Pricing a construction tender is the most stressful step of the response: it commits the firm for 6 to 24 months, it makes or breaks the margin, and it's often done in a rush 48 hours before submission. A 3% error on a €2M project means €60,000 lost.
AI in 2026 doesn't replace the estimator's expertise. It does something more useful: it absorbs the mechanical part of pricing (reading the priced bill of quantities, applying ratios, computing coefficients), to free up time on the 20% of items that make the difference. Here's the 2026 method.
Why pricing fails.
Across the 12 construction SMEs I've worked with in 2024-2026, the main causes of pricing failure are the same:
A.Insufficient time. A tender file arrives Monday, submission is Thursday. The estimator has 4 other files going. They patch, paste from a previous tender, miss 2 lines of the bill of quantities.
B.No up-to-date ratio library. 2025 material prices aren't the same as those of the last 2023 project. Many firms price "by feel" without an updated baseline.
C.The bill of quantities misread. A "no object" line, a unit (lm vs m²) confused, a total that doesn't match the quantitative survey: those are the errors that lose 3 to 8% of margin.
D.Poorly calibrated overheads. Many SMEs apply a single coefficient (15%, 20%) regardless of project type. Result: overpricing on simple projects (lost tenders), underpricing on complex projects (lost margin).
E.No safety net. A bid priced too low (eagerness to win) goes through without anyone challenging it. A bid priced too high goes through without anyone re-checking the ratios. There is no review step.
"I won a €2.3M project. I lost €90,000 on it because I forgot to price the demolition of an existing slab. Written on page 47 of the technical specs. After 3 sleepless nights, it goes." — Construction CEO, 2024.
The AI method in 5 steps.
The point isn't to replace the estimator. It's to give them a fully structured pricing draft in 30 minutes, that they validate or adjust in 1-2 hours, instead of starting from a blank page.
STEP 01Extraction of the priced bill of quantitiesThe tender file is analysed by AI, which extracts the 3 financial pieces in spreadsheet format (Excel or JSON) with: item number, description, unit, quantity, and reference to the technical specs. This step, which takes 1-2 hours manually, takes 2 minutes with AI, with 95-99% accuracy.
STEP 02Automatic application of trade ratiosAI applies the firm's known ratios (€/m², €/m³, €/lm, €/U) on each line. The ratio library is stored in an Excel or Airtable owned by the firm. Standard items priced in 5 minutes instead of 1 hour.
STEP 03Detection of uncertain itemsOn items with no ratio in the database, or with a wide range (±30%), the AI marks the line in red with a comment: "Specific material: supplier consultation recommended". These items become the estimator's priority.
STEP 04Overheads & margin calculationThe AI applies the coefficients configured by the firm depending on project type (rural/urban, new build/renovation, public/private). Site overheads, general overheads, net margin, contingency provision, are computed automatically with the option to adjust line by line.
STEP 05Line-by-line human reviewThe estimator opens the pre-filled bill of quantities in Excel, sees lines in red (to price) and green (AI pricing ready), validates each line. For specific items, they make their classic supplier call. Review time: 1-2 hours instead of 4-6 hours.
Result: final pricing in 2-3 hours instead of 4-8 hours, with at least equal quality (sometimes better: lines aren't forgotten, the right ratios are systematically applied). Margin protected, tenders better targeted, estimator less burnt out.
Trade ratios to know (2026).
Here are the average ratios observed in metropolitan France for the main construction items. Important: these ratios are orders of magnitude. Each firm must calibrate its own based on its area, suppliers, client profile.
Item
Unit
Ratio 2026 (excl. tax)
Standard reinforced concrete structural work
m² gross floor
€340-480
Interior demolition (partition + floor)
m²
€45-95
Single-skin BA13 plasterboard partition
m²
€52-78
Suspended ceiling 60×60 tile
m²
€38-62
Porcelain stoneware tiling 60×60 laid
m²
€62-95
Matte acrylic paint (wall+ceiling)
m²
€14-22
Interior wood joinery (standard door)
U
€380-650
Tertiary high-current electrics
m²
€85-145
Office sanitary plumbing
m²
€45-80
HVAC: VRV multi-split
kW
€1,200-1,850
External insulation, 14 cm polystyrene
m²
€110-160
Flat roof, two-layer waterproofing
m²
€95-145
Asbestos removal SS3 leaded partition
m²
€180-340
Source: aggregation of 12 construction SMEs 2024-2026, margins and contingencies excluded. Ranges reflect variability by region (Paris region +15-25%), accessibility (floor without elevator +20%), and market type (private +10% vs public).
