Eficients
CAESEN

CS · Real cases · 3 sectors

Real cases.
No inflated figures.

All anonymised at the clients' request: competent descriptor and verified figures, no logos. Here you have the challenge, what we did and what they gained.

CASE 01

Geotechnical engineeringFrom ~5 h to ~15 min per report

+25 years · national reference · ~1 report per day

REAL ARTEFACT
Complete geotechnical report
— display pending client permission

The challenge

Each geotechnical report took hours of repetitive work (borehole data, classifications, memo structure) in a firm that does roughly one a day. Work with consequences, where an error costs dear —and where the engineer's judgement is non-negotiable.

What we did

An AI configured with their technical code and their criteria, connected to their sources (cadastre, geological data, municipal data). Generates the report and flags where decision is needed; quality controls at each step; the engineer validates and signs.

Results

  • · From ~5 h to ~15 min per report (generation + your review)
  • · 70-85% match with the reference report, depending on source quality
  • · Rigour and professional judgement, preserved at 100%

What we learned

Rigour comes from combining programmed determinism with the engineer's judgement. AI doesn't compute 'in your place': it removes the mechanical part so they dedicate their head where it matters.
CASE 02

Architecture11 councils with adapted documentation

+25 years · structures, national prestige · had tried a custom app that fell short

REAL ARTEFACT
Municipal document folder + BC3 output
— display pending client permission

The challenge

From BC3 to estimate by hand, hours in Excel. And each permit's documentation, redone municipality by municipality with different formats. They had tried a custom app built specifically for them; it solved the first pass but fell short as soon as a council changed what it asked for.

What we did

The companion reads their BC3 and produces the estimate with their items; assembles the document folder for each council with the format each one asks for. All by talking to it in natural language — when what's needed changes, no form is rewritten.

Results

  • · From BC3 to estimate in hours, not days
  • · 11 councils with adapted documentation, no redoing each time
  • · Less Excel, fewer data-retyping errors

What we learned

A companion that adapts beats a rigid app. When what's needed changes, you tell it and it does — no rewriting a form or touching an app's code.
CASE 03

Construction~10 HOT and ~12 WARM filtered every week

+45 years · leader with international patent · technical debt at back-office, betting on AI to clear it

REAL ARTEFACT
Voice CRM screenshots + tender radar
— display pending client permission

The challenge

Estimates, tender capture and follow-up, with a team on site —not in front of a form. Much info in the notebook and nothing tidy. A leading firm with patent but with accumulated technical debt at the back-office; they chose to invest in AI precisely to clear it without forcing a cultural change on field staff.

What we did

A CRM that understands voice: you talk and it logs everything (leads, notes, follow-ups). And a radar that filters public tenders and alerts on the ones that fit.

Results

  • · Each week, ~10 HOT and ~12 WARM opportunities already filtered
  • · Relevant tenders detected without watching portals
  • · Site info tidied by talking, not writing

What we learned

Voice fits teams that live on site: friction drops to zero and information stops being lost. The AI doesn't make them change the way they work; it adapts to it.

On anonymityAll clients cited exist and are in production with the figures published. Company policy is not to name them until we have explicit permission. When they authorise, we'll replace the competent descriptor with the real logo and we'll be able to show the artefact (with project data covered). If you want to see them with one-time permission on a call, we can show you.

Next step

Your case can be the next one

Let's talk 30 min →