A new campaign kicks off. The brief is a fresh document, because briefs are always fresh documents. Someone remembers the team ran something similar, maybe eighteen months ago, so the excavation begins: a search through Slack, a folder of old decks, a CRM report nobody fully trusts, a spreadsheet named FINAL_v3 that is neither final nor version three.
The deck turns up. It has the results. It does not have the reasoning. Nobody can reconstruct why the team chose that segment, what they rejected, what surprised them halfway through, or which conclusions they would still stand behind. The person who could answer left in the spring, and her context left with her.
So the team does what every team does. They rebuild from fragments and instinct, run the campaign, learn some of the same lessons at full price, and file the results in a new deck that the next team will excavate in another eighteen months.
I have watched this scene for fifteen years, across media agencies, consumer brands, and B2B SaaS demand generation. The companies change. The stack changes. The scene doesn't.
The false diagnosis
When teams notice the symptoms, they almost never name the disease. They call it a performance problem and change agencies. They call it a tooling problem and buy a platform. They call it a reporting problem and rebuild the dashboard, which produces a better-looking view of the same amnesia.
Here is the test I use. Ask a marketing team what their best-performing campaign of the last two years was, and most can answer. Ask why it worked, what the team believed going in, what turned out to be wrong, and what they would never repeat, and the room goes quiet. Or worse, everyone answers differently, with confidence.
There is a second tell. In most GTM organizations, the institutional memory is the org chart. Ask where the learning from the last big ABM push lives, and the answer is a person's name. That works right up until the name is on a farewell card. Organizations that store knowledge in people do not have a knowledge base. They have a risk register.
Neither of these is a performance problem. They are a memory problem, and it has a specific shape: the organization is generating knowledge faster than it can keep it.
Where the knowledge actually lives
Campaign knowledge does not disappear because people are careless. It disappears because it was never stored in one place to begin with.
Strategy lives in documents. Execution lives in the marketing platforms. Attribution settles in the CRM weeks later, in a report that disagrees with the platform numbers. Context lives in Slack threads that scroll away. Judgment lives in people's heads. And the lessons, if anyone wrote them down at all, live in a retrospective deck that was presented once, praised politely, and never opened again.
Every system in the stack captures its stage of the campaign. No system captures what the campaign taught. That sentence took me years to see clearly, and once you see it you cannot unsee it: the modern GTM stack is superb at recording activity and almost incapable of retaining understanding.
The retrospective deserves special mention, because it looks like the solution. Teams do write retros. The problem is retrieval. A retro nobody reads before the next decision is a diary, not a memory. Storage was never the hard part. Reading it back at the moment of decision is.
This is also why more data does not fix it. Data tells you what happened. It does not tell you why the team decided what it decided, which alternatives were on the table, what changed mid-flight when the first week's numbers came in, or which conclusion should shape the next brief. Attribution is the sharpest example. I have sat in rooms with two attribution reports telling two different stories about the same quarter, and the meeting ended with a decision about which report to believe rather than a decision about what to do. Attribution can assign credit. It cannot preserve judgment. Those are different jobs, and we keep asking the first to do the second.
The tax, itemized
Because nobody names the problem, nobody prices it.
The tax gets collected in ways that never appear on the same page:
- Experiments run twice. A test that failed two years ago runs again with new branding and the same result, because the failure was recorded as a number, not as a reason.
- Messaging recycled into the same wall. The segment that never converts gets a fresh sequence every few quarters, from a new hire who has no way of knowing it's a wall.
- Onboarding that takes two quarters. New leaders don't inherit context. They inherit dashboards. So they spend six months re-running the previous leader's playbook to find out, at full cost, what it already proved.
- Handoffs that lose the plot. Between marketing and SDRs, between SDRs and sales, the what transfers and the why doesn't.
- Conflicting truths. Ask three functions why the quarter missed and you get three confident, incompatible answers, each supported by its own dashboard.
- Strategy that resets with the org chart. When leadership changes, direction changes. Not because the evidence changed, but because the evidence was never in a form a new leader could interrogate.
Individually, each looks like the normal cost of doing business. Added up, it is a tax on every dollar the team spends, levied quarterly. And it compounds in the wrong direction: the faster you run campaigns, the faster you pay it.
What a team that remembers looks like
The fix is not another dashboard, and it is not a wiki that goes stale by June. It is an operating rule: every campaign has to leave something behind, and the next campaign has to start by reading it.
The record I have landed on, after years of versions that didn't stick, is short enough to actually get written. The hypothesis: what we believed and why. The decisions: what we chose, what we rejected, and the reasoning at the time, written when it happened, not reconstructed in the retro after the outcome has quietly edited everyone's memory. The evidence: what the numbers said, including which numbers we didn't trust. The surprises: what nobody predicted, which is usually where the real learning is. And the implications: the open questions, and what the next brief should do differently because of all of it.
Then one rule with teeth: no new brief gets approved until it cites the record. Not "reviewed the folder." Cites. Which experiments it inherits, which walls it will not run into again, which assumption from last time it is deliberately re-testing.
The cheapest version of this I have ever run is almost embarrassing. One structured document per campaign, five headings, written in the last thirty minutes of the closing meeting while the context is still warm, filed where the planning template points to it. No platform. The expensive part was never the writing. It was the organizational agreement that the next brief has to read it.
Do this for four quarters and something changes in kind, not just degree. Campaign five starts smarter than campaign one, not because the team got smarter but because the organization did. That distinction, team intelligence versus organizational intelligence, is the whole game. Teams turn over. Organizations, in theory, don't have to.
Why this is finally an AI-shaped problem
For most of my career, the honest objection to all of this was cost. Structured records are expensive to maintain and hard to retrieve, and under quarterly pressure the discipline loses to the deadline. That objection is what has changed.
Language models are genuinely good at exactly the parts that made memory expensive: structuring messy inputs into a consistent record, retrieving the relevant precedent when a new brief starts, comparing this quarter's pattern against last year's, and surfacing the contradiction between what the team believes and what its own record shows.
They are the wrong tool for the parts that matter most: deciding what the evidence means, choosing what to do next, and owning the call. The failure mode I see everywhere right now is teams using AI to produce more, faster, with the same amnesia underneath. More content, more campaigns, more output, none of it leaving a trace. That is not an AI-native operation. That is the old problem at a higher speed.
The right division of labor is boring and strict: models keep, retrieve, and surface. Operators judge, decide, and own. Every AI system I run is built on that split, and it is the split I would audit first in anyone else's. If you are evaluating an AI tool for GTM right now, one question separates the useful from the theatrical: does it make your organization remember anything it didn't remember before?
Why I started building Teamless
I kept hitting this problem in my own work, so I started fixing it for myself. First as the manual discipline above. Then, because I wanted the discipline to survive without me enforcing it, as software.
Teamless is that attempt: a way to preserve the context, decisions, and learning that disappear between campaigns, connected to the systems teams already run rather than replacing them. It began as a passion project. V1 is built, its underlying system is in active use in my own work today, and I am opening it to private-beta users. I am now building V2, testing whether the learning mechanism actually changes what teams do next, and working to turn it into a SaaS product for demand generation teams. That is the honest status, and this essay is not a pitch for it. The thesis stands whether or not the product does: you can run the discipline with a document and a rule, starting Monday.
Which is the point I want to leave you with. The strongest GTM teams of the next decade will not be the ones that run the most campaigns, or even the ones with the best models. They will be the ones that get smarter with every campaign they run. Most teams are one operating rule away from that. The rule is simple: stop paying for the same lesson twice.