Blog · Published · Updated
Polish and English in one note: will AI preserve the meaning?
Short answer
With notes written in Polish and English, AI has to keep the state of a case: „ordered” and „ETA to be confirmed” do not mean „in transit”. In a scenario of 800 service notes a month, a glossary and instructions cut errors from 12% to 6%, and fine-tuning to 2%, which together save 6.7 hours of corrections, or €233 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
Bilingual text and AI: statistics 2026
6.7 h
fewer corrections a month at 800 service notes: a glossary and instructions cut errors from 12% to 6%, fine-tuning to 2%, at 5 min per correction. Scenario.
40%
According to a study published in Science (2023), ChatGPT cut the time 453 professionals spent on writing tasks by 40%.
Source: Noy and Zhang, Science (2023) ↗
15.1%
According to Eurostat (2025), 15.1% of people aged 16 to 74 in the EU used generative AI tools for work.
−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), studies with a control group found smaller time savings, from 14% to 56%.
Source: Anthropic, Estimating AI productivity gains (2025) ↗
Figures checked: 22 October 2026
What the note actually says about the part
A ticket holds this note: „Replacement part ordered. ETA nadal do potwierdzenia. Klient prosi o update po potwierdzeniu terminu”. The Polish half says the ETA is still to be confirmed and the customer wants an update once the date is confirmed. ETA means the estimated delivery date. The AI summary reads: „The replacement part is already on its way to the customer”. That changes the state of the case: the part was ordered, but nothing confirms it is in transit. The correct version: „The replacement part has been ordered. The estimated delivery date needs confirming. The customer asks for an update once it is confirmed”.
In a mixed-language team a summary may be needed in Polish or in English, and both have to keep the same distinction between ordering, shipping and delivery.
- Entries not used to adapt the model
- Summaries in both languages you need
- Corrections marked by staff who know both languages
- Errors in the state of a case counted separately from errors in names
Microsoft describes the dynamic dictionary in Translator, which keeps company part names. A glossary looks after names, but it will not keep the meaning of a whole sentence: the part name can be translated perfectly while „ordered” still turns into „in transit”.
What a simpler change gives, and what training adds
- Today: 12% errors × 800 × 5 min
- 8 h (€279)
- After a change without training (a glossary and instructions): 6%
- 4 h (€140)
- After fine-tuning: 2%
- 1.3 h (€47)
- The simpler change saves 4 h, fine-tuning adds 2.7 h (€93) a month
Compare fine-tuning with the cost of examples, training and a new evaluation
The error shares and the correction time 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 glossary and instructions give a large part of the effect for a fraction of the cost. Training makes sense at a larger scale or when the remaining errors are expensive.
- Evaluation cases kept apart from 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 kind of mistake
- A plan for a new evaluation after a change of base model or 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 put that cost in the quote together with the training itself.
Try the existing model first
Before you order more training, state the task plainly: a short summary of the given note that keeps the state of the case. When errors keep coming back, you need specific corrections from staff who know both languages and the service work. AWS describes supervised model customisation on such examples. Fine-tuning changes how a note is processed, but the current delivery confirmation still has to be in the material.
Errors in part names
a company glossary
no training
Errors in the state of a case are rare with good instructions
instructions with examples
no training
The same state error keeps coming back despite instructions
fine-tuning on the team's corrections
after a trial
Where to start with fine-tuning for mixed-language notes
The first trial covers the entries of one service team in both languages. We adapt models as part of our private LLMs, and we build the app where an employee sees the result next to its source as part of our AI apps. For the free analysis, a few examples of an error that keeps coming back and a note of who corrects it today are enough.
Worked example · Scenario
How we count the payback: AI in service work
We run the same count for your notes in the free analysis.
€3,490
a month for writing the reports for 400 jobs
Payback: 1 mo.
15 minutes to write a work report 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: reading a photo at €0.0042 (0.004734 USD), measured in our field documentation system, plus the description (Claude Sonnet 5.5, about 2,000 + 400 tokens), at the ECB rate of 1.1205 USD per euro on 22 October 2026.
See the full calculation →
