25:25 deliver outcomes.

25:26 There is no ceiling, again, no ceiling in terms of activities we can deliver.

25:35 The only ceiling is how much can we ask for?

25:39 Well, once you've made that investment infrastructure, I'm just excited to see what the velocity

25:45 is that your team comes up with, and what the new agents you build are.

25:50From engineering to founding Clevy to Fountain

25:50 I would love to dive into a little bit-- you have an engineering background, founding company

25:58 background.

25:59 Now you're running marketing, and RevOps, and all these other things.

26:03 But what was the through line, and how did you get to what you're doing today?

26:09 My life has been quite right.

26:15 So yeah, I started in engineering, and I love engineering, for what it's worth.

26:20 I absolutely love it.

26:23 I co-founded my company that we sold three years ago to found them.

26:28 I co-founded it with two other great co-founders.

26:31 One was a CEO of that company, clearly, and one was the CTO.

26:37 And so I kind of organically moved to marketing, because the three founders are as much engineers

26:44 as they are into business.

26:46 You just need to just move as fast as possible.

26:49 And at some points, moving as fast as possible meant having me working on marketing, and

26:59 applying the system methods that I have thanks to engineering, applying that into marketing.

27:10 When we got acquired by found them, I actually went back to engineering.

27:17 And I stayed there for like, I stayed in engineering for about three to four months, I guess, before

27:24 being pulled back into into marketing, because again, we needed to build systems.

27:30 And so I've been working in marketing at found them for for three years now.

27:36 And yeah, to me, marketing is really about finding the systems that will lead you to

27:45 deliver what the customer wants.

27:47 I do think systems thinking can be really helpful in a number of business functions.

27:54 What in particular attracts you to marketing over all the other ones?

28:00 In marketing, there is something really amazing.

28:03 It's all about the data.

28:06 There is I mean, sure, there is some creativity, and I'm very, very thankful to be to have

28:11 a team that is exceptional and that in creatives and in in copywriting and the totally balance

28:19 my my lack of creativity and my obsession for systems.

28:24 I would say that's because everything is about data, everything is about data, or at least

28:29 improvements about data.

28:32 It just makes sense for for marketing to be led by systems, because the more efficient

28:39 you are in finding sense into the data, the more efficient you are into implementing the

28:44 changes, the recommendation and therefore the changes that come from those learnings.

28:51 And the faster you will go into into growing your pipeline, the faster you'll go into understanding

28:59 what your, what your prospects.

29:03 The faster you're going to launch a new initiative, that's really where you get, you get leads,

29:10 you get them, you get leads that actually convert that are qualified and that will,

29:16 that will convert into into pipeline.

29:19 So system to me, marketing is just a way of it's just a fast track to fast track to success.

29:26 Now that being said, doesn't disqualify quality, so you still need to have high quality content

29:31 system to have high quality creative high quality high quality copywriting.

29:37 And to be honest with you, I love to be able to find to find individuals in my organizations

29:43 that are exceptional at it.

29:46 And that will just make those systems super successful, being able to bring high quality

29:53 content high quality visuals.

29:56Brand versus demand

29:56 I would love your take on, on one debate I always end up having with marketers or even

30:03 thinking about how we run LeanScale to brand versus demand.

30:09 There's some things that you can do that can be difficult to have attribution, but you

30:13 just kind of know that you should be doing them.

30:17 And I think it lives in that having high quality content, copywriting, everything, look and

30:22 feel the aesthetic of what you're, you're building.

30:24 I think that there's importance there.

30:26 How do you think of maybe from a investment gauge, like how much should you be investing

30:33 in brand versus demand?

30:36 And then are there any things about a business that might change what that equation would

30:41 look like?

30:42 Yeah, that's a, that's a great question.

30:47 And, and I would love to be able to, to give like the perfect answer that everyone would

30:53 be, we'd be able to relate to 30% of marketing budget here, 70% there.

30:59 And don't, and don't move, let's do that now.

31:01 Now it's, I think it's way more complicated because it's highly depends on your, on your

31:08 overall strategy as a company.

31:10 It highly depends on your, on your audience as well, obviously for us at Fontaine being

31:18 selling to an audience that are mostly C-level or C-level minus one or minus two.

31:25 It's key for us to be top of mind for them.

31:28 It's key for us to, to be out there and for, for, for prospects to at least know what Fontaine

31:35 does.

31:36 So yes, we are, we are investing a fair amount of budgets, a fair part of our budget into,

31:43 into awareness, into bronze, but it's not only because it's a, not only because we think

31:49 it's good, it's only, it's also because we see that it works really well.

31:54 And what does that mean?

31:55 It means that we do have a large part of our pipeline that comes from inbound and it just

32:02 happened that the pipeline grows when we, when we invest more on, on, on awareness.

