Loom Intelligence
Research that takes weeks, in an afternoon.
You are sitting on a mountain of figures.
Every answer costs an afternoon of searching.
And then you have one.
Loom Intelligence does the research.
A sample looks at 2%.Loom looks at everything.
No human reads 800,000 entries. This engine does — and millions of rows just as well.
Size is not the limit: the calculations run on the file itself rather than reading everything through a language model — whether that is thousands of rows or terabytes.
We asked it one thing
Does money circulate between entities, and does it ultimately return to the entity it came from?
| Finding | Result |
|---|---|
| Closed money loops | 253 |
| Steps within those loops | 1,646 |
| Amount in circulation | €156.0 m |
| Direct return patterns | 140 |
| Amount in direct returns | €15.3 m |
The largest loop closes after eight entities
BV001BV029BV023BV004BV063BV053BV021BV048BV001
A route that returns to where it began, with a total of €1.54 m. For every finding the underlying calculation is preserved: from question to research to evidence.
The same engine ran test cases on warehouse and supply-chain data, on sales figures and on an e-commerce administration of a hundred thousand orders over two years.
You do not get a table.You get a report.
Three steps
You put your files in.
Simply what you already have: Excel, CSV, an export from your accounting package, your order system or your inventory management. It need not be cleaned up — the engine looks at what is in it and tells you if something is off: duplicate rows, empty fields, odd dates, columns that are not what they claim to be.
You ask your question in plain language.
“Which customers are leaving?” is enough. Just as much as: “Go through the entire ledger for entries that do not fit the pattern.” You need not know how it should be calculated.
You get a report, not a table.
Not a raw printout you still have to make sense of, but a readable document: the conclusion at the top, the reasoning beneath it, the figures alongside. Ready to forward to your board, your accountant or your client.
Your data remains yours
Sensitive data is recognised and masked before it enters the engine.
Names, account numbers, addresses and other identifying fields are replaced by a token. What remains is exactly what the research turns on: the amounts, the dates, the relationships.
If you work in an environment where that does not go far enough, you decide for yourself: field by field, you can manually mark what should additionally be tokenised, before anything is processed.
Nor is masking blunt. If a protected field contains something the research needs — a country code, a category, a year — you can release precisely that part while the rest remains a token. Only with your explicit consent, field by field.
| Field | What you supply | What the engine sees |
|---|---|---|
| Name | J. de Vries | PERSON_01 |
| Account | NL00 BANK 0123 4567 89 | ACCOUNT_04 · NL |
| Email address | j.devries@example.nl | EMAIL_02 |
| Amount | € 12,480.00 | € 12,480.00 |
The country code NL is still there, because it was released for this research. Without that consent, only the token remains.
And what need not go, does not go.
On the infrastructure that is already there
Loom Intelligence is not tied to a single language model. If your organisation already works with an AI environment of its own — with your own cloud provider, within your own network, with a model you manage yourself — the engine runs inside it.
The model is thereby a component you choose, not a party you are tied to. If your organisation later moves to a different model, the research simply moves with it.
It is the same idea as in our engagements: we adapt to the infrastructure that is already there, not the other way round.
Four examples
Screening your ledger
Every row checked, not a sample. Amounts outside the bandwidth, entries at odd moments, items that do not reconcile, reversals that cancel each other out, accounts that suddenly behave differently from the eleven months before.
Detecting fraud and anomalies
Holding the entire file up against itself: duplicate payments, a supplier with an employee's account number, invoices that stay just under an approval threshold, amounts that are suspiciously often round, customers that exist for only one quarter. You get the signals together with the underlying rows, so that you can verify them yourself.
Checking your supply chain
Where is the delay really, with which supplier, in which month? Which items are structurally late, which stock lies idle, which supplier became more expensive without anyone noticing? Orders, deliveries, stock and invoices are set side by side — even if they are four separate exports from three systems.
Reading customers, revenue and margin
Who is leaving and since when, where the margin really is, which product carries the growth, and which discount costs more than it earns.
This is not AI. AI is a component of it.
Loom Intelligence is a research engine. AI reads your question and recognises what is in your columns — the calculation is done in code, on your own file.
Anyone can ask an AI something about a spreadsheet these days. The problem is that you never know whether the figure is real.
No invented figures.
Every figure in the report comes from a calculation on your file and can be traced back to the rows it came from. Nothing is estimated and nothing is filled in — and what is not in there, you are told.
The figures are actually calculated.
Every figure in the report comes from a calculation on your file and can be traced back to the rows it came from. Nothing is estimated and nothing is filled in.
You see the assumptions before anything is calculated.
“I am assuming this column is your revenue, per order.” Assumptions of that kind are shown to you while you can still correct them. So you never depend on a guess.
It tells you when the answer is not in there.
If your data cannot answer a question, you are told so, with the reason. Better an honest “I cannot see that here” than a handsome figure that means nothing.
It is built to be token-efficient.
The calculations run on your file itself; no more data passes through a language model than is needed to understand the question. That is why the size of your file is no limitation: the model does not read everything, the engine calculates everything.
For those who want to know their own figures
Accountancy and audit firms, internal control and finance, forensic investigators, procurement and supply-chain managers — and any business that wants, for once, to really know what is in its own administration.
And no, it does not replace your accountant or analyst. It hands them the detective work ready-made, so that their time goes to judgement instead of searching.