FR EN
B2B distribution · Office supplies & furniture

Three time-consuming chores, turned into AI assistants.

At this company, a distributor of office supplies and furniture, repetitive high-volume tasks were eating hours every week and leaving revenue on the table. We did not “digitalise” them — we compressed them. Here is what it produced, measured.

0%
accuracy on product coding
vs the in-house expert — on a task that had been 100% manual
0
of at-risk revenue detected automatically
across 545 customers analysed, 296 follow-ups pre-drafted
0
tender lines matched
2 lots analysed, go/no-go verdict + environmental score
The detail, chore by chore
The starting point

A company that works — and is drowning in the invisible.

The company was not short on expertise. It was short on time. Like any distribution SME, its team repeated tasks every week that were individually low-value but very high in volume: tracking down product references, answering public tenders, following up with the right customers. Necessary work, slow, and never finished.

🏢 The client

  • B2B distributor of office supplies and furniture, western France.
  • A catalogue of 20,330 references in active use (the product “Bible”).
  • A sales and customer-service team, well equipped but with no in-house AI expertise.
  • A clear intent from management: save time without stripping the human out of customer relationships.

What it was costing

  • Translating every customer request by hand into catalogue item codes.
  • Pricing public tenders line by line — hundreds at a time.
  • Spotting in raw exports which customers were drifting away, before they left.
  • Knowledge concentrated in a few key people, hard to scale.
The method

You don’t “add AI”. You kill one specific chore.

No sweeping transformation programme, no six-figure platform. One simple loop, repeated on every task that bleeds time.

01

Spot

Identify the repetitive, high-volume task that eats hours for little cognitive value.

02

Compress

Build a bespoke assistant, plugged into the company’s real data, that does the bulk of the work.

03

Measure

Compare the AI’s output against the human expert’s. A number, not a promise.

Principle
AI proposes, the human decides. Nothing goes out and no pricing decision is made without a review. The machine does the chore — the person stays in control.
The results

Three assistants, three chores that no longer cost a day.

Assistant 01 — Product coding

From a customer request to the right product code, in seconds.

A salesperson receives “quote for 5 ergonomic chairs + 2 pedestals for the town hall of X”. Previously this meant digging through the catalogue by hand. The assistant finds the catalogue codes on its own, from plain language, and flags whatever it is not sure about. Measured against quotes validated by the in-house expert: 83.9% accuracy, without missing a single line that needed manual handling.

Catalogue of 20,330 items Semantic search + AI validation
First version (word-for-word)9,7 %
Deployed version83,9 %
Accuracy measured on a reference corpus of 31 lines taken from real customer requests; ground truth = quotes validated by the in-house expert.
545 customers
the universe analysed
185 priority accounts
genuine drop-off detected
296 emails ready
follow-ups pre-drafted
0
of revenue
at risk, made visible
Assistant 02 — Customer retention

The revenue that walks out without warning, made visible.

From a simple sales export, the assistant spots the customers who are dropping off — not those who simply bought once, but the real signals. It surfaced €186,000 of at-risk revenue across 545 customers, ranked the 185 priority accounts, and prepared 296 follow-up drafts personalised and ready to review. All inside an interactive dashboard.

Interactive dashboard 296 follow-ups in under a second
Assistant 03 — Public tenders

A public tender taken apart while you drink your coffee.

A tender means hundreds of lines to price and a go/no-go call to make. The assistant matches every line against the catalogue, calculates the coverage rate, researches the contracting authority online, assesses the environmental score and returns a clear verdict per lot. On a supplies tender for a local authority: 671 lines processed across 2 lots, in a single pass.

🟡 Lot 01 — worth a look (73.8% coverage) 🟢 Lot 02 — go for it (91.3% coverage)
74%coverage
Lot 01 · Supplies
478 / 648 lines
91%coverage
Lot 02 · Stationery
21 / 23 lines
Every line of the tender is matched against the catalogue, with online research on the contracting authority and an estimate of environmental compliance (French AGEC law).
Why this transfers

The same loop works in any small business.

This company is not unusual. Any business that distributes, quotes or handles tenders carries the same invisible chores. What changes from one to the next is the catalogue and the tools — not the method.

🎯

Bespoke, not off-the-shelf

Plugged into your real data, your vocabulary, your customers. Not a generic piece of software.

📏

Proven by a number

We measure the output against your expert. If it does not hold up, you see it straight away.

🤝

People stay in control

AI prepares, your teams approve. No blind automation, no loss of control.

Quick to test

A first useful assistant within weeks, on one well-chosen chore. Not a year-long project.

What about you?

Which chore is costing you a day a week?

We pick one. We build the assistant. We measure what it saves you — and you decide what comes next.

Alexis Pineau
Designing & deploying AI assistants for small businesses