The cohort problem: why training providers outgrow their CRM before they outgrow their market
8 September 2026
At HubSpot’s Spotlight this week, one slide summed up a mistake we see constantly in AI adoption, before the company had said a word about why it happens.
Under “many popular use cases drive low value” sat a list that felt uncomfortably familiar: creating content, updating content, researching the market, sending sales emails, planning resources, prepping meetings. The next slide made the contrast explicit: “high value use cases are data and context rich”: prioritising pipeline, flagging at risk deals, scoring leads, resolving tickets, analysing customer feedback.
Look at the two lists side by side and the pattern is hard to miss once you see it. The popular use cases are popular because they’re easy to bolt AI onto without touching a business’s actual data. Draft an email. Write a blog post. Summarise a call. None of it needs to know anything specific about your customers, your team or how you actually sell. That’s exactly why it’s low value; it’s AI doing something generic and adjacent to the business, rather than something embedded in it.
The high value use cases are the opposite. You cannot flag an at risk deal, score a lead properly or resolve a support ticket well without genuinely understanding the specific business, the specific customer and the history between them.
These use cases are valuable precisely because they demand context, and that’s exactly why most businesses haven’t got there yet. It isn’t that businesses picked the wrong priorities. It’s that the easy use cases don’t require the hard work, and the valuable ones don’t work without it.
What “context” actually means
HubSpot put a genuinely useful definition on stage, and it’s worth leaning into to understand your context position. Good context has three parts.
Business context, brand, voice and positioning: what makes your company yours, not a competitor’s.
Customer context, conversations, personas and buying signals: the accumulated, specific history of every real relationship, not a generic persona document.
Team context, roles, goals, approvals and methodologies: how your business actually works, who decides what, and what “done well” looks like inside your specific team.
What bad context looks like in practice
HubSpot shared a striking finding alongside this – AI running on bad context produces worse outcomes than having no AI at all. Not slightly worse, worse than doing nothing. It’s worth being specific about why, because the mechanism is simple once you see it. A person working with incomplete information hesitates, asks a question, or does nothing. An agent working with incomplete information acts anyway, confidently and immediately, at whatever scale it’s been given access to.
Missing business context looks like an agent drafting outreach that references pricing which changed months ago, or positioning against a competitor your business quietly partnered with last quarter. A person would sense that something was off before hitting send. An agent has no sense to check against.
Missing customer context looks like an agent sending a renewal upsell to an account that told your support team last week they’re cancelling, because that conversation lives in a ticket the agent was never connected to. Or it asks a customer something they’ve already answered, which reads to them as “you weren’t listening,” at exactly the moment you needed them to feel understood.
Missing team context looks like an agent approving a discount level that’s meant to need director sign-off, because nobody told it the rule changed last quarter. A person doing this by hand would pause and ask whether it was still right. An agent doesn’t pause.
None of these are a single bad email. They’re the same mistake, made identically, every time the trigger condition repeats, until someone notices the pattern, by which point it has usually already happened to several customers, not one.
The floor on building has come down. The ceiling on thinking has gone up.
This is the part of the shift most conversations about AI miss entirely and it’s the reason the readiness question matters most.
Building has been commoditised. Anyone in a business can now describe a problem in plain language and get a working agent, a workflow, a piece of content, without writing code or waiting on IT. At the same time, what’s now possible to attempt has expanded significantly: more ambitious analysis, more complex coordination across functions, work that used to need a specialist team to even scope properly.
Put those two together and the obvious assumption about AI turns out to be backwards. If building is now cheap and available to everyone, building stops being the differentiator. It was never the scarce resource, it just used to look like one because it was hard. What’s scarce now is judgement – knowing which agent is actually worth building, on what data, governed by what rules, and accountable to whom, before anyone opens a builder tool.
That judgement is exactly what an honest audit of your business, customer and team context produces. It’s not the boring preamble to the interesting AI work. In a world where anyone can build anything, it’s the only part of this that a competitor can’t simply copy by turning the same feature on.
Where this leaves you
For most mid-market businesses, at least one of the three context types is genuinely missing. Business context often exists but isn’t written down anywhere structured. Customer context is scattered across inboxes and call notes rather than the CRM. Team context, the unwritten rules about who approves what and how deals actually get won, is almost never documented anywhere at all.
That’s the starting point. Before choosing which agent to switch on, it’s worth an honest audit of which of the three types of context your business genuinely has in a usable state, and which exists only informally, in someone’s head.
That audit is worth more than any single AI feature you could turn on this quarter, because it decides whether every feature after it actually works, and because it’s the one advantage in this shift that building faster can’t buy you.
