Skip to content
Blog

Can ChatGPT analyse a customer specification?

An 80-page customer specification, and the quotation has to go out by Friday. What ChatGPT does well in the analysis, where it reaches its limits and what to settle before a customer specification ends up in a chat window.

Yanik YeganehfarCo-Founder & COO5 min read
Quick answer

ChatGPT can summarise a customer specification, list requirements and draft questions for the customer. It reaches its limits with long documents full of tables and drawings, with reliable page references and with comparisons against your earlier projects. Because specifications are confidential, one belongs only in an approved company account, never in a private one.

Contents
  1. Can ChatGPT analyse a customer specification?
  2. What ChatGPT does well here
  3. Where does ChatGPT reach its limits with a customer specification?
  4. How do you use ChatGPT sensibly, if you are allowed to?
  5. ChatGPT and a specialised AI compared
  6. How does KoAssist analyse a customer specification?

Thursday afternoon. The customer specification for an inspection system runs to 84 pages, plus three annexes with the layout, part drawings and the customer's plant standard. The quotation has to go out by Friday. The obvious idea is to drop the PDF into ChatGPT, type "Summarise the requirements" and be done.

The idea is not wrong. It just has a few catches you should know about before the quotation is built on that summary.

Can ChatGPT analyse a customer specification?

Yes, in part. Current language models read long PDFs, recognise the structure of a specification and can turn it into a usable overview. For a first look, that helps.

An analysis for clarifying a quotation is more than a summary, though. It has to capture every requirement, find contradictions and gaps, prepare questions for the customer and pull out deadlines and acceptance conditions. How such an analysis works by hand is described in the article on analysing customer specifications. Any tool has to be measured against that standard.

What ChatGPT does well here

Summarising. An overview of the scope of supply, the chapters and the main points of the specification in a few minutes.

Listing requirements. Turning running text into a list, sorted by mechanical, electrical, software and documentation.

Drafting questions. Turning an unclear passage into a polite, precise question for the customer.

Explaining terms. For example when a plant standard uses abbreviations nobody in your own company is familiar with.

Where does ChatGPT reach its limits with a customer specification?

Sources. For a quotation, what counts is not only what a requirement says but where it is written. Without a reliable reference to page and section, someone has to look up every important detail in the original. For numbers, tolerances and references to standards that should happen anyway.

Tables, annexes and drawings. Many requirements are not in the running text but in tables, in the layout or in a footnote in an annex. This is where details are easily missed or mixed up.

Contradictions across documents. The specification says IP54, the layout puts the installation site in the splash zone, the plant standard calls for IP65. Contradictions like these arise between documents, and they are the most important findings of an analysis.

Your own experience. ChatGPT does not know your earlier projects. The question of how you solved the part feeding on the last order for the same customer stays open.

Confidentiality. A customer specification contains the customer's information, often under a confidentiality agreement. It does not belong in a private account. What to settle before using ChatGPT in the company is covered in the GDPR check on ChatGPT for internal documents. The overview of GDPR-compliant ChatGPT alternatives shows what else is available.

How do you use ChatGPT sensibly, if you are allowed to?

Assuming your company has an approved account and the agreement with the customer allows it:

  • Ask for sources. Have it name the page and section for every requirement, and spot-check them in the original.
  • Work in sections. Analyse chapter by chapter rather than the whole document at once.
  • Treat annexes separately. Query tables and layouts on their own and compare them with the main text.
  • Always check the numbers. Cycle times, tolerances, protection ratings, deadlines.
  • Treat the result as a draft. The analysis is a starting point for clarification within the team, not a finished functional specification.

ChatGPT and a specialised AI compared

Task ChatGPT on an approved plan Specialised AI with project folder
Overview of a customer specification good good
Drafting questions for the customer good good
Reference to page and section on request, check it for every detail
Contradictions between specification, layout and emails only what was uploaded across the project folder
Questions about earlier projects no yes, from your own documents
General texts, translations very good not what it is for

Many teams use both: a general tool for texts and a specialised AI for working with project documents.

How does KoAssist analyse a customer specification?

KoAssist is the AI agent for machine and plant builders. Two functions help with a customer specification.

The knowledge assistant is live. You ask questions of the specification and the project folder, such as "What protection rating is required for the control cabinet?", and get the answer with file, page and section. This also works with scans and drawings and across earlier projects.

Project Memory fills a project profile from the specification, with every detail backed by its source. Whatever is missing or contradictory goes on a list of questions for the customer. Project Memory is currently in pilot operation with existing customers.

The data is held in data centres in Germany, and training on customer data is contractually excluded. The details are on the security and EU hosting page. If you would like to see how this works with one of your own customer specifications, you can book a demo.

Written by

Yanik Yeganehfar

Co-Founder & COO

On LinkedIn ↗

Frequently asked questions

Yanik Yeganehfar, co-founder of KoAssist

Find out with Yanik whether KoAssist fits your projects.

Request a demo