Blog · Published · Updated
Two addresses on an insurance form: can AI tell them apart?
Short answer
With insurance forms the most common error is a value read correctly but put in the wrong field, such as the head office entered as the warehouse location. First check a reading template and a ready-made extractor with examples. Adapt the model when assignment errors return on legible forms. At 2,000 forms a month, an extractor cuts errors from 15% to 7% and fine-tuning to 3%, saving 24 hours, or €838 a month.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- model on your server or in the cloud in the EU
- quoted after the review
- same work in application with EU cloud model
- from €3,000
- task, data and equipment analysis
- €0
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 3 min read
Statistics
Insurance and AI: statistics 2026
24 h
fewer corrections a month at 2,000 forms: a ready-made extractor with examples cuts errors from 15% to 7%, fine-tuning to 3%, at 6 min per correction. Scenario.
14–56%
According to Anthropic (2025), studies with a control group found smaller time savings, from 14% to 56%.
Source: Anthropic, Estimating AI productivity gains (2025) ↗
90%
According to the European Commission (2026), 90% of companies using AI say people take the decisions, and 2% of decisions are fully automated.
Source: European Commission, Flash Eurobarometer 578 (2026) ↗
20.0%
According to Eurostat (2025), 20.0% of EU companies with at least 10 employees used AI.
Source: Eurostat, 20% of EU enterprises use AI technologies (2025) ↗
55.0%
According to Eurostat (2025), 55.0% of large EU enterprises with 250 or more employees used AI.
Figures checked: 22 October 2026
The head office and the warehouse address
A commercial property insurance form contains two addresses. The first is in the company details section and gives the head office, the second in a separately marked section about the warehouse being insured. The tool read the address with no typo, but entered the head office as the warehouse location too. The fix means going back to the document.
A correct record takes the warehouse location from the warehouse section and shows, beside every value, the passage of the form it was read from. If the warehouse section is empty, the head office address does not fill the gap, and the employee sees that the location needs clarifying. A form with two warehouses needs two groups of data, with the address and the description at the right location.
- 01Readingthe page text, including from scans
- 02Sectionscompany details kept apart from locations
- 03Fieldseach value with its source passage
- 04Emptya gap marked for clarification
- 05Importas many locations as the form has
How much a simpler change gives, and how much training adds
- Today: 15% errors × 2,000 × 6 min
- 30.0 h (€1,047)
- After a change without training (a ready-made extractor with examples): 7%
- 14.0 h (€489)
- After fine-tuning: 3%
- 6.0 h (€209)
- The simpler change gives 16 h, fine-tuning adds another 8 h (€279) a month
Compare fine-tuning with the cost of examples, training and re-evaluation
The error shares and the time per correction are the scenario's assumptions, which we measure on the client's cases not used for training. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat.
In this count a ready-made extractor with examples gives most of the effect for a fraction of the cost. Training makes sense at a larger scale or when the remaining errors are expensive. We count errors separately for misreading and for wrong assignment, because each is fixed in a different place.
- Evaluation cases kept apart from the training examples
- The same set for the current model, the simpler change and the adapted model
- Correct answers prepared by the people who correct the results today
- Errors counted separately for each type of mistake
- A plan to re-evaluate after a change of the base model or the process
The set of training examples has to be maintained: when the process, the naming or the base model changes, the evaluation is repeated on the same test set. We include that cost in the quote together with the training itself.
What to check before training the model further
With a fixed form layout, an existing reading template is enough. If the company collects answers through a web form, separate fields for the head office and the locations will reduce the need to extract data. With scans, tell misread text apart from correct text in the wrong field, because OCR that loses a section heading spoils the assignment that follows. Google describes the Custom Extractor, built on a foundation model that can be adapted with examples. If the import into the system allows only one location, improving the model will not remove the problem.
A fixed form layout
a reading template
configuration
Varied layouts, typical errors
a ready-made extractor with examples
an app from €3,000
Assignment errors on legible forms
a model adapted on the team's corrections
priced after a comparison
Where to start
We start with the forms employees correct regularly. The team explains what belongs in each field and how it splits the locations, and we assess the trial on forms not used for adapting the model. We adapt models as part of private LLMs, and we build the app where an employee sees the result next to the source as part of AI apps. For the free analysis, a few examples of an error that keeps coming back and a note on who corrects it today are enough.
Worked example · Scenario
How we count the payback: data from forms
We run the same count for your team in the free analysis.
€1,745
a month for copying 1,000 documents by hand
Payback: 2 mo.
3 minutes to copy the fields from a document and checking in 15% of the old time are our assumptions. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat, for companies with 10 or more employees. Model cost about €0.007 per document: OCR in our system cost €1.17 for 181 files, and a separate read about €0.0008, converted at the NBP rate of 21 October 2026 and the ECB rate of 1.1205 USD per euro on 22 October 2026.
See the full calculation →
