The full estimate came to 3,393 hours. The client had expected a number several times smaller.
At that moment it is easy to trim the estimate down to the budget, win the contract, and move the hard conversation to the middle of the build. We left the number where it was. The work stopped for 270 days, and the relationship survived: the client came back on their own and continued with the same engineers.
Snapshot
| End-client sector | manufacture and installation of glass structures |
| End client | under the code-name Glass Configurator, Russia |
| Engagement | direct contract; discovery paid for as a separate stage |
| Task | pre-project discovery and an estimate for an industry-specific product |
| Full estimate | 3,393 hours, 8 to 18 months |
| Reduced estimate, August 2023 | 1,196 hours, 3 to 6 months |
| Outcome | the client got a specification, a prototype, a working drawing generator and a defensible estimate; after the pause they returned to the same team |
The problem
A manufacturer of glass structures came to us. The industry tool their process rested on had left the Russian market and ran only from the cloud. Installing it on their own server was not an option. They needed a tool of their own.
The client handed over a real order as a PDF to work from and opened access to their production staff. The request was short: build the same program. It carried no boundaries, no priorities and no budget.
Why this is hard
A task like this cannot be estimated by analogy. Glass sizing with tolerances, cut-outs for hardware and a file a cutting machine will accept look nothing like ordinary web development. The error here is measured in multiples rather than percentages, and it gets paid for either by the contractor in unbilled time or by the client in an unfinished product.
For the buyer the picture is simpler and worse. They do not know what their own idea costs, and there is no reference price anywhere around them. Until the contractor shows what the work is made of, any figure looks like an opening bid rather than an engineering result.
How we did it
1. Estimated the whole scope, including the unpleasant part. The full version came to 3,393 hours and 8 to 18 months of work. At the review it became clear that the client had been counting on a number several times lower.
2. Refused to trim the figure. We named the risk plainly: a lowballed estimate does not reduce the scope. It moves the gap to the middle of the project, where the contractor is already working at a loss, the schedule stretches, and the client is left holding an unfinished product.
3. Sold discovery as a separate paid stage. In 78 days, from 2 May to 19 July 2023, we produced a specification broken down by module, an interface prototype, a working drawing generator and 3 revisions of the estimate. Closing that stage separately was the client’s own idea. The invoice was paid the next day.
4. Showed the structure of the estimate, not the total. Almost 79% of the hours sat in the product configurator. Inside it were 3 independent product types, each with its own geometry, hardware and tolerances. From there, cutting became a deliberate act: drop a feature and see immediately how many hours it releases.
5. Recalculated the reduced scope the same day. 22 days after the review the client asked to shorten the first stage. By that evening the estimate had come down from 3,393 to 1,196 hours, not through a discount but through features moved to later.
Incidents and response
A scale defect on the very first drawing. We sent a generated file to production and got an error back: the height came out at 1000 mm instead of the size required. The fix took 4 minutes, after which production confirmed the file was correct. That is what early contact with a real shop floor buys: the mistake surfaced where fixing it took minutes instead of months of rework.
We had to study the reference program from a screen recording. Access to the industry tool had already been switched off. Part of the domain knowledge was reconstructed from a recording of the client’s own employee working in it, and we wrote that uncertainty into the estimate rather than hiding it.
Results
| Metric | Value |
|---|---|
| Full estimate | 3,393 hours, 8 to 18 months |
| Reduced estimate, August 2023 | 1,196 hours, 3 to 6 months |
| Discovery stage | 78 days; paid separately, invoice settled the following day |
| Working materials | specification by module, prototype, drawing generator, 3 estimate revisions |
| After the estimate | 270 days of pause, then the client returned to the same team on their own |
An honest estimate cost 270 days of waiting. It did not cost the relationship. When the client came back, the restart did not begin with a blank document: we pulled up the specification, the prototype and the generator code, then fitted them to the new scope.
Team
- Analyst-developer from the bureau, domain knowledge and estimation. On the project throughout, from May 2023 to June 2026
- Front-end developer from the bureau, screen prototypes
- Anton Hersun, Xaver Pro, project manager
If an estimate has come back to you that does not fit the budget, send us the specification and the range you need to land in. We will tell you what can be cut without losing the product, what cannot, and return a breakdown in hours. The first assessment is free.