A large infrastructure contractor kept winning big public tenders, and watching its gross margin shrink anyway. The pricing was done by hand, on last year's numbers. We grounded every bid in what the work actually costs, and put a hard floor under the profit.
Management called us in after a quiet, uncomfortable discovery: the company was winning plenty of large tenders, and its gross profitability was eroding all the same.
The cause was in the pricing room. Bills of quantities (the thousands of priced line items behind every tender) were worked out in hand-built spreadsheets, leaning on cost estimates from past jobs that no longer held. The method couldn't price a real-time jump in raw materials (concrete, steel, asphalt) or the labour shortage biting the Israeli market. So bids went out too low, and some projects were losing money from day one.
It was also slow. A team of engineers and estimators spent two to three weeks reading thousands of pages of technical specs, special conditions and drawings just to price a single tender. And the people pricing the next job at head office had no live link to the real costs of the last one, what a subcontractor or an hour of heavy equipment had actually cost in the field.
Real field costs, current material prices, and a hard line the margin can't cross.
We connected the actual execution and procurement figures from across the company's past projects (alongside current supplier price lists and construction-input indices) into a single, living cost record the pricing team can trust.
A language engine reads and parses the raw tender documents and bill of quantities in seconds, pulling out every priced item and the conditions attached to it, work that used to swallow whole weeks.
The model cross-checks each line item against what that work has really cost the company on site, and produces a data-based estimate, not a guess anchored to a job from three years ago.
We set an iron rule in the system: if a line carries a high risk of overrun, or the pricing drags gross margin below an 8% floor, it halts the submission, flags the item in red, and pushes an alert to the bid manager with a recommended budget buffer.
Live material spikes and labour scarcity feed straight into the estimate, so a jump in the price of steel or asphalt shows up in the bid, before it shows up as a loss in the build.
The engine reads, prices and flags; the estimator reviews and commits. The judgement stays with a person, freed from the keying and cross-checking to do the work that actually moves margin.
The hit rate held. The margin stopped leaking.
Loss-making submissions were eliminated entirely, and grounding every bid in real costs lifted gross margin across the company's winning projects by about 3.5%, straight to the bottom line.
Preparing, analysing and pricing a large infrastructure tender fell from three weeks to about 48 hours, and the estimators it freed moved from keying spreadsheets to hard commercial negotiation with subcontractors and suppliers before each bid closed.
The win came from two disciplines working together: grounding every price in the company's real field costs, and a hard margin floor that refuses to let a loss-making bid out the door. Neither is a demo trick. They're the same guardrail-first thinking behind our retail pricing work, where a hard margin floor blocks any loss-making promotion before it reaches the till. The estimator still signs every bid; the system just makes sure the number is honest.
A large US retailer was losing money to spoilage and empty shelves. Per-branch demand forecasting a buyer approves cut waste ~30% and out-of-stocks 60%, and dropped buyer admin from 4 hours a day to 15 minutes.
Read the case → Dynamic pricing · RetailAn Israeli retailer was running promotions that quietly lost money. Real-time margin costing with a hard 12% floor blocks any loss-making deal before it hits the till, lifting gross margin by around 4%.
Read the case →
Reef TRH