File Understanding
Answering questions about a document somebody just sent, from the document itself rather than from your general policy — a distinction that matters more than it sounds.
Answering questions about a document somebody just sent, from the document itself rather than from your general policy — a distinction that matters more than it sounds.
What it does
File understanding is the capability that lets an assistant answer about an uploaded document. It combines several others — reading the file, extracting text, finding the relevant passage, composing an answer — into the behaviour a customer experiences as "it read my contract".
The critical property is attribution. An answer about a customer's document must be clearly about that document, not blended with your general material. Conflating the two is how an assistant tells somebody their specific agreement says something it does not.
Where it shows up in a conversation
Contracts
Answering what a specific agreement says about a specific clause.
Bills and statements
Explaining a charge from the document the customer is looking at.
Forms
Checking completeness before something enters a queue and is rejected later.
Supporting documents
Confirming the right type of document was provided and is legible.
Technical documents
Answering from manuals and specifications nobody reads end to end.
Claims and cases
Extracting the facts that determine next steps from a bundle.
Why it is harder in two languages
The bilingual difficulty compounds because several capabilities stack. A scanned Arabic document requires recognition that handles cursive script and missing vowel marks; then retrieval that matches an informally phrased question against formal document language; then an answer composed in the customer's language.
Each step has an Arabic penalty, and they multiply rather than add. A pipeline where every component is slightly weaker in Arabic produces an experience that is substantially weaker end to end, which is why testing with real Arabic documents matters more than testing any single component.
Bilingual documents are the hardest case and are extremely common here. An official letter with Arabic body text, an English reference number and a table running the other direction requires the system to handle both directions on one page and keep the association between them.
What to look for when you evaluate it
| Real document handling | Tested with photographs and scans of the kind you actually receive. |
|---|---|
| Attribution clarity | Whether the answer distinguishes the customer's document from your policy. |
| Bilingual pages | Whether a mixed-language document reassembles correctly. |
| Malware scanning | Whether uploads are checked before anything opens them. |
| Retention | How long files are kept and who can reach them. |
| Poor-quality behaviour | Whether it asks for a better copy or answers from partial text. |
Where it should not be trusted alone
It should not be used to give advice about a document. Explaining what a clause says is helpful; advising what somebody should do about it is a different act, and in regulated sectors it may be one you are not licensed to perform through an automated channel.
It should not conflate sources. An answer drawn from the customer's file and an answer drawn from your policy are different answers, and the customer needs to know which they received.
And it should not be trusted where recognition was poor. Where a scan could not be read cleanly, answering from what was recovered produces a figure that was never in the document — with no signal that anything went wrong.
How this fits the whole
No single capability answers a question on its own. A conversation passes through several in sequence — understanding what was asked, finding the relevant material, composing a reply, checking it before it is sent — and the quality of the result is set by the weakest link rather than the strongest.
That has a practical consequence for evaluation. Vendors tend to demonstrate their strongest component, and a system with excellent retrieval and weak language handling will demo beautifully and disappoint in production, because the demo was conducted in the language and phrasing the system handles best.
It also explains why grounding matters more than raw capability. A more capable model that answers from its own training rather than from your material produces confident, fluent, unsupported answers — which is a worse outcome than a less capable one that retrieves correctly and declines when it cannot. Capability without grounding increases the cost of being wrong rather than reducing it.
The pieces are described separately here because that is how they are evaluated and configured. They are not experienced separately: a customer asking a question encounters one system, and judges it as one.
Where this sits in a conversation
- Something arrivesA typed message, a spoken sentence, an uploaded document. The first job is establishing what was actually received — which language, which script, whether it is legible, whether it is one question or several.
- It is interpretedWorking out what is being asked rather than what words were used. The same question arrives phrased twenty different ways, in two languages, often mixed, frequently misspelt, sometimes as a fragment.
- Material is retrievedThe relevant part of your own approved content is found. This is where answer quality is largely decided: a correct answer assembled from the wrong source is still wrong.
- A reply is composedIn the language of the question, in the configured tone, from the retrieved material and nothing else. Where the material is silent, the reply says so rather than filling the gap.
- It is checkedBefore sending, the reply is examined for claims the retrieved material does not support and for personal data that should not be repeated back. This runs on every reply.
- It is deliveredAs streamed text, as speech, or as a structured response — and the conversation is recorded so that what happened can be reviewed later.
Why this is built rather than bought in
Every capability described here could, in principle, be obtained by calling somebody else's service. That is how most assistants in this market are assembled, and it is why so many of them cannot answer the question that decides regulated procurements: where does the data go.
When a component is a call to an external provider, every question that passes through it leaves the organisation. That may be perfectly acceptable for a retailer answering questions about opening hours. It is usually not acceptable for a bank explaining a transaction, a hospital confirming an appointment, or a public body handling a citizen case — and the constraint is architectural, so it cannot be configured away later.
The second consequence is control over behaviour. A capability obtained as a service behaves the way the provider decides, changes when the provider changes it, and cannot be tuned for the specific way your customers speak or write. A capability that runs inside your own estate can be adjusted for dialect, for your terminology, and for the particular failure modes your material produces.
The third is continuity. External services are deprecated, repriced and withdrawn. A deployment whose core capability depends on somebody else's roadmap inherits that roadmap, including the parts nobody announced in advance.
Measuring this component honestly
| Measured in which language? | Nearly every published benchmark in this field was developed against English. An overall figure that averages across languages hides exactly the gap that matters in this market. |
|---|---|
| Measured on what input? | Clean or realistic. Read speech or spontaneous speech. Typed text or phone-typed text with autocorrect. The difference between these is usually larger than the difference between vendors. |
| Measured by whom? | Self-reported figures with an undescribed method are marketing. Ask for the test set and the procedure. |
| What counts as correct? | A surprising number of published figures count a partially correct answer as correct. Ask what the scoring rule was. |
| How does it fail? | A component that degrades gracefully is worth more than one with a higher headline figure that fails silently and confidently. |
| Does it report uncertainty? | A component that surfaces low confidence lets the system confirm rather than guess. One that hides it produces confident errors downstream. |
What to ask about this specifically
Ask for the Arabic figure
Not the overall figure. Nearly every published benchmark in this field was developed against English, and the gap between the two languages is where deployments disappoint.
Ask how it was measured
Against what test set, by whom, on what kind of input. A number without a described method is marketing rather than measurement.
Ask what it does when uncertain
Whether uncertainty is surfaced or hidden. A component that hides low confidence produces confident errors downstream.
Ask about real conditions
Noisy audio, poor scans, dialect, informal phrasing, mixed languages. Laboratory conditions are not predictive of contact-centre reality.
Ask where processing happens
For any component touching personal data, this is a procurement question rather than a technical curiosity in regulated sectors.
Ask what is retained
What data this component keeps, for how long, and whether it is used to improve anything.
Common questions
Office documents, PDFs, images and phone photographs. Requiring conversion defeats the purpose, since most real uploads are photographs taken on a phone.
From the document, and it says so. Conflating the two is how an assistant tells somebody their agreement says something it does not.
Yes, before anything opens them. An unscanned upload path is a security hole with a friendly interface.
They are not kept. Elbi reads the file to answer and never writes it to disk — uploaded documents are among the most sensitive data the assistant touches.
Common here and the hardest case — Arabic body text with English references and tables running the other way. Worth testing with your own material specifically.
It explains what the document says. Advising what to do about it is a different act with different consequences, and the boundary is set deliberately.