Tools 2026.
a. For reading the priced bill of quantities
DCE Analyzer (free up to 3 analyses/week) extracts lots, the priced bill of quantities in JSON, and flags alerts. For intensive use: Claude API + a custom n8n workflow (8-15 days of setup).
b. For the ratio library
Airtable (€12/u/month) or a simple shared Google Sheets is enough. One line per item, low/mid/high ratio, last update. For very small firms, a local Excel with Dropbox versioning is enough.
To go further, paid pricing libraries (€95-150/month) provide pre-calibrated ratios. Relevant for highly diversified firms.
c. For final pricing
Excel remains the estimator's reference tool. AI produces a pre-filled Excel file the estimator opens, validates, adjusts. No workflow change, just a different starting point. For larger contractors, dedicated estimating software can be integrated via API.
d. For quality review
Good practice in 2026: have the AI re-read the pricing with a prompt "identify items potentially under or over-priced relative to the trade ratios". This catches 1 in 3 errors a human would have let through.
Pitfalls to avoid.
A.Blindly trusting the AI. AI can "hallucinate" an item that doesn't exist, or mistake the unit. Always check line by line at least on the first 10 items of the bill.
B.Using competitor ratios. Tempting: open an online construction price site, copy ratios. False good idea: those ratios are public, used by everyone, so "market average". You lose your competitive edge.
C.Not updating the database. Material ratios varied +30% between 2021 and 2025. A ratio database not updated every 6 months becomes poison for pricing.
D.Underestimating variable costs. Specific site costs (difficult access, security, site facilities) are missed by AI if the tender file doesn't explicitly mention them. Always verify on a site visit.
E.Forgetting the margin on subcontractors. If you subcontract a lot, add 8-12% margin to the subcontractor quote. AI is unaware of this commercial logic, it's up to the human to add it.
Going further.
Complementary articles on construction tender response:
No, and you shouldn't want it to. AI in 2026 is excellent at pre-pricing: extracting quantities, applying known trade ratios, flagging uncertain items. Final pricing always requires a human eye for site specifics (access, basement, phasing constraints) and supplier negotiations.
What accuracy to expect from AI pricing?
On standard items (structural work, standard finishing trades), AI accuracy is around ±5-10% with good trade ratios. On specific items (special mechanical, custom fit-out), supplier consultation is needed. AI saves 50-70% of the time and frees human expertise for the critical 20%.
How to feed the firm's trade ratios to the AI?
Either via an Excel file (the simplest) with one line per item, the ratio in €/m² or €/U, and the variation range. Or via an Airtable / Supabase database queried by the AI agent. Or via an existing library according to your subscription. The first option is enough for 90% of construction SMEs.
Self-host the AI or use an online service?
For 95% of construction SMEs, an online service (DCE Analyzer, Claude API, OpenAI API) is enough: GDPR-compliant, European hosting available, data not reused for training. Self-hosting (open-source models like Llama 4 or Mistral) only makes sense beyond 100 pricings/month or for defence / nuclear markets.
Can AI detect an error in the priced bill of quantities?
Yes, that's actually one of its strengths. AI can flag: an inconsistent unit (lm vs m²), an aberrant quantity (paint on 10,000 m² for an 800 m² project), a line with no unit price, a heading missing relative to the technical specs. These checks, rarely done by humans, take 30 seconds in AI.
How much does AI pricing cost in API?
For a priced bill of quantities of 100-200 lines analysed via Claude Sonnet 4.5 or GPT-4o, the API cost is €0.05 to €0.20. Even at 200 pricings/year, total cost is below €50. ROI isn't even a debate.
Test AI extraction of a priced bill of quantities in 2 min?