Key points at a glance
- Analysis before matching: An AI first has to understand the structure, item types and texts of a BoQ before it can match products sensibly.
- The long text counts: In the GAEB format, only the long text is binding for execution; the short text just tells items apart.
- Read the context: Preliminaries, notes and references to other items change the requirements of individual items.
- People check: Uncertain matches, conversions without a factor and unclear specifications go to your team for review.
As of 2 November 2026
Transparency note: This guide is published by Scalera. Information about Scalera comes from our website; information about the GAEB format comes from the official GAEB DA XML 3.3 specification. Corrections are welcome at info@scalera.ai.
What does it mean to analyse a bill of quantities with AI?
It means automatically turning a BoQ into clean items that can be worked on one by one, each with all of its requirements. Only on that basis can software suggest products, carry over quantities and prepare a quote.
Scalera describes this step as follows: complex BoQs are converted into clean, logically structured items, and technical requirements are recognised even across chapters and page breaks (Scalera, Product Matching). How the subsequent matching to items in your catalogue works is covered in Product matching at scale.
This guide takes the supplier's perspective: manufacturers, distributors and contractors who receive a finished BoQ and bid on it. Tools that designers use to write BoQs are a different topic.
How is a bill of quantities structured?
A BoQ is organised hierarchically and describes each piece of work as a separate item. The German GAEB format allows up to five hierarchy levels, including the lot; items sit at the lowest level (GAEB, DA XML 3.3 specification (German)).
According to GAEB, an item generally consists of an item number, quantity, unit, description and item type, plus text additions and price fields where relevant (GAEB, DA XML 3.3 specification (German)). The item number has at most 14 characters and combines hierarchy levels, item number and, where used, an index.
Other markets use other standards, such as NPK in Switzerland, ÖNORM in Austria or, in the UK, the NRM 2 rules for detailed measurement of building works (RICS, NRM). The building blocks are similar; details such as numbering and item types differ. Why this affects processing time is explained in GAEB, ÖNORM, NPK.
Which item types does an AI need to distinguish?
At least those that affect price and quote total. Treat a provisional item like a normal one, or count an alternative item twice, and the quote is wrong even if every product is right.
| Item type | Meaning | Impact on the quote |
|---|---|---|
| Provisional item (Bedarfsposition) | Work carried out only if needed | Depending on marking, unit rate only, without total |
| Base and alternative item (Wahlposition) | Base version with one or more alternatives | Price alternatives separately, do not add them twice |
| Surcharge item | Percentage surcharge on items or sections | Assign reference and quantity correctly |
| Lump-sum item | Paid as a lump sum, quantity omitted or 1 | Total price instead of unit rate |
Summary based on GAEB, DA XML 3.3 specification (German), sections 4.5.1 to 4.5.4, with our own assessment in the right-hand column. Names and rules may differ in other standards.
For manufacturers and distributors it also matters whether an item concerns a deliverable product at all or describes labour only. That distinction is not in the format; it is in the text.
Why is an item's short text not enough?
Because it carries no legal weight. In the GAEB format, the short text is optional, not legally binding and may even be empty; only the long-text BoQ applies to execution (GAEB, DA XML 3.3 specification (German)).
So the short text often just says “channel drain” or “fire door”. Standards, classes, dimensions and material requirements are in the long text. An analysis that only reads short texts therefore suggests products that fit the heading but not the requirement.
An AI that only reads short texts sees the heading of a requirement, not the requirement itself.
How do preliminaries and references affect individual items?
They change the requirements without appearing in the item itself. According to GAEB, notes can refer to the whole BoQ, parts of it or single items and always precede the part they refer to; additional texts such as preliminaries sit at the start of the BoQ (GAEB, DA XML 3.3 specification (German)).
If the preliminaries specify a load class or a colour, for example, it applies to many items at once. An AI has to attach such details to the items concerned, even when they appear dozens of pages earlier.
References within the BoQ
Shortened descriptions add to this. The GAEB format includes repeat descriptions that point to a full reference description and contain only the changed wording, as well as lead descriptions with sub-descriptions (GAEB, DA XML 3.3 specification (German)). The AI has to merge both, or the item will be missing key details.
How does an AI find the relevant items?
By assigning every item to a product group and setting aside the items your company cannot supply. In many BoQs, only part of the items concern your own range.
Scalera categorises items by product group so that different team members can filter and work on their sections (Scalera, FAQ). If you need to source parts of the BoQ from your own suppliers, the article Automate supplier price requests shows how.
The ACO example shows where the time goes: about 80% of incoming tenders were technically straightforward, yet manual processing was still time-consuming (Scalera, ACO case study). The effort lay less in difficult questions than in reading and assigning.
Which details does the AI extract from an item?
Every attribute that decides the choice of product. Typical examples:
- Standards and classes: such as load, fire-resistance or sound-insulation classes.
