AI quotations for the skilled trades
Prepare quotations with AI: organise enquiries, draft line items and assess drawings. Documented examples and a process that can be checked through to approval.
First version: Updated:
Read articleKnowledge for companies
AI in skilled trades: uses in the office and on the job
AI in the skilled trades can organise customer enquiries, prepare reports and make invoice data available for review. A useful starting point is a recurring task with clear inputs and a result the business can check. Responsibility for measurements, costing, appointments and approvals remains part of the workflow.
In daily work
How to use the solution.
A photograph without measurements, an incomplete call note or an invoice without a purchase-order reference creates follow-up work. Assess AI by whether it helps the next person receive a more complete record.
Overview
Build on the digital work already in place
In Bitkom’s 2025 survey, 4 percent of trade businesses said they used AI. Sending quotations digitally was already common at 68 percent, as was sending invoices digitally at 62 percent. The representative weighted survey covered 504 German trade businesses with at least one employee, interviewed by telephone in calendar weeks 23 to 29 of 2025. The report was published in 2026.
Start by reviewing your existing trade software. A shared folder, a required enquiry field or a notification after approval may solve the immediate problem. AI becomes relevant when the task involves free text, images or varied documents. The survey figures describe technology adoption, not the financial returns from using it.
Overview
Which task makes a suitable starting point?
Use this mapping to assess work in your own business. It describes possible support and the checks to plan alongside it.
| Task | Possible support | What people need to check |
|---|---|---|
| Enquiries | Turn free text and notes into a structured enquiry | Address, trade, scope and unanswered questions |
| Quotations | Find approved line items and prepare a text draft | Quantities, prices, technical specification and terms |
| Work reports | Organise notes and associate them with the job | Work actually performed, materials and time details |
| Telephone | Capture the request and callback details | Correct recipient, urgent cases and handover to a person |
| Invoices | Extract unstructured data and flag discrepancies | Purchase-order match, prices, duplicates and approval |
Overview
Clarify customer wishes before the consultation
In DigiCast on 18 March 2025, Anke Freund described Tischlerei Dein Freund’s furniture chatbot: it collects preferences and generates images for a consultation. Images can be wrong. This is an individual reported example, not a T-NEX reference.
A similar process starts with the questions your staff already ask. For a fitted wardrobe, these include the installation setting, intended use, dimensions and access to the room. Pass the answers and concept image on with the enquiry. An elegantly written conversation record offers little help if the customer has to explain the installation location again at the appointment.
Overview
Complete office work in the right job record
A language model can turn notes into an email draft or an organised work report. A useful instruction identifies the recipient, available facts and type of text required. Missing information should remain a question. An assumed task, material used or appointment must not become a statement that the work was completed or the date agreed.
Filing is part of the task. An approved report needs the correct job, processing status and underlying notes. If staff generate a text and then reconstruct it across several windows, much of the office work may still remain. Test the complete path to the saved or sent version.
Overview
Use the telephone to prepare a useful callback
A phone assistant can be assigned a narrow task: collect the enquiry, job location and callback details, then pass them to the responsible person. Its greeting should make clear that the caller is speaking to an automated assistant. Customers need a simple way to reach a person when the conversation cannot progress.
Test the process against situations from your trade: an unclear address, background noise, an interrupted call and a question about an existing job. The assistant should confirm an appointment only when it can actually check the information authorised for that purpose. Urgent requests need a predefined route to the right person. Also decide which conversation data will be stored and who may handle it.
Overview
Include the work of checking invoices
On DigiCast, 15 October 2025, Heiko Süthoff described invoice matching at Günter Terfehr Bautechniker: extracted items are compared with prices; staff review and approve. Special items require clarification. This interview example is not a T-NEX project.
This suggests a practical approach: trace one invoice from receipt to approval. Record which information is missing, which prices form the comparison and who decides what happens to a discrepancy. Extraction is one component of that process. With structured invoice data, most of the remaining work may concern matching records and handling exceptions.
Overview
Set up the first trial with the people doing the work
Choose a recurring task whose result someone with relevant expertise can assess. Collect typical cases alongside those that currently require a follow-up question. For the first trial, use documents whose use in the proposed system has been cleared. Define the expected output and the circumstances in which the application should return the task to a person.
The person who writes quotations, answers calls or checks invoices every day should help select the test cases. Allow time for comparing results, explaining errors and trying changed procedures. A named deputy prevents questions and corrections from stalling when that person is at a job site or on leave.
- Choose one recurring task and the person responsible.
- Select typical records and difficult exceptions with the team.
- Agree access, confidentiality, storage and permitted use of the data.
- Write down the expected result and the route for follow-up questions.
- Compare the complete task, including correction and filing.
Overview
Measure the benefit in completed work
For the same type of task, record active processing time before and during the trial. Include follow-up questions and corrections. For a phone enquiry, check whether the note reached the correct job and whether the callback could proceed with the information needed. For a quotation, assess the approved line items as well as the speed of drafting the text.
Compare the result with ongoing maintenance, technical support and usage costs. Expand where the full process works better and staff can handle its exceptions. When adding another document type or trade, reassess its requirements using examples specific to that work.
FAQ
Frequently asked questions
Does a small trade business need its own AI model?
A first text draft generally does not require a model trained by the business. Start with existing software and a defined task. An additional application becomes useful when data must be brought together from several systems or specific business rules need to be implemented.
Which task makes a good starting point?
A recurring task with accessible information and a result that can be checked. A complete enquiry or a verifiable invoice comparison is easier to assess than the broad goal of automating the office.
Can AI images be used directly as manufacturing drawings?
A generated image can explain a design idea. Dimensions, materials, construction and installation must be specified and checked separately. Manufacturing needs an appropriate, technically approved document.
How can we tell whether the trial helps?
Compare completed tasks of similar difficulty. A benefit emerges when total effort, including corrections, falls or results become more reliable. The speed of text generation alone does not answer that question.
What does introducing AI in a trade business cost?
Costs depend on existing software, required connections, data preparation and ongoing support. Ask for implementation and recurring costs separately. A limited trial should establish whether the benefit after review and corrections justifies those costs; a general saving cannot be established in advance.
Sources & further reading
- Bitkom: Digitalisation of the skilled trades, 2025 study, 2026 report, pp. 15, 22 and 33 (German)bitkom.org
- Mittelstand-Digital Zentrum Handwerk: DigiCast episode 41 with Anke Freund, 18 March 2025 (German)open.spotify.com
- Mittelstand-Digital Zentrum Handwerk: DigiCast episode 47 with Heiko Süthoff, 15 October 2025 (German)podcasts.apple.com
- T-NEX: AI automation for SMEsT-NEX
Related options
You may also be interested in these.
Automating incoming invoices
Automate incoming invoices from receipt and data checks through approval to accounting: review matching, duplicates, exceptions and the final handoff.
First version: Updated:
Read article
AI chatbot for companies with source references
Internal knowledge search and chatbots built around your content: approved answers or RAG with a language model. T-NEX scopes access, data and implementation.
Learn moreAn AI pilot with defined evaluation criteria
Evaluate one AI use case with T-NEX: limited scope, suitable data and agreed criteria for expansion or stopping.
Explore serviceWhich task would you like to solve next?
Bring a concrete task. Together, we will define what the application needs to do.
Discuss a task from your business