Most AI use at this company still starts with zero company context, two years into serious adoption.

By all means, this isn’t a company that has been sitting still. The entire company is enthusiastic and moves at a rapid pace. They’ve deliberately combined multiple tools, done custom work and built their own internal chatbot. They also have an active group of AI ambassadors structuring their efforts.

Yet people in customer service, sales and other parts of the business aren’t consistently using the same context or sharing the lessons that truly differentiate the company. And I think this will seriously hamper how much value they can extract from AI.

What happens without context

Because if you ask a model to do something without any organizational context, without guardrails and without the things that make your company unique, the result might be fine. It might even be perfectly written. But it could have come from any company.

Ask it to respond to a complaining customer and it can write a polite email. But does it know your return policy? Does it know how you handled a similar case last spring? Does it understand what kind of decision your company would actually stand behind? Probably not.

You see it most clearly when two people ask the same question and get two different answers, both of them reasonable. Nothing is wrong with either output. There is just nothing holding them to the same line.

And that problem doesn’t disappear, even when a new frontier model seems to be released every other week.

The company vault

The fix is unglamorous. You need a place where the things that make your company specific are written down, in plain files a model can read. Your policies, the decisions you have already made and the reasons behind them, the phrasing you use with customers, the cases that set a precedent.

Most of this already exists somewhere. It sits in a policy document, in a closed ticket, in the head of the person who has been there longest. The work is collecting it in one place and keeping it short enough that people maintain it.

Give every part of the vault an owner and a review date. Without those two things it goes stale within a quarter, and a stale vault is worse than none, because people stop trusting what it says and go back to explaining everything themselves.

The personal vault

Alongside that, people need their own. How they write, which accounts they handle, what they are working on this month. This is the part individuals will actually keep up to date, because they feel the benefit the same day.

The two work together. The company vault holds what everyone should be consistent about. The personal vault holds what makes one person’s work different from their colleague’s.

Turning expertise into something the team can use

Once the context is written down, you can go a step further and package it. If your best person at handling complaints has a way of working, that can become a skill the whole team uses, with the policy and the precedents already attached to it.

That is the point where AI stops being a faster typist and starts carrying some of what your company knows. It also means a new joiner gets the good version from their first week instead of after a year.

If every employee still has to explain the company from scratch, you simply haven’t scaled organizational intelligence.

Originally posted on LinkedIn