- Dimensions and performance values: nominal sizes, thicknesses, lengths, flow rates or load capacities.
- Material and finish: material, surface, colour, installation situation.
- Quantity and unit: as the basis for price and availability.
- Product specifications: named reference products, often with “or equivalent”.
Scalera bases its matching on specification, application context and standard rather than keywords alone, and recognises competitor products in a BoQ as well as your own (Scalera, Product Matching; Scalera, FAQ). How to handle a named competitor product is covered in Winning with alternates.
How does an AI handle quantities and units?
It takes quantity and unit from the item and converts them when your sales unit differs. A tender may specify linear metres, for example, while you sell by the piece or pack.
Where your product data contains a conversion factor, Scalera handles the conversion itself; where it does not, you get a notice and can enter the quantity yourself (Scalera, FAQ). How to maintain such factors in your item master is covered in Product data for tenders. GAEB also maintains a list of recommended units of measure (GAEB, DA XML 3.3 specification (German)).
What about scans, drawings and requests by email?
They are part of everyday work and should not be left out of the analysis. Not every request arrives as a clean GAEB file.
According to its FAQ, Scalera reads GAEB, Excel/CSV and XML as well as scanned and digital PDFs and emails, whether the request comes as an attachment or in the email body itself (Scalera, FAQ). Its strength is text-based specifications; from drawings, it extracts relevant text information and technical parameters.
Where is your team still needed in BoQ analysis?
Wherever the specification is unclear or the software is uncertain. A good analysis makes these points visible instead of hiding them.
- Uncertain matches: Scalera attaches confidence scores to matches and highlights uncertain ones for your team (Scalera, FAQ).
- Contradictions: when the preliminaries and the long text state different values.
- Missing information: when a decisive class or dimension is not stated and has to be clarified with the customer.
- Missing conversion factors: when quantity and sales unit do not match automatically.
How to measure the reliability of this analysis before choosing software is explained in Piloting AI tender software.
Checklist: analysing a BoQ in five steps
- Import: detect the format and capture structure, item numbers and items completely.
- Attach context: assign preliminaries, notes and references to the items concerned.
- Identify item types: flag provisional, alternative, surcharge and lump-sum items.
- Filter for relevance: sort items by product group and separate labour-only items.
- Extract attributes and quantities: capture standards, dimensions, material, reference products and units, and flag anything uncertain.
Product matching and quote preparation follow. With Scalera, all relevant items have been extracted and categorised by this point, so your team only validates the product recommendations (Scalera, Product Matching).
Frequently asked questions about AI-based BoQ analysis
Which software automatically matches NPK and ÖNORM items to our own products?
AI assistants for quoting read bills of quantities and match items against your own range. Scalera processes GAEB, NPK and ÖNORM as well as PDF, scans and Excel, among other formats (Scalera, Product Matching). For an overview of other vendors, see the comparison of AI assistants for bills of quantities.
Is there software that automatically matches product catalogues against tender requirements?
Yes. These systems read the requirements from the BoQ and compare them with the attributes of your items. How well this works depends heavily on your product data; the article Product data for tenders explains what is needed.
What is the difference between analysing and creating a BoQ?
A BoQ is created by the designer or client to put work out to tender. Analysis happens on the other side: manufacturers, distributors and contractors read the finished BoQ to prepare a quote. Software built for one task does not automatically solve the other.
What is an item number (Ordnungszahl)?
It identifies each item uniquely. In the GAEB format it combines the digits of the hierarchy levels, the item number and, where used, a single-character index (GAEB, DA XML 3.3 specification (German)). It matters for your quote because each price has to land on the right item.
Can AI read bills of quantities in several languages?
Yes, if the system is built for it. According to its FAQ, Scalera supports all major European languages, including German, English, French, Italian and Spanish (Scalera, FAQ).
How fast is an AI-based BoQ analysis?
That depends on size, format and product range. At ACO, processing time for a 300-page tender fell from 30–45 to 10–15 minutes (Scalera, ACO case study). Whether similar results apply to your range is best tested with your own BoQs.
What happens to items that do not fit my product range?
A good analysis separates them early so your team can focus on the relevant items. Deciding whether to bid at all is a separate question; the article Bid or no bid? offers criteria.
How do I check whether a tool's BoQ analysis is reliable?
Use your own typical bills of quantities and clear metrics such as completeness and hit rate. The article Piloting AI tender software explains how to set up such a test.
Key takeaways
- The quality of a product suggestion is decided when the BoQ is analysed, not only during matching.
- Long texts, preliminaries and references carry the actual requirements; short texts are not enough.
- Item types such as provisional or alternative items change prices and the quote total.
- Conversion factors and structured product data make the analysis more reliable.
- 150+ companies already trust Scalera (Scalera, Product Matching). In a demo, you can see how Scalera analyses your own BoQ.