A placement manager has a renewal due to go to market in six days. The client’s information has arrived as an email chain, two spreadsheets and a PDF, and the figures in them do not quite match. Nobody has made a single judgement about the risk yet. The first day is spent working out what the submission actually says.
That work does not show up on a placement timeline because it is not underwriting, pricing or placing. It is preparation. It is also where every later step gets its data from, which is why brokers put it near the top of what they want automated.
Brokers Name Intake as the Most Valuable AI Use Case Today
The clearest signal comes from brokers themselves. In Guidewire’s London Market Tech Barometer 2026, 42 percent of surveyed brokers cited automating submission intake and data extraction as the most valuable AI use case today, with a further 38 percent pointing to exposure management. Guidewire reads this as evidence that brokers see the quickest return from AI where it deals with messy, unstructured information early in the workflow. How extraction works in practice is covered in 5 Ways to Automate Insurance Document Workflows.
The survey covered 251 brokers who place primarily with London Market insurers, so the finding describes that market rather than every broking desk everywhere. The logic behind it travels further. Intake is where unstructured information first meets a structured process, and every later stage reads whatever intake produces.
That explains why brokers want it. It does not yet explain why it should come before the uses that attract more attention.
Intake Is the Right Place to Start Because Its Errors Can Be Seen and Corrected
Trust is the constraint on AI in broking, and intake is where it is easiest to earn. The same Guidewire survey found that 55 percent of brokers are concerned about algorithmic bias and a lack of transparency in AI-driven risk selection. Guidewire concludes that confidence is highest where AI supports a decision rather than making it.
Intake fits that description. An extracted value can be checked against the document it came from. A wrong value is visible, and a person can correct it before anything depends on it. A risk selection offers a much harder check, because there is no source document to compare it against.
The conversations at BIBA 2026 pointed the same way. Genasys, a policy administration software provider, reported that the AI uses brokers raised with its team were narrow and practical. Document ingestion, bordereaux processing and renewals triage came up repeatedly, as high-volume tasks where an error can be seen and put right, and nobody it spoke to wanted AI making unsupervised underwriting decisions.
Brokers are not asking AI to decide first. They are asking it to read first.
The obvious pushback is that intake is clerical work and the real value sits in underwriting judgement. That holds for value per decision. It does not hold for where trust can be built first, and trust decides how far any desk will let automation go.
But a checkable task is not the same as a checked one, and the gap between the two is where the benefit can stop.
Checkable Only Stays Checkable If the Review Is Designed In
An extracted value can be checked against its source. That is a property of the task. Whether anyone actually checks it depends on how the workflow around the tool is built.
Three things decide it. Each extracted value needs to stay linked to the source document it came from, so a reviewer can see the proof without searching for it. A named person needs to confirm the value before it is relied on. And when that person corrects a value, the correction needs to reach every place the value is held, so it is not fixed in one system and left wrong in another.
UK brokers will recognise the accountability point. The FCA says it does not plan to introduce extra regulations for AI and will rely on existing frameworks, and its page on AI notes that its rules emphasise accountability for senior managers and are relevant to the safe use of AI. It sets no intake-specific standard. That leaves the design of the review to each firm’s accountable people, which is why it belongs in the buying decision and not after it.
Genasys drew a similar practical conclusion from its BIBA 2026 conversations: use AI to take over a known, repetitive task, leave a person answerable for the outcome, and track the results.
A task being checkable does not mean it is being checked. The difference is whether the review step was designed before the tool was bought.
A desk that cannot show what a value was read from has automated the reading and lost the proof.
Conclusion
Nearly 90 percent of the brokers Guidewire surveyed describe the market as soft or softening, and they are paying closer attention to efficiency and ease of doing business when they choose insurers. In that market, hours of rework at the start of a placement are harder to absorb, and harder to explain to a client.
The desks that gain most from intake automation will not be the ones with the best extraction tool. They will be the ones that decided before buying it who confirms each value and against what, so the time saved at the start is never repaid as rework later.
Key Takeaways
| Five things to retain from this article |
|---|
| 01 42 percent of surveyed London Market brokers named automating submission intake and data extraction as the most valuable AI use case today (Guidewire, 2026) |
| 02 55 percent of the same brokers are concerned about bias and transparency in AI-driven risk selection, so AI that supports a decision earns more confidence than AI that makes one (Guidewire, 2026) |
| 03 Intake suits early automation because an extracted value can be checked against its source document and corrected before anything depends on it |
| 04 A checkable task only stays checked if the review is designed in: values linked to their source, a named person confirming them, and corrections applied everywhere the value is held |
| 05 The FCA says it does not plan extra AI regulations and relies on existing frameworks, and that its rules emphasise senior manager accountability (FCA, AI and the FCA: our approach) |
Frequently asked questions
Genasys’s practical position is to use AI where it removes a known, repetitive task, keep a person accountable for the outcome and measure the result. For intake, one simple measure is to compare a sample of extracted values against their source documents and record how many needed correction, by type of field, over time. That method is a suggestion drawn from the argument above, not a sourced standard.
Not in so many words. The FCA states that it does not plan to introduce extra regulations for AI and relies on existing frameworks, and its page on AI says its rules emphasise accountability for senior managers. It sets no intake-specific requirement. Whether and how a given value is checked is a decision for the firm’s accountable people.
No. It comes from 251 brokers who deal primarily with London Market insurers, surveyed by Censuswide in November 2025. It describes that market. Brokers elsewhere may rank use cases differently, which is why the argument in this article rests on how intake behaves, not on the number alone.
Glossary
| Key terms used in this article |
|---|
| Insurance premium tax (IPT) An indirect tax on most non-life insurance premiums, set nationally in Europe, usually settled by the insurer and passed on to the policyholder. |
| Location of risk The rule that decides which country may tax a premium, based on where the insured property, vehicle or policyholder is situated rather than where the broker or insurer is based. |
| Debit Note The document a broker issues to a client showing premium, taxes and charges due, and the starting point for reconciliation with insurer statements. |
| Effective-dated tax rate A tax rate stored with the dates it applies from and to, so the system charges the correct rate based on when a premium falls due. |
| Parafiscal charge A levy added to an insurance premium to fund a specific public purpose, such as fire services or guarantee funds, on top of the standard premium tax. |
| Regulatory return A periodic report a broker submits to its regulator, in the format and on the schedule that regulator sets. |
Ready to see how intake data is handled in practice?
When your intake tool reads a submission, who confirms what it read, and against what?
Agiliux was built by people who worked inside broking and reinsurance, with integration designed into the architecture from the outset. Its data and document automation ingests documents and flows the structured data into the core system, so what is read at intake is the record the rest of the workflow works from.
Want to see what that looks like for your own submissions?
Sources cited
- Statistic: Guidewire Software, “London Market Tech Barometer 2026: Key Survey Insights,” Guidewire blog, 24 February 2026 (survey of 251 insurance brokers who deal primarily with London Market insurers, conducted by Censuswide, 19-27 November 2025, methodology per the Business Wire release of 19 February 2026). GuideWire and BusinessWire
- Qualitative finding (not a statistic, vendor-authored account of conference conversations): Genasys, “BIBA 2026: 10 Valuable Insights from the Conference,” reporting conversations at the British Insurance Brokers’ Association conference, Manchester, 13-14 May 2026, published 19 May 2026. Genasystech
- Regulatory reference (not a statistic): Financial Conduct Authority, “AI and the FCA: our approach,” first published 8 September 2025, page last updated 2 October 2026. fca.org