32:08 So the correlation really, you know, really works out for us.

32:12 So there is, the reason why we're investing is also because of empiric learnings.

32:18 We learn that the more we, the more we invest, the more pipeline we, we get, but I would

32:26 say that it depends.

32:27 But for my previous company, for instance, awareness in bronze was not like a big game

32:32 changer for us.

32:35 The little growth activities, like the, because we were very, we were very early stage, right?

32:41 So like the smaller activities, like the growth hacking, where the activities that brought

32:46 80 to 90% of our pipeline at Fontaine, it's, there is a much larger part that is due to

32:55 awareness for sure.

32:58 Yeah.

32:59AEO: there is no recipe and nobody has one

32:59 One of the biggest questions a lot of the companies we're working with a lot, I have

33:05 personally is how to optimize for AEO.

33:10 I'm finding myself pretty much exclusively going to the LMS to search for products just

33:16 personally or even in a business context.

33:20 And I'm actively leveraging that way of finding products I want to buy.

33:27 What from a marketing perspective, what's the best way to optimize?

33:32 I know it's very early stages, but anything that you have that helps people get an edge

33:36 on AEO would be super helpful.

33:39 Yeah, absolutely.

33:41 So the one learning that I got from, I don't know, like a year, maybe like two years or

33:48 two years, just meeting companies that are experts in AEO.

33:53 The one learning that I got is that there is no recipe for you.

33:58 Nobody knows exactly how it works.

34:00 No magic words.

34:02 The only thing that we can do is just watch the trends and just react as fast as possible

34:11 to those trends.

34:14 And that can be by publishing an article, that can be by editing the website, can be

34:20 by just doing a press release for whatever product you have or use you have.

34:27 But down the line, the only thing that works is just being extremely attentive to what

34:37 the market sense and to signals.

34:41 So looking back to, settling back to Fontaine Watch, which again is that layer at Fontaine

34:49 that just collects signals from everywhere, from the markets.

34:55 What we'll do is that we will, so we will collect those signals and let's say, I'm going

35:00 to give you an example.

35:01 Let's say there is a new regulation that come out that is going to affect, to have an impact

35:10 on a segment of our customers.

35:13 What's going to happen is we have marketing rights that is going to pick up that signal

35:20 and it's going to tell us that it has two actions for us, two recommendations.

35:26 Recommendation number one is to edit our, like, marketing websites page, like it's going

35:35 to pick up the page that talks about regulations, it's going to say, hey, you need to address

35:39 this regulation.

35:40 It could be, you know, other collections that says, hey, you should really talk to us to

35:45 learn more about how you can mitigate your risk given this new regulation.

35:52 It can be just some, just contents addressing this regulation.

35:58 And number two is going to be maybe a blog post.

36:02 So Marketing OS is going to say, okay, this is a regulation.

36:05 This is everything I know about this regulation.

36:08 This is how it's going to impact your customers.

36:11 There is no legal advice at all.

36:14 It's just purely facts and I'm going to make that recommendation to write a blog post.

36:21 So then it's passed on to the action agents, the action agents is going to get the information,

36:28 write the blog posts.

36:30 We have the human in the loop at some point.

36:32 So we are going to say, yes, I approve that.

36:34 It's approved and it goes to the website.

36:37 So within a few hours from the regulation going live, we have a blog post that is published,

36:47 the website that has occurred to action, that is directly to demo of the product that we

36:53 have that help our customers.

36:57 And that will rank in church PT, in Claude, in the Netherlands in general.

37:05 If that one doesn't rank, it will be the next one.

37:07 If it's not the next one, it will be again the next one, but AEO is a game of volume.

37:12 AEO is a game of getting the right contents at the right time, type your brand.

37:22 And if you do that relentlessly, you get a good ranking on it.

37:28 You get a good ranking.

37:29 You're going to get a good ranking if customer asks for the brand in your category, but you're

37:33 also going to get a good ranking when anyone asks a question about that regulation because

37:38 church is going to quote your article.

37:42 And I don't know what the recipe is in terms of the phrasing and how we should put titles

37:49 and subtitles and all that stuff.

37:51 I have no idea.

37:52 What I know is that the faster I am picking up the signals and publishing, the higher

37:57 I am in church PT ranking.

38:00 And that's just good enough for me.

38:02 It's worked really well so far.

38:03 So we are just keeping, we just keep doing the same.

38:08Why velocity beats the two-week sprint

38:08 That is the first time I have heard the importance of velocity and it's making a ton of sense.

38:16 Everyone's talking about volume.

38:18 Everyone's talking about the hacks.

38:19 Oh, right.

38:20 This type of content, right?

38:21 That type of content, right?

38:22 This or that.

38:23 But if you're the first person to write content and the second people are searching about

38:30 it, then you probably already stacked up people looking at that article, looking at whatever

38:35 you put out into the ether.

