Blog · Published · Updated
Local AI for manufacturer warranty documents
Short answer
A local AI assistant puts the serial numbers from the proof of purchase, the rating plate photo and the inspection report side by side in one view and flags a mismatch before an employee assesses the warranty claim. It does not decide whether the claim is justified. With 500 claims a month at 8 minutes of comparing each, a €3,000 app with a GPU server pays back in 2.0 months.
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
Warranty claims and AI: statistics 2026
2.0 mo.
payback of a €3,000 local app at 500 claims a month: €2,327 of comparing documents today, €796 with checking and a GPU server. Scenario.
47%
According to the European Commission (2026), customer service is the most common use of AI among EU SMEs and mid-caps that already use it, at 47%.
Source: European Commission, Flash Eurobarometer 578 (2026) ↗
15%
According to Brynjolfsson, Li and Raymond (2025), customer service agents supported by AI resolved on average 15% more issues an hour, in a study of 5,172 people.
Source: Brynjolfsson, Li and Raymond, The Quarterly Journal of Economics (2025) ↗
−87%
According to Anthropic (2025), AI cuts the time of tasks on invoices, memos and other documents by an estimated 87%.
Source: Anthropic, Estimating AI productivity gains (2025) ↗
14–56%
According to Anthropic (2025), controlled studies found smaller time savings, from 14% to 56%.
Source: Anthropic, Estimating AI productivity gains (2025) ↗
Figures checked: 22 October 2026
The same model name, two serial numbers
In the files of one warranty claim, the proof of purchase shows serial number A and the workshop inspection report shows serial number B. The model name is the same. The employee wants to compile the material as the documentation of one unit and notices the difference only at the third file.
The assistant shows both numbers next to their sources and flags the mismatch. The employee opens the originals, checks the markings and asks for a report on the unit from the proof of purchase or for an explanation of the difference. A matching model name does not confirm that the inspection concerned this unit.
| Document | What the assistant reads | What the employee sees |
|---|---|---|
| Proof of purchase | model, serial number, date | the value next to a crop of the document |
| Rating plate photo | serial number | an enlarged part of the photo, an uncertain character marked |
| Inspection report | unit number, fault description | the mismatch with the proof of purchase |
| Corrected report | the new number | the version next to the previous one, with the employee's note |
- 01Openall the claim documents in one view
- 02Comparethe numbers and model from each source side by side
- 03Mismatchthe assistant flags the difference and uncertain characters
- 04Clarifythe employee asks the customer or the workshop for the right document
- 05Assessthe person handling the case judges whether the claim is justified
What image quality changes
If the number in a photo is illegible, the app leaves it to be checked in the original. A confident-looking, wrongly read character would make it harder to find the right document. Reading text from an image is a separate part of the app, and local processing covers both that stage and the model that compares the text.
The system links documents to the serial number and opens them side by side
the system's current features
no build
Codes in attachments described differently, compiled by hand
a local assistant with a source view
from €3,000
Recurring faults across many claims
grouping descriptions, a separate project
priced after the review
What local AI for warranty documents costs and when it pays back
- Today: 500 × 8 min
- 66.7 h × €34.90 = €2,327
- Checking the comparison, 25% of today's time
- €582
- Hetzner GEX45 server
- €214
- Saving
- €1,531 a month
Payback of €3,000
2.0 months
The volume, the time and the checking share are the scenario's assumptions. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat. Price of the GEX45 with 24 GB of GPU memory from Hetzner's website on 22 October 2026, without the one-off setup fee of €209. Through the API, Claude Sonnet 5.5 would cost €10 a month at 8,000 input tokens and 600 output tokens per claim, converted at the ECB rate of 1.1205 USD per euro on 22 October 2026.
The difference between the server and the API is the price of keeping customer documents on your own infrastructure. At 150 claims a month a local server pays back in 9.7 months, and below about 135 claims it takes more than a year. Then the API works better, if your data rules allow it.
The local model and the rest of the app
Ollama describes a local mode that switches off cloud models and web search. That is a property of the program that runs the model. Where attachments are stored, the history of questions, backups and service access need agreeing for the whole app, because the same permissions cover the attachments and the comparisons prepared from them.
The first scope covers one type of claim and the documents the team compares most often. We build the local model as part of private LLMs, and the case view as part of AI apps. For the free process review, a description of a case where it took a long time to establish whether the attachments concerned the same unit is enough.
Worked example · Scenario
How we count the payback: reading data from documents
We run the same count for your claims in the free process review.
€2,618
a month for copying 1,500 documents by hand
Payback: 1.4 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 →
