Being right doesn't matter if nobody acts on it. That's true whether you're presenting to executives who need to move, or building a pipeline that runs while you sleep.
March 2016: the math was wrong
I created a presentation in March 2016 at Banyan showing that the company had been calculating its churn rate incorrectly. The standard metric — clients lost per month over total clients — didn't account for the twelve-month contracts clients signed, nor the rapid expansion the sales team had after rounds of funding.
Here's what that hid: we had cohorts of clients who would all age out of their contracts within months of each other. The older, pre-funding cohorts showed cohort-level retention of roughly 30%, but those cohorts were small enough that the blended number looked fine. The denominator kept swelling with new clients who couldn't churn yet, so monthly churn looked manageable right up until the big post-funding cohorts hit their contract anniversary.
My estimate in March: without strong retention strategies — critically, a retention team that could spearhead tactics and research — we were going to lose potentially 50-60% of our revenue by the end of the year.
The executive response: the presentation was "overly doom-and-gloom."
By December, the company had churned so many clients that a prospective buyer backed out at the last second. By January, the team was cut from roughly 130 employees to less than 40. Banyan went out of business after I left, largely because they ignored the warning and never recovered.
All of those numbers are my recollection; I don't have access to Banyan's financials to verify them now. What I do have is the lesson: having the right information doesn't matter if people aren't willing to act on it. I learned a lot about the importance of presentation of knowledge, and using Socratic tactics to create an inception of ideas, where others might believe they arrived at a conclusion independently.
The retention research that produced the warning also produced the Brown Score, which account managers later worked from. The analysis was good. The persuasion wasn't.
The check nobody built
I built a job-search pipeline for a family member. The system discovers job postings, scores them against a candidate profile across thirteen weighted dimensions, routes each into a pursuit strategy, generates tailored application materials, and tracks every job through a tiered human-approval gate. It's parked now, but it did work: 25,605 lines of Python across 56 files, 445 test functions in the current suite, no database, no server process — just two third-party packages and plain JSON files on disk.
One job made it all the way to the final approval gate with a strong score and polished materials before I caught something: none of the thirteen scoring dimensions had evaluated whether the candidate would actually want to work at that company. We determined that a company was morally misaligned with the candidate's values, seconds before the application went out.
A red_lines field already existed on the candidate's profile. It held exactly the values information that would have flagged the mismatch. Nothing in the pipeline read it.
The fix added an ethics_review gating stage between materials generation and approval. The code itself was straightforward — the important part was that the gate had to run, and it had to block if the check failed.
What I learned twice
A correct warning nobody acted on and a values field nobody's pipeline read are the same shape of failure: the fact existed and wasn't load-bearing in the decision.
With people, the fix is persuasion. You present the knowledge so it lands. You ask the questions that let them reach the conclusion themselves. The March 2016 presentation had the right math and the wrong approach.
With agents, the fix is structure. You don't persuade a pipeline; you add the gate that must run and must block. The ethics_review stage reads the red_lines field and stops the job before it reaches approval if there's a conflict. It can't be skipped and it can't be talked around.
The principle is the same in both cases: information only matters if the system — human or automated — is built to act on it. In 2016 I had the numbers but not the persuasion. In 2026 I had the field but not the gate. Both times, the fix was structural, not informational. One required changing how people thought about the problem; the other required changing what the code had to check before it could proceed.