Trade and wholesale firms spend hours per quote on a task that is mostly lookup and arithmetic: which articles, which quantity, which price for this customer today. The inputs are scattered across survey notes, article catalogues, discount agreements, a daily metal quotation and old quotes. An AI step can assemble a draft with a price derivation for every line; a sales clerk or engineer checks and releases it. This is suggestion software, not an autonomous price setter.
Problem & value
- Today: a plumber's office (SHK) turns site notes into 30-100 quote items by hand, searching Datanorm catalogues of several wholesalers. A wholesaler's inside sales calculates customer-specific prices from discount groups, scale prices, project prices and a metal surcharge for cable. An engineering office derives a fee from construction cost, fee zone and service phases under the HOAI.
- AI value: faster first drafts, fewer forgotten items (fittings, small parts, disposal), fewer outdated or mistyped prices, consistent margins, and the know-how of the one experienced estimator becomes reusable. The shadow-process research puts the manual effort at roughly 18 h (wholesale price and terms calculation), 10 h (SHK quote items) and 8 h (HOAI fee) in the sampled companies.
- Scope boundary: not dynamic B2C pricing (see OPT-004), not make-to-order RFQs with CAD (AGENT-027, PRED-032), not quantity takeoff from plans (DOC-030). This use case is item and price assembly for catalogue-based trade.
Variants
- SHK / electrical trades: survey notes or voice memo → article proposals from the Datanorm catalogue.
- Wholesale (Grosshandel): price and terms calculation per customer, incl. scale, project and metal-surcharge logic.
- Engineering offices: HOAI fee proposal from project data; since 2021 the fee is freely agreed, so the tool proposes a range, not a "correct" value.
- Not covered: consumer price personalisation, tenders with GAEB bills of quantities (see DOC-015).
How it works
- input (survey notes / voice memo / photos / customer enquiry / project data)
- extraction of work items, quantities and units (LLM, structured output)
- article matching (embeddings + keyword search in Datanorm / supplier catalogue; similar past quotes)
- price rules (customer discount group, scale and project price, validity date)
- surcharges (metal quotation x copper figure, freight, minimum quantity)
- deterministic calculation (rules engine, not LLM) incl. margin check
- draft quote with price derivation and flags
- human review and release
- transfer into ERP / quote document
- LLM role: understands notes, proposes items and substitute articles, explains deviations. All arithmetic and price lookups are deterministic in the ERP or a rules engine; an LLM must never invent a price.
- Metal surcharge: the surcharge is a formula (copper figure in kg/km x (quotation + procurement cost - copper base) / 100, quotation in EUR per 100 kg, result in EUR per km; typical manufacturer catalogues, e.g. Lapp). The daily quotation comes from the wholesaler or an index; the model only selects the cable and the parameters [3][4].
- HOAI: the fee tables serve as orientation; a proposal comes from the base fee rate and agreed surcharges, the engineer decides [2].
- Human-in-the-loop: every line shows its source (catalogue, customer condition, past quote); the clerk accepts or corrects. Corrections feed the matching.
Data preparation
Which data: article master data and price lists (Datanorm 4/5 files, supplier portals), customer conditions (discount groups, project and scale prices), past quotes and orders with final prices, daily metal quotations, surcharge parameters per cable, and for engineers HOAI fee zones, past fee agreements and project data.
Where it lives: the ERP or craft software (e.g. Business Central, SAP Business One, weclapp, proALPHA, Lexware handwerk, Sander & Doll, Plancraft), wholesaler data portals, email attachments with price lists, and Excel sheets that hold the "real" discount rules.
Typical gaps: catalogues are imported as snapshots, so prices are only as current as the last file [1]; article descriptions differ between wholesalers; discount agreements live in emails or heads; nobody can export the final price logic from the ERP; past quotes were changed by phone without being updated.
