The company you never had on the list

A tariff schedule changes on a Tuesday. By Thursday a company you have never held is up nine percent, and someone in the investment meeting asks whether it had been looked at.

It had not, and no screen you ran would have surfaced it. The company sits in a sector classification with no obvious connection to the exposure, and the exposure itself is not recorded in any field a conventional screen reads. It exists in one place only, which is what the management team said about their input costs on a call in February.

This happens across the industry constantly, and the reason has very little to do with how anyone was working.

Classification systems were built for a different job

Sector classification exists so that companies can be compared, indexed, and reported consistently. It does that job well. It was never designed to answer the question of who is exposed to what, and it struggles badly when asked.

Three things get in the way.

Assignment follows majority revenue. A company earning most of its revenue from one activity carries that code, even where a quarter of the business sits somewhere entirely different. The exposure that moves the share price can live comfortably inside a minority of revenue while the classification reports only the majority.

Revision follows a review cycle. Classification updates run on periodic schedules. A company that spends two years repositioning its business keeps its original code for most of that journey, and the code changes long after the market has repriced the story.

Exposure travels across sectors. A duty on a raw material reaches the industrial firm that buys it, the logistics operator that moves it, and the customer three tiers downstream who absorbs the cost. Only the first of those is likely to appear in a sector screen for the theme. The others are exposed in substance and invisible in structure.

None of these are failures of the classification standards. They are simply the standards doing what they were designed to do, being asked to do something else.

Where the exposure actually lives

The company that moved on Thursday told the market about its exposure months earlier. It appeared in a paragraph on an earnings call about renegotiating supplier terms. It appeared again in the risk section of the annual report, in language a keyword search would only find if you already knew the phrasing the company happened to use.

That is the pattern in almost every case. Exposure is disclosed before it is classified, and it is disclosed in prose. Companies describe what they buy, who they sell to, which contracts are up for renewal, and which regulatory changes they are watching. Very little of that becomes a database field, and the part that does becomes one slowly.

So the practical question is not whether the information was available. It was. The question is whether the process had a way of reaching language rather than fields.

There is one more thing worth saying, and it applies even where the classification is perfectly maintained. Companies keep moving, so a screen describes the world only as it stood the day you ran it. That deserves an article of its own, and it gets one later in this series.

What reaching language looks like in practice

The alternative to screening on fields is screening on what companies disclose, which means running the search across filings, transcripts, and disclosures directly.

Practically, that changes three things.

You describe the exposure instead of selecting it. Rather than picking a sector and a geography, you state what you are looking for in your own words. Semantic search finds companies describing the same idea even where their wording differs from yours, which matters because no two management teams describe the same risk the same way.

Sector stops being a boundary. A company qualifies because of what it disclosed, so the logistics operator and the downstream customer appear alongside the obvious industrial name. The list stops being a list of the sector and becomes a list of the exposure.

Every name arrives with its evidence. A match returns with the passage behind it, so the first thing you see is the sentence the company actually wrote. That is what makes the output usable, because a candidate list is only worth reading if the reason each name is on it can be checked in seconds.

The reading, the judgment, and the decision about whether an exposure matters all stay exactly where they were. What changes is the size and honesty of the list those judgments get applied to.

The names that were always findable

The uncomfortable thing about the company that moved on Thursday is that nothing about it was hidden. It filed its documents. It answered the question on a call. It described its supply chain in ordinary language in a public report.

It was reachable the whole time. It simply was not reachable through a system that reads fields, in a universe organized by a taxonomy built to compare rather than to find.

That gap is closing, and it is closing because the documents themselves have become searchable at scale. Orbit is an award-winning AI research platform built for the buy side, and Orbit Screening runs across the full document record for the listed universe, so a screen can work on strategy, disclosure, and management commentary alongside the numbers.

Which means the next time the tariff schedule changes on a Tuesday, the list can include the names that were never in the sector.