The Founder's Pivot: What a 12% Completion Rate Taught Me About the Training-Video Industry's Real Business Model

Every workforce training vendor knows the number. Corporate e-learning finishes at around 12%. Buy a library of AI-fluency videos for a thousand employees, and roughly a hundred and twenty of them will reach the end. The rest click into lesson one, get pulled into a meeting, and never come back.

For years I read that number the way the industry wants you to read it. As an engagement problem. A production-quality problem. Something a better nudge email or a shorter runtime could fix. Then I spent enough time on the other side of the table, selling to L&D and risk teams, to read it differently. The 12% is not a defect the industry is failing to repair. It is a number the business model does not need to repair.

That is why Cogito is no longer the company I started.

The number everyone misreads

Completion rate gets treated like a gym treats attendance. Low attendance, sick business. But a training-video vendor is not paid on attendance. It is paid on seats. You license a course library for your headcount, the invoice clears in the first quarter, and whether anyone opens lesson seven in the third quarter changes nothing about revenue that already landed.

Run that forward. If the money is fixed at the moment of purchase, what is the rational amount to invest in making lesson seven genuinely change behavior? Enough that procurement renews next year. Not a dollar more. Completion sits downstream of cash that has already been collected, which makes it the one metric a vendor can afford to be relaxed about. So they are relaxed about it.

This is not cynicism, it is incentive design. Show me how a company gets paid and I will show you what it actually optimizes, no matter what the landing page says it cares about. A vendor paid on licenses optimizes for the purchase. A vendor paid on outcomes has to optimize for the outcome. Those are different companies, even when they sell what looks like the same thing.

What the industry actually sells

Here is the part it took me too long to say out loud. The training-video industry does not sell learning. It sells a record that training was offered.

That record has real value, just not the value you were hoping for. When something goes wrong, when an employee pastes client data into a public chatbot or ships a fabricated number to a customer, the first question from legal, from the regulator, from the insurance carrier is always the same: what training did you provide? A completion certificate answers it. It is a receipt. It says the training was offered, so the failure belongs to the individual, not the institution.

That is a product, and a valuable one. A defensible-looking paper trail is worth real money to a general counsel, and the industry prices it accordingly. The trouble is that it is defensibility theater. It looks like evidence of capability and it is actually evidence of distribution. Those are not the same thing, and the distance between them is exactly where your exposure lives.

Your carrier is asking the same question as your auditor

This stopped being an abstract concern in the last two years. General liability and errors-and-omissions carriers have been tightening AI-related coverage. Some are excluding AI-related claims outright. Renewals increasingly ask for evidence of workforce AI-fluency training, and "evidence" is doing real work in that sentence. An underwriter who has watched AI claims climb is not satisfied by a count of assigned videos.

So now two parties want the same thing your audit team wants. The examiner reviewing your EU AI Act literacy obligations or your NIST AI RMF program, and the underwriter deciding your renewal terms, are both asking whether your people can actually evaluate the AI output they touch every day. A completion report does not answer that for either of them. It answers a different, easier question, the one nobody important is asking.

Why 12% is fine for them and a problem for you

Put yourself in the review. It is 2027. The literacy obligations are being enforced, your model-risk program is under examination, and you get the question everyone is now getting: can your workforce evaluate the output of the AI tools it uses?

You hand over the LMS report. Ninety-two percent of staff "completed" AI-fluency training. Then comes the second question, the one the certificate cannot touch. Show me one of them doing it. Show me an employee catching a fabricated citation, the way the attorneys in Mata v. Avianca did not before they filed it. Show me someone refusing a confident, wrong chatbot answer, the way Air Canada's customer-facing bot was not refused before it cost the airline in tribunal. Show me an engineer who declined to paste source code into a public model, the lesson Samsung learned the hard way. Show me the judgment, not the attendance.

The certificate has nothing to say. You did not buy capability. You bought a 12%-completed library and a spreadsheet of green checkmarks, most of which represent a person who watched ninety seconds of a prompt-safety video and closed the tab to get back to real work.

A completion certificate proves the training was offered. It says nothing about whether anyone can do the thing the training was about. In an audit, that distinction is the whole game.

A Fortune and Protiviti survey of more than 1,500 board members and executives named this directly. AI is exposing a critical-thinking gap that threatens an organization's ability to oversee the systems it is deploying. You cannot close a thinking gap with a video the workforce does not finish, and you certainly cannot prove you closed it with a log of play buttons.

The pivot

Cogito started as a tool for building critical-thinking skills. The mechanics were good. The market I had aimed them at was not where those mechanics mattered most. The people losing sleep over "can our workforce evaluate AI output, and can we prove it" were not where I was looking. They were in risk, in L&D, in the general counsel's office at regulated companies, staring at a renewal and a coverage letter and an audit calendar.

So I rebuilt the company around one idea. The product's value has to be the evidence, not the completion. Every session a person works through ends in a Coaching Session Report. Not a checkmark. A scored breakdown of how they reasoned through a scenario specific to their industry and role: where their judgment held, where it slipped, and what they would do differently on the next real decision. The report is the artifact. It goes in the audit file. It is the thing the certificate was only pretending to be.

The format forces the issue. A coaching session is not a video you can leave running in another tab while you answer email. The coach pushes back on weak reasoning and names the specific move the person missed. You cannot finish it without actually doing the work, which means a completed session is, by construction, a record that the work happened. That is the mechanism, and it inverts the video model on the one axis that matters: completion and capability become the same event.

What evidence of judgment actually looks like

It is worth being concrete, because "evidence of judgment" can sound like a slogan. A certificate records a single fact: this person reached the end of the assigned content. A Coaching Session Report records a reasoning trail. It shows the scenario the person faced, the call they made, where the coach challenged them, and how they responded under that pressure. It scores the dimensions of the decision rather than the fact of attendance.

Put the two documents in front of an examiner and watch which one survives a follow-up question. The certificate generates one. "Completed" invites "completed what, and can they do it?" The report closes the loop instead of opening it, because the answer to "can they do it" is the document itself. That is the difference between a receipt and a record.

The tell: whose margin survives if the learner learns nothing?

If you want to know what a vendor really sells, ask one question. Whose margin survives if the learner learns nothing at all?

For a video library, the margin is fine either way. License sold, revenue booked, completion optional. For us it is not fine. A Coaching Session Report that shows no reasoning is worthless to the buyer and embarrassing to us, because the report is the product. We do not get to separate "they paid" from "they could actually do it." Our incentive and your audit exposure point in the same direction, and that alignment is the entire reason I made the switch. I did not want to sell a better receipt. I wanted to sell the thing the receipt was supposed to stand for.

That is also why I stopped competing on price with the video libraries. We are not a cheaper way to generate certificates. We are a different category. The AI Fluency course is built on that distinction, anchored to the real cases your risk team already cites, ending in a verifiable certificate that points back to scored evidence rather than a play-button log.

What this means for your next renewal

You are going to renew an AI-training line item this year. Before you sign, run the audit in your head. Pull your current completion report and ask it the second question. Can it show one employee actually evaluating an AI output, or can it only show that the video was assigned? If the honest answer is the second one, you are paying for defensibility theater, and the day an examiner or an underwriter looks closely, the theater does not hold.

The 12% completion rate is not the training-video industry's problem to solve. It is their tell. It tells you the product was never really about whether your people can think. The good news is that the fix is not more discipline about finishing videos. It is buying the evidence directly. That is a different purchase, and it is the one your audit file, and your carrier, actually need.