How Coulon built its AI roadmap on a field audit
A few weeks of audit inside an estimating department, to sort what deserves automation, what must stay under human review, and what needs tidying up before anything can even start.
Estimating departments that only hold together by hand
Construction estimating departments generally run on proven but poorly standardised processes from one business unit to the next: manually creating project folders in multiple copies, copying hundreds of quote lines by hand into each architect schedule, dozens of emails a day per site manager to sort and file, site photos scattered across galleries and inboxes. Each task is manageable on its own, the whole is not. Underneath that operational layer sits a more structural problem: eighteen years of documents accumulated project after project, in considerable volumes, scattered and unstructured.
of emails a day per site manager
To be sorted and filed by hand, with no shared rules from one business unit to the next.
of quote lines copied by hand
Every architect schedule was rebuilt line by line from the quote, project after project.
of unstructured documents
Considerable volumes accumulated project after project, scattered and out of reach for any AI.
Coulon, VanvesCoulon
Coulon is a leading company in the Paris region for the renovation and energy performance of building envelopes: facade restoration, insulation and cladding, waterproofing, rehabilitation. Founded in 1920 and based in Vanves, it employs around 120 people.
Its estimating department plays a central role in winning and securing projects. More than a century of activity also means a considerable documentary heritage: it was precisely that untapped resource, long unusable as it stood, that made the company fertile ground.
Three intervention modes, and a document foundation before anything else
The audit did not try to automate as much as possible. It tried to sort: what is repetitive volume, what requires human validation, and what is only feasible once the data has been tidied up.
A field audit over a few weeks
Interviews with the estimating team and site managers, a review of existing quality documentation, and a mapping of the real processes, business unit by business unit.
Prioritising the use cases
Three combined criteria: frequency, impact, implementation complexity. Many ideas set aside, keeping only those that hold up and can be sequenced.
Three intervention modes
Autopilot for high-volume repetitive tasks, Copilot for high-stakes tasks requiring human validation, Hybrid where the process combines both. Every selected case falls into one of the three.
A shared document foundation
Migration of all project folders into a target structure, in phases, with no data ever moved without a safety copy. This foundation gates the majority of the selected cases.
A roadmap you can actually hold, not a shopping list
Use cases prioritised and sequenced, split across three intervention modes depending on the level of human validation required, for an estimated gain of more than a thousand hours a year. The clearest gains are on automatic site email sorting, time entry, copying quotes into architect schedules and instant querying of technical documents: a 200-page specification becomes readable in seconds. Everyone in the estimating department is affected, to varying degrees. The document foundation remains the prerequisite: without it, most of the selected cases do not hold.
Want us to map your requests?
Free audit of your processes, we identify high-ROI use cases together, and quantify the gains before deploying anything.
More case studies
All case studiesA national joinery player
↗Ops
One form, a complete SEO article published as a draft in the CMS
Médident
↗Audit
A 15-day field audit, several costed use cases
Santiane
↗Ops
Part of the support desk automated, without health data leaving the perimeter
Moose Knuckles
↗Audit
Three AI opportunities, quantified on the brand's own data