Preparation steps: 1. Automate the Datanorm import and note the price date in each quote (PREP-005). 2. Document customer conditions and surcharge formulas as rules (PREP-016). 3. Export 200-500 past quotes with final prices as examples (PREP-020). 4. Build a test set of 30-50 real enquiries with the correct quote (PREP-018). 5. Clean article duplicates and unit mismatches (piece, m, kg).
Tools: ERP export, Datanorm import tool, spreadsheet for the rules and test set.
Rough share of effort: about 50 percent of the project. Getting data and rules out of the ERP is the main hurdle, not the model. See GUIDE-003.
Implementation paths
Which path for whom
- SMB - Path A: craft and quote software increasingly has AI drafting built in, uses your own Datanorm data and avoids integration work. Reported examples (vendor and blog statements, not independently tested) include a Sander & Doll AI module [5]; for Plancraft, ask the vendor which AI functions exist.
- Enterprise, developer and public audiences are not the target here; large wholesalers use their ERP's pricing engine and belong to a different problem.
Path A - Buy (recommended)
- Craft and quote software with AI drafting: Plancraft, Sander & Doll, and ERP-near tools with Datanorm and GAEB import (Lexware handwerk, TAIFUN, Craftview). AI functions differ by vendor and should be asked for explicitly [5].
- Check: Datanorm version and price-date handling, EU hosting and DPA (AVV), whether the AI proposes from your catalogue only, traceable price derivation, export of data.
- Choose A when: standard trade processes and one main software system.
Path B - Build on cloud APIs
- Pattern: LLM with structured output for item extraction, embeddings plus keyword search over the catalogue, a deterministic price service on the ERP API (Business Central, SAP B1, weclapp, proALPHA), n8n or Power Automate as workflow.
- Choose B when: wholesale price logic is complex (scale, project, surcharge) and the ERP has a usable API.
- Keep the price calculation in the ERP; the LLM only calls it.
Path C - Self-host
- Open-weight LLM (e.g. Qwen or Mistral family, served via vLLM or Ollama) plus an open embedding model and OpenSearch or pgvector. HW2 is enough for a 7-30B model because the volume is small. Choose when customer terms must not leave the premises.
Path D - Custom training
Rarely sensible. A tabular model for margin or win-probability suggestions needs several thousand quotes with outcomes; most SMBs do not have them.
State of the art (2026)
- Feasible today as a suggestion step: drafting items from notes and matching them to catalogue articles works in products; reported time savings per quote come mainly from vendors and are not independently verified [5].
- Weak points: unit and quantity errors, wrong substitute articles, outdated prices from old catalogue files, rare special items, and cases with implicit customer-specific rules.
- Maturity: early-adopter. Drafting is productised in craft software, but wholesale price-logic with scale, project and metal surcharge, and the HOAI fee proposal, are mostly custom projects.
- Limit: LLMs are not calculators. Price, discount and surcharge must come from deterministic code or the ERP.
Shared resources & combinations
- Same stack (llm-general + embedding-text + vector-db): DOC-014 (catalogue and PDF price list parsing), AGENT-027 (RFQ-to-quote for make-to-order), AGENT-005 (supplier quote comparison).
- Upstream: DOC-015 (enquiry and tender analysis), DOC-030 (quantity takeoff from plans), PRED-008 (metal and raw material price outlook).
- Downstream: DOC-001 (supplier invoices check against quoted purchase prices), LANG-003 (cover text and offer wording), OPT-004 (price-strategy guardrails).
- The same catalogue index serves quote drafting and article search for the sales desk.
Effort, cost & KPIs
- Path A: weeks; subscription of the craft software, extra for AI module. Path B/C: 2-5 months including ERP integration and data clean-up.
- KPIs: time per quote, share of lines accepted without change, price error rate found after release, quote-to-order rate, margin of released quotes vs. floor, share of quotes with outdated price date.
