Every company knows less about itself than it thinks.

The people are careful, and the systems do the jobs they were built for. But what actually runs a business lives outside them: in the emails that negotiated the deal, in the meeting where someone agreed to a change, in the PDF a vendor sent eighteen months ago that nobody filed anywhere the ERP could see. All of it is real, and almost none of it is read.

The person who has to answer for the business is always answering with less than the full picture, and everyone has learned to treat that as normal.

I think that should end. Cotiss reads everything a company already produces about its vendors and its commitments, works out what each fact actually means under that company's own rules, and turns it into one current, sourced picture that people, and increasingly agents, can act on. Say go, and it acts.

That's what we built Cotiss to do, and this letter is the argument behind it.

The pressure from outside

Nineteen line items on a cost-out, three weeks out from the board meeting. Seven are tracking, and the other twelve are somewhere else entirely: in vendor emails, in the integration team's notes, in a tariff letter that arrived last week and hasn't been read past the first page. Six business days left, and the board has made it clear it doesn't want the story afterward.

Nobody in that room did anything wrong. This is simply what happens when work answers to someone outside the team.

Most work doesn't. A tightened process, a faster workflow: if that slips, nobody beyond the team ever knows. But a board wants proof on a date, an auditor wants the trail behind it, a regulator's deadline was written into law, and an acquisition promised savings for a named quarter. None of them will wait while the answer gets rebuilt.

And rebuilding is what it takes, because the systems never held the whole story. Procurement and finance know this better than anyone. They're the most process-minded operators in the company, and they've spent years working around the gap without ever naming it. Each workaround looks small on its own: ten minutes to walk to the desk of the one person who remembers, an evening on the extra spreadsheet, an hour of inbox searching nobody budgeted, and then the judgment call that finally closes the question, which costs nothing at all unless it turns out to be wrong.

The workaround holds right up until the date lands, and then twelve line items are due in six days.

That deck is why Cotiss exists.

How it shows up

Customers never describe this as a data problem. They describe what it feels like on a Tuesday, and it turns out to be the same gap wearing three different faces.

The numbers. Somewhere, the company has committed to one: a profit target set by its owners, a public promise to cut cost, the synergies a deal was priced on, sometimes just the number that keeps a covenant intact. The value behind that number is real and it's hard to see, because the levers sit scattered across contracts, invoice lines, ledger codes, and emails nobody reads twice. Finding it today usually means weeks of analysts building a spreadsheet by hand.

The deadlines. A date gets set from outside and it doesn't move: the first audit after going public, a regulator's rules, the integration savings a deal was signed on. Each one demands a trail: who you're committed to, what those commitments actually say, and what's happened since. Then you discover three companies on three systems, the same vendor showing up three different ways, and the person who negotiated the original contract left the business six months ago. The auditor's question isn't complicated. Which version is right?

The risks. A risk doesn't show up on a calendar or in the ERP. A sanctions list changes overnight and a vendor is suddenly off-limits. A cyberattack hits a vendor you signed a data agreement with years ago, and nobody remembers what that agreement actually promised. Spotting the signal is rarely the hard part, and knowing what it means is: whether there's a right to terminate, a liability cap, a notice window that has to be hit this week. The question runs all the way from the portfolio down to a single clause.

Three different shapes, and the same demand underneath all of them: a current, evidenced picture of where things stand, in time to act on it rather than explain it afterward.

The problem, named

The facts that run the business already exist. They were written into emails, signed into contracts, said out loud on calls, and they stayed where they landed, apart from each other, unread as a whole, because no system the company owns can see them together. Every hard question gets answered by gathering the pieces by hand, under pressure, and nobody can say for certain what the answer missed.

Why this isn't already solved

If the gap is this obvious, why is it still open?

Every tool a company owns made the same bet: people will type what happened into a form. They typed just enough to keep work moving, and the rest stayed where it was made.

Follow one deal through the stack. The negotiation lives in email, on calls, and in a vendor's PDF. The contract tool holds the signed document without the reasoning behind it. Whatever fit the procurement platform's form fields is in there, and the accounting system has the transaction, because guarding the ledger is its whole job. By the time spend analysis reads that ledger, the deal is one line and everything that explained it is gone. Generic AI reads any one artifact well, and holds no current, source-backed picture of the whole business, resolved into that company's own vendors and rules.

None of that is a failure of effort. There was never a tool whose job was reading where the work happens, and until very recently there couldn't be: reading what a real company writes down each year only became affordable once models got cheap and accurate.

Start from what actually accumulates. After a year inside one business, Cotiss has read that company's entire commercial history, resolved it into a single picture under the company's own rules, and kept a source on every fact. Every month of use sharpens what's there. None of that arrives with a model release, and it's why I don't expect mature companies to close this gap on their own just because reading got cheap. Cheap reading is what everyone has now, and what nobody gets off the shelf is the understanding that has already compounded around their own business.

Bolting a model onto a workflow tool doesn't change its bet: the form fills in faster, and the rest still lands on the floor. Building it in-house hits the same wall, because the model is the commodity and the memory, the routing, the business logic and the evidence trail underneath them are not.

Even the most generous future for AI leaves the gap standing, and moves the whole game onto whoever owns the context. The work is heading to agents, which sharpens the point: an agent can only act on what it can read.

What I believe

Real work happens in email, on calls, and in documents, and AI is sending more of it there every year. That's where we read. The reasoning behind a deal survives in the thread and the vendor PDF, and it starts dying the moment somebody paraphrases it, so we keep the original and give every fact a path back to it. Checking what Cotiss tells you takes a click.

Reading is only half of it. A clean answer from a single source can be right on the page and still wrong for the business, so what a fact means under your rules matters as much as what it says on paper. Cotiss starts from a working model of companies like yours, then learns your vendors, your categories, and the language your team actually uses, and it keeps learning as you grow.

What a business knows should outlast the software it happens to be running. We build that layer, and whatever tool comes next plugs into it.

What changes

The shift is simple to state and hard to live without: the scattered record of what the business knows becomes a connected one, living and current, updated as the facts change instead of rebuilt from scratch every time someone outside asks.

Five things change first.

Money stops leaking unnoticed. The value that used to slip out of vendor spend a little at a time surfaces as it happens, not months later in an analyst's spreadsheet after it's already gone.
Procurement sees the deal while it's still forming, early enough to shape the outcome.
Knowledge survives the person who held it. The deal whose details left with the person who negotiated it stops being a risk, because the record doesn't depend on anyone's memory.
Audit prep and diligence stop being a fire drill twice a year and become a state you're always in.
Agents finally have something worth acting on: a picture that's complete, current, sourced, and shaped by the company's own rules, rather than fragments they have to guess around.

Savings is usually where it starts. What it grows into is control across cost, risk, continuity, and compliance, all resting on the same understanding of the business.

Where this goes

Every company knows less about itself than it thinks. That doesn't have to stay true.

The teams who close the gap first get something simple: when the board finally asks, they already know the answer.