Palantir built an ontology for billion-dollar companies. We built the small one.

SECTION 01The word, and what it costs

There is a word for the thing that lets an enormous company see across its own disconnected systems. The word is ontology, and Palantir sells the famous one.

Strip the mystique and it is a description of a business in terms of the real things the business deals with, rather than in terms of tables. Palantir's own documentation defines an object type as "the schema definition of a real-world entity or event" and a link type as "the schema definition of a relationship between two object types".[3] Objects, properties, links, and then actions, which are the things staff are allowed to do. Nouns and verbs. Palantir calls the result a "digital twin of the organization".

It is a genuinely good idea and it works. It is also priced like this, from Palantir's own published rate card for United Kingdom government buyers:[1]

Read the last two rows together. Three million a year buys the licence. The people who build the thing are priced separately, at a hundred and fifty thousand per person per quarter, which is six hundred thousand a year for each engineer deployed. An ontology is not something you buy. It is something a team builds for you, and the team is the expensive part.

SECTION 02What an ontology is actually for

It exists because a large organisation's information is scattered across dozens of systems that were never built to talk to each other, so no single one of them can answer a question that crosses the others.

Which part sat on which aircraft, on which route, on the day a delay cost us money. Which patient is in which bed, against which supply, against which shift. The question is not hard. Getting it to travel across five systems is hard, and that is the whole job.

Now the part that settles our argument. Palantir's own annual report describes a strategy that "targets large-scale, hard-to-execute opportunities at large government and commercial institutions", and says plainly that "high installation costs, high failure risks, complexity of data environments" and long sales cycles "raise the barriers to entry for competition".[2]

The barriers are not a problem they are working on. The barriers are the strategy.

That is a perfectly coherent way to run a company and we are not complaining about it. We are pointing at the consequence. A business doing $6 million a year has exactly the same shape of problem, four or five systems that do not speak, and was never going to be anywhere near this.

SECTION 03The small one

So the question we set ourselves was what survives when you take that idea down to a business with four systems, no engineers, and an owner who is also the person who opens up.

Most of the machinery does not survive, and it does not need to. Nobody here has 180 data sources or a hundred thousand users. What survives is the point of it: the real things the business deals with, joined, so a question can cross a boundary it could not cross before.

And then something becomes answerable that was not answerable in a table. Start from an uncomfortable premise: the owner already knows more about their business than we ever will. They know which regular stopped coming and roughly why. They know which of their people is carrying the shift. Thirty years of that is not a dataset and it is not beatable.

So anything we hand over that they could have worked out by walking around is a report, and reports get read once. The only work that survives is work that was structurally out of reach from where they sit.

We expected that to be a long and growing list. It is four things.

SECTION 04One: you can only stand in one place

An owner can be in one shop. Ride with one technician. Sit with one rep. Watch one register. Whatever they are looking at, they are not looking at the other five.

So anything that requires holding their own units side by side is invisible to them and almost trivial for us. Revenue per hour by technician. Margin by rep. The same product priced two different ways in two of their own locations, which happens constantly and embarrasses everyone when it surfaces.

This is the easiest of the four and it is where most outside analysis stops, because it is the one that falls out of a standard report. It is also the one owners are least surprised by. They suspected. They just could not prove it, or rank it, or put a figure next to it.

SECTION 05Two: you have no organ for slow change

Nothing about a 4 percent price increase looks wrong on the invoice it arrives on. The invoice is correct. The vendor is not doing anything unusual. Every individual document in the chain is unremarkable, and the increase only exists as a shape across four quarters.

Human perception of gradual change is not merely imperfect, it is measurably terrible. In one experiment, roughly half of people giving directions to a stranger failed to notice when that stranger was swapped for a completely different person mid- conversation.[2] If a whole person can be replaced without detection, a vendor moving a line item by a few percent a quarter is not a fair fight.

Drift is not a failure of attention. It is a category of fact that attention is the wrong instrument for.

The fix is not vigilance, and telling an owner to watch more carefully is worse than useless. The fix is that somebody plots the series. A trend line is visible in one second and invisible in a thousand correct invoices.

We wrote about the most expensive version of this in public: a brand that lost roughly 6 percent a year for fifteen years while every single quarter had a reasonable explanation attached to it.

SECTION 06Three: nothing records what stopped happening

This is the big one, and it is the one no system a business owns will ever show them, for a reason that is structural rather than lazy.

Every operational system is built to record events. A sale happened. A job was completed. An invoice went out. That is what the tables are for, and they are good at it.

But the most valuable facts in most businesses are not events. They are the absence of events, and absence does not write a row.

• No customer sends an email to announce they have stopped coming.
• No maintenance agreement announces that it lapsed.
• No completed job raises its hand to say it was never invoiced.
• No quote reports that it has gone stale.
• No product category files a note saying it stopped moving in March.

Finding any of those means building the rows yourself: constructing what should have been there, then subtracting what is. That is a fundamentally different operation from querying, and it is why absence stays hidden in businesses with perfectly good record keeping.

The canonical version of this problem is eighty years old and was not about business at all. In 1943 a statistician working for the United States government was asked where to add armour to bombers, given the damage on returning aircraft.[1] The bullet holes clustered in certain areas, and the intuitive answer is to armour where the holes are.

The correct answer is the opposite. The data only contained planes that came back. The areas with no holes were the areas where a hit meant the plane was not in the sample.

The armour goes where the holes are not.

Every business has a version of that aircraft. The customers who left are not in the customer file. The quotes that died are not in the sales figures. The jobs never invoiced are not in revenue, by definition. The most informative part of the business is systematically the part that is missing.

SECTION 07Four: the one we missed

For a while this framework had three kinds, and we were pleased with it. Across, drift, absence. It felt complete. It was not.

All three are backward-looking. Each describes something already true that the owner cannot see. What kept turning up in real work was a fourth thing that fit none of them, and it kept turning up because it is one of the things owners ask for most.

What is about to run out. Which agreements end in the next sixty days. The week where demand meets a hard capacity ceiling.

None of that is hidden. An owner could work out that one item has six days of stock left. They could do it for a second item. What they will never do is that division for every item, every week, and then rank the answers.

It is not concealed. It is tedious at a scale humans do not operate at, which from the owner's seat is the same thing.

That produces the rule we like most of the four: if a fact is trivially computable and nobody computes it, the absence of effort is itself the opportunity.

We are telling you the framework was wrong at first because that is the part that makes it worth anything. A taxonomy that was never contradicted was never tested against a real business.

SECTION 08Why four and not four hundred

A list of things a business could examine is effectively infinite, and useless for exactly that reason. Every consultancy has one. It is why their first meeting is a discovery workshop: the list does not narrow anything, so the client has to.

The four kinds are not a list of things to look at. They are a filter that runs the other way, against one question asked of every candidate finding: could this owner have seen it from where they stand?

If yes, it does not go in the envelope, however interesting the chart. If no, it is worth sending, even when it is small, because it is genuinely new information to the only person whose opinion counts.

What makes this workable is that the four kinds are industry-agnostic in a way that subject-matter checklists never are. A dental practice, a roofing company and a chain of shops have almost nothing in common at the level of what to measure. They have exactly the same four blind spots, because the blind spots come from the geometry of being one person in one seat, not from the trade.

SECTION 09The point

Most businesses do not have a data problem in the sense they think they do. The numbers are already there, usually in better shape than the owner fears.

What they have is a geometry problem. One person, one vantage point, a finite amount of attention, and a business that generates more facts per week than anyone can hold. Given that, the same four categories of thing go unseen, forever, in every business we have ever opened up.

Knowing that the list is four items long is most of the work. The rest is arithmetic.