People & change
Involve early: management (margin floor, who may release), the head of sales or the estimator as process owner, IT for ERP access, data owners of the article and customer conditions, and the DPO if customer data goes to a cloud model.
Works council: no in the usual case; the tool prices articles, not employee behaviour. If logs of individual clerks are evaluated, §87(1) no. 6 BetrVG may apply; §90(1) no. 3 asks for timely information anyway (GUIDE-004).
Data protection: quotes contain business contact data and, for sole traders or private customers, personal data; use an AVV (Art. 28 GDPR) and send only needed fields (GUIDE-005).
AI literacy: short training for estimators on typical errors (units, substitute articles, old prices) satisfies part of Art. 4 AI Act (GUIDE-006).
Pilot: start with one product group or one trade, run AI drafts in parallel with the manual quote for a few weeks, and measure line acceptance. Concern to address: experienced estimators fear loss of know-how value; position the tool as a draft that captures their rules (GUIDE-008, GUIDE-009).
Risks, regulation & compliance
- EU AI Act: minimal risk; no Annex III category. AI literacy duty (Art. 4) applies.
- Price errors: a wrong released price binds the firm. Four-eyes release, margin floor and price-date display are mandatory controls.
- Pricing law: B2B quotes are free of the consumer price rules, but competition law applies to price coordination with competitors; do not feed competitor price agreements into the tool. Quotes to consumers need gross prices (PAngV).
- HOAI: since 2021 the fee tables are no longer binding minimum and maximum rates; the former minimum is a non-binding "base fee rate" that applies if no fee agreement was made (agreements need only text form, e.g. email, and apply to contracts concluded from 1 Jan 2021). The tool must not present the table value as legally required [2].
- Metal surcharges: the DEL copper quotation was suspended on 14 Feb 2022 and there is no single successor; wholesalers use their own or derived quotations (e.g. Woehner switched to the LME quotation times a correction factor from 1 Mar 2022). The index and procurement-cost percentage must be taken from the supplier's terms [3][4].
- Data protection: customer conditions are trade secrets; prefer EU hosting and contractual no-training clauses.
- Mitigations: deterministic price service, source shown per line, audit log of suggestion vs. final quote.
Getting started
- Measure the baseline: quotes per month, minutes per quote, error and rework rate.
- Check whether the current software already offers AI drafting and automatic Datanorm updates.
- Write down discount, scale, project and surcharge rules for the top 20 customers or articles.
- Test on 30 real past enquiries; compare the draft with the quote that was actually sent.
- Go live with human release only; review line acceptance monthly and fix rules, not prompts.
Sources
- Wikipedia (DE) - Datanorm (file format for article and price catalogues in trade; prices only as current as the imported file) - https://de.wikipedia.org/wiki/Datanorm
- Goerg - Die neue HOAI (2021): Preisorientierung statt zwingendem Preisrecht - https://www.goerg.de/de/aktuelles/veroeffentlichungen/02-03-2021/die-neue-hoai-2021-trat-am-1-januar-2021-in-kraft-preisorientierung-statt-zwingendem-preisrecht
- Handwerksblatt - Ohne Kupferpreis-Notierung ist keine Kalkulation moeglich - https://www.handwerksblatt.de/handwerkspolitik/kupferpreis-notierung
- Woehner - Change of the copper price quotation DEL (DEL suspended 14 Feb 2022; wholesaler switches to LME with correction factor) - https://www.woehner.de/es/novedades/novedades-es/change-of-the-copper-price-quotation-del
- Skill Sprinters - KI im Handwerk 2026: Fuenf Use Cases vom Angebot bis zur Wartung (vendor-near blog; claims not independently verified) - https://skill-sprinters.de/blog/branchen/ki-im-handwerk-2026-fuenf-use-cases/
- Handwerk-digitalisieren.de - SHK-Software 2026: Branchensoftware fuer Heizung & Sanitaer - https://www.handwerk-digitalisieren.de/handwerkersoftware-shk/