The Board-Meeting Use Case We Never Pitched — And the Support Ping That Revealed It

Our chat agent crashed mid-session.

Within seconds, a customer pinged me directly. Not a ticket. Not a form. A direct message to me.

"Your agent stopped. You're disrupting my board prep."

He was the CEO of a Swedish company that sells local services. He was not on our marketing team. He was not a user I had ever demoed. And what he was doing in Cogny when it crashed was something we had never pitched, never documented, and never designed for.

He was running user cohort analysis for his board deck.

I was, at that moment, looking at our own positioning deck. The headline on it said "AI Agents for Marketing."

That support ping rewrote it.

What "AI Agents for Marketing" Means — And What It Doesn't

If you read the marketing-AI category at face value, you know what it's supposed to do.

Campaign analysis. Creative iteration. Budget reallocation. Attribution modelling. Maybe a copilot for a Meta Ads manager. Maybe a chat that sits next to GA4 and saves you from writing SQL.

That's what we shipped. That's what we pitched. That's what our landing page said.

The thing is — by the time this customer had been live for a while, here is what he had actually plugged in:

His CRM — every customer record, every service booking, every status change
Google Ads and Meta Ads for paid acquisition
Google Search Console for organic demand
His email platform — every send, every open, every click
Google Analytics 4 for session-level behavior

Five sources. One business. One agent that could query across all of it in natural language.

At that point, the label on the box is a lie. You have not built a marketing tool. You have built a CEO-level data layer and labeled it a marketing tool.

The customer figured that out before we did.

The Discovery: A Ping, Not a Sales Call

Here is the thing I keep going back to.

We did not pitch board prep. There is no onboarding flow in Cogny that says "great news, you can use this for your next board meeting." There is no template called Board Deck Cohort Analysis. There was no sales call where he asked, "can your product help me talk to my investors?" and we said yes.

He just started doing it.

And the way we found out was not a glowing NPS survey or a case-study interview we set up. It was a support ping, written with the impatience of a CEO who was in flow and had been interrupted.

That impatience is the signal. When a customer is annoyed that your product stopped working because they were trying to use it for something you did not advertise, you have learned something that no amount of positioning strategy work would have surfaced.

The positioning exercise says: here is who we think the user is. The support ping says: here is who the user actually is.

When those disagree, the support ping wins. Every time.

What He Was Actually Doing

He was running a user cohort analysis.

For a local-services business, cohorts are where the money is. Customers are not one-off transactions. They rebook. Or they don't. They refer. Or they don't. They respond to a follow-up email three months after their first service. Or they don't. Your unit economics live inside that pattern.

To answer a question like "how do our customers from paid search compare to our customers from organic, on rebooking rate and email engagement, over the first 90 days?" you need:

Your CRM to define who the customers are and when they first booked
Your ad platforms to know which channel acquired them
GA4 to connect the session that led to the booking
Your email platform to see what they did with follow-up sends
A way to join all of that together in a single analysis, cohorted by acquisition month

Inside a previous stack, this question takes a week of analyst time. Maybe longer — data sits in five places, the join keys are inconsistent, and by the time you have the answer the board meeting is over.

Inside Cogny, he was doing it himself, the night before the board meeting, in natural language. That is why he pinged me the moment the agent stopped. His BATNA was not "send a tweet." His BATNA was "manually stitch five data sources together at midnight."

The Repositioning Lesson

Here is what one support ping taught us.

When you build a connected data layer, your ICP quietly shifts on you.

We thought we were building for the marketing manager. The marketing manager is a real user and a great user. But the highest-leverage user of a product that sits on top of CRM + marketing + email + search + GA is not the marketing manager. It is the person in the company who needs to answer cross-system questions under time pressure, with no analyst in the room.

In a small-to-mid company, that person is often the CEO.

In a business preparing for a board meeting, it is definitely the CEO.

And the cost of mislabeling this is not that a few landing pages were wrong. The cost is that our own team anchors on the smaller user. We would have under-built cohort features. We would have under-invested in exec-level narration. We would have under-priced the product. We would have written a roadmap for the marketing manager and quietly missed the CEO standing two meters away asking for a board deck.

A caveat, honestly: not every marketing-AI tool is secretly a CEO tool. This one was — specifically because of the breadth of data one particular customer chose to connect. If you have a narrow AI copilot wired to one channel, the CEO use case does not light up. The signal lives in customer behavior, not in the category label.

The practical version of the lesson, for any founder reading this:

Your support inbox is more honest about your ICP than your positioning doc is.
The pings you care about are the ones that mention a use case you did not pitch.
When you find one, do not dismiss it as "edge case." Ask how often it is happening, quietly, to people who did not happen to hit a bug.

What You Can Try

If you are a founder or CEO of a business where your CRM, ad platforms, email, and analytics have never all been in the same place before — and you have a board meeting in the next 30 days — here is the cross-system question I would try first:

"Of our customers acquired in the last 6 months, break them into cohorts by acquisition channel. For each cohort, show me rebooking rate, average revenue per customer at 90 days, and email engagement on our follow-up sequence. Flag any cohort that is underperforming the others by more than 20%."

If your existing stack cannot answer that in an afternoon, that is the signal. It is not "your team needs a better BI tool." It is that the question lives across five systems and, until those five systems share an agent, the question is unanswerable at board speed.

Our customer answered it the night before his meeting.

He was annoyed when it broke for ninety seconds, because at that point the product had already changed what board prep was allowed to feel like.

How this article was reported

This post is about a real customer interaction. What is quoted above ("Your agent stopped. You're disrupting my board prep.") is paraphrased from a direct message received during a chat-agent outage. The customer is described by role and company shape only; their name, their company, and the specific city they operate in are withheld intentionally. The claim about five connected data sources reflects the integrations this customer had live at the time of the incident. The specific cohort question shown is illustrative of the category of question he was asking, not a verbatim transcript of their exact query.

Berner Setterwall is co-founder of Cogny. He has spent eleven years in growth, most of them arguing that the data layer is the product.