38:38 And I think the importance of speed here has to be so much more valuable than I think a

38:44 lot of people are thinking.

38:47 It's key, but it's not only key for AEO.

38:52 It's also key for SEO, even, I mean, SEO is declining, but it's still out there, right?

38:58 So, yeah, I work prospects and customer studios being studios, Google.

39:07 So it's still out there.

39:08 So it's still useful for SEO.

39:11 It's also useful for campaigns, being able to come up with the right awareness campaign

39:18 or the right ABM campaign, targeting very specific segments before all of our competitors

39:29 plays into the success.

39:31 So velocity, speed to me, is at the core of marketing with DNA.

39:41 Marketing OS is not only about volume, sure, volume is there, but it's also about velocity.

39:47 Like I don't want to have to wait two weeks until the next sprints or a week before, until

39:51 the next sprints, in order to consider making an article that will talk about something

40:01 that was, or the signal that was found like a week ago, it's, I think it's a cost of opportunity.

40:11 The one huge advantage of agents is that they can just run super fast and they can be proactive.

40:20 So whenever they see something, whenever they see an opportunity, they will come knocking

40:24 on our door, on my door, on my marketing manager, or like a member of my team and say, you really,

40:33 really, really need to pay attention to this because this will grow your pipeline.

40:41 So yeah, velocity is key and it's not only, you know, it's not only a DNA that we have

40:45 in marketing, it's actually something that we, we leave by the fountain.

40:52 Velocity is everything, and I think all of the hard work in infrastructure that you put

40:59 in has really unlocked and enabled that.

41:02 I think some people are trying to be really fast, but they don't have the foundations

41:06 in place, and this is another excellent example of being able to leverage that infrastructure

41:13 to have high velocity for creating content that helps with the AEO, along with all the

41:17 other things you're doing in paid ads and everything else.

41:20 So I think if teams are wanting to do this, you got to start with that infrastructure

41:25 layer.

41:26 That's a hard part.

41:27 And then you can get really creative on how you execute against it.

41:32The agent builder inside Marketing OS

41:32 Totally. And, and to be honest, velocity is not only something that we use for doing

41:41 marketing activities, creating campaigns and iterating over campaigns and other contents.

41:46 It's also something that we, we also leverage in, in the architecture of MarketingOS.

41:54 In MarketingOS, for instance, if you have an idea of an agent, you don't need to go

42:00 include and try to figure out how to build it and then somehow add it to MarketingOS.

42:05 You can just open MarketingOS and say, "Hey, I want to build an agent that will do whatever

42:14 is your idea.

42:15 So I want an agent that will go and find such signal and based on that signal, we'll launch

42:24 campaigns for such ACP."

42:26 And MarketingOS is actually built to build agents.

42:29 We have an agent builder inside MarketingOS, which is exceptional because all of our agents

42:33 have the same taxonomy, have the same, have the same, like, prompts, I mean the prompt

42:42 architecture, the same one across every single agents.

42:46 It's the same, the skills are shared between, the skills are shared between all agents.

42:53 So it's really amazing that you can just create any, any agent that you want without having

42:57 to worry about anything really.

42:59 You can just describe your, you can just describe your use case and just, and marketing ways

43:03 we build the agents and we're decorating production.

43:05 That's huge.

43:06 So you're not needing to gather, like you said, all the skills, all the data connectors,

43:11 all the context, "Oh, I'm packaging up this new thing," and then the handoff takes time

43:16 and then like, "Oh, okay.

43:17 Well now we're going to like reshape it for this environment and make it production grade."

43:21 Being able to build it directly from MarketingOS is very impressive.

43:25 Yeah, it's huge difference for us.

43:28Wrap

43:28 Bastian, I really appreciate everything you shared on this podcast.

43:32 It is awesome what you're building with MarketingOS and I think the infrastructure plus the execution

43:38 layer, all of the hard work that you put into it, and then getting real use cases.

43:42 I mean the LinkedIn ad comparisons, the awareness that you're pulling from the market and immediately

43:48 building AEO content is just, it's very, very impressive.

43:53 And I love the approach of moving from engineering, but finding love in the marketing world and

44:00 really applying that systems thinking to something that really has a lot of data to work with

44:06 and has a lot of systems that you can build and I'm going to assume there will be a lot

44:13 more engineers falling into the marketing field in the future as well.

44:16 So I just want to thank you again for everything you shared.

44:20 I've always loved working with you and I can't wait because I know this is only the beginning

44:26 of so much stuff you're going to build to see what you do next and follow all of Fountain's

44:30 success.

44:31 Of course, thank you so much Anthony.

44:36 It's been great.

44:37 I love this.

44:38 So thanks for having me and yeah, you're on.