Challenge
Listening to every customer, without tying up a team for six months.
The manufacturer's customers talk about its products continuously. Their reviews say what satisfies, what disappoints and what breaks down. It is the most direct form of the voice of the customer, and it remained largely untapped: at this volume, no team could read every verbatim.
Generative AI opened the way to an exhaustive reading. Management had to answer three questions before investing.
- Value. Check that the analysis surfaces actionable insights for product and quality teams.
- Timeline. With a traditional organization, a first answer required six months.