Text Summarisation
Compressing something long into something useful. Genuinely valuable for handover and review, and genuinely risky wherever the detail that was dropped was the important part.
Compressing something long into something useful. Genuinely valuable for handover and review, and genuinely risky wherever the detail that was dropped was the important part.
What it does
Summarisation condenses a longer text into a shorter one that preserves what matters. In an assistant its highest-value use is handover: giving a human agent the state of a conversation in a few lines rather than a transcript they will not read.
The risk is inherent to the task. Summarising means deciding what to omit, and a summary that drops the one detail that mattered is worse than no summary, because it looks complete.
Where it shows up in a conversation
Handover
Giving an agent the state of a conversation quickly, so they continue rather than restart.
Case notes
Turning an interaction into a record somebody can act on later.
Document questions
Condensing a long document into an answer to a specific question.
Review
Making a large volume of conversations reviewable at all.
Escalation context
What was tried, what was established, what remains open.
Reporting
Aggregating themes across conversations for people who will not read them.
Why it is harder in two languages
Summarising across languages is where this most often goes wrong. A conversation conducted in Arabic summarised into English loses nuance, tone and frequently the precise phrasing that made a complaint a complaint — and the agent reading the summary then handles a different situation from the one that occurred.
The right default is summarising in the language of the conversation and routing to an agent who reads that language. Translation should be a deliberate exception rather than a convenience, because the cost falls on exactly the interactions that most need care.
Arabic summarisation also has to handle register. A conversation containing formal courtesy and a direct complaint should preserve the complaint, not average the tone into something neutral that reads as routine.
What to look for when you evaluate it
| Faithfulness | Whether the summary contains only what the source contained. Added detail is the most dangerous failure here. |
|---|---|
| Omission behaviour | What gets dropped, and whether anything signals that detail was lost. |
| Same-language summarisation | Whether Arabic conversations summarise into Arabic by default. |
| Tone preservation | Whether a complaint still reads as a complaint after compression. |
| Length control | Whether summaries are consistently useful rather than variably long. |
| Handover usefulness | Measured properly: does the agent still have to ask the customer to repeat? |
Where it should not be trusted alone
Summaries should not be the only record. Where a decision, a complaint or a commitment occurred, the full text has to remain available, because the summary is an interpretation and the original is evidence.
They should also not be trusted for anything consequential without the source. An agent deciding a case, a manager reviewing a complaint or a compliance function examining an interaction should be reading what was said, not a compression of it.
And summarisation should never introduce. The most dangerous failure is a summary containing a plausible detail that was not in the conversation — it is fluent, it fits, and there is no signal that it was invented.
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
Handover, above everything else. An agent who reads three lines and continues the conversation is the difference between a good handover and a customer repeating themselves.
Yes — that is inherent to summarising. Which is why the full transcript remains available and why summaries are not the record.
Not by default. Nuance and tone are lost, and the agent handles a different situation from the one that occurred. Route to an agent who reads the language instead.
That is the failure to guard against. A summary is checked against its source, because a fluent invented detail carries no signal that it was invented.
No. Where something matters, the original text is the evidence and the summary is an interpretation of it.
Yes, and answering a specific question from a long document is usually more useful than a general summary of it.