Customer proof

Proof, from the teams who deployed it.

From TotalEnergies to major players in energy, industry and logistics — here’s what K-AI detects, fixes and makes trustworthy across real document estates.

Field report · TotalEnergies Retail Power & Gas

The essential layer of an AI chatbot you can trust.

Station F, Paris — June 2026. RPG customer chatbot (individuals, professionals, businesses, local authorities), ~500 pages of official web documentation. Accelerated by TotalEnergies On.

~500

web pages mapped and audited by K-AI

19%

of pages to fix, identified by K-AI

53%

of cases resolved in 3 weeks, prioritizing the most critical

What K-AI brought

Govern

Map and audit the entire existing corpus, with no migration, turning it into a trusted reference base.

Clean

Make reliable

Near-instant detection of inconsistencies: 19% of pages to fix, including cases impossible to spot by eye at scale. Each case is routed to the right expert with a diagnosis already prepared — the update that follows is just as fast.

Activate

A trustworthy customer chatbot: a fragmented base becomes a reference base reliable for AI and for customers alike. Accuracy is measured before / after.

The challenge

Already in production, the RPG customer chatbot answers in natural language from about 500 pages of official web documentation. A living base, enriched over the years, ends up holding divergent versions that are hard to spot manually — yet an assistant serves them back with confidence, at the risk of eroding customer trust.

The result

In three weeks and with limited resources — 1 to 2 experts, half a day to a day per week — the teams focused on the priority cases: 53% of cases were already resolved. A fragmented base turned into a trusted reference base, reliable for AI and for customers alike. At closeout, the chatbot’s accuracy will be measured before / after remediation.

How it works

Analysis plugged into existing sources, no migration.

Automated detection of inconsistencies, with recommendations.

Each case routed to the right expert, with a diagnosis already prepared.

Business experts stay in charge of the decision. K-AI prepares, they decide.

« La démarche a permis d’identifier des conflits que nous aurions difficilement repérés à l’œil. Les recommandations font voir le problème immédiatement et savoir vers quel expert l’orienter. »
“The approach let us identify conflicts we would have struggled to spot by eye. The recommendations make the problem immediately visible and show which expert to route it to.” (translated from French)
Thomas Bensoussan — Head of Digital Products, TotalEnergies Retail Power & Gas

↓ Download the field report (PDF)

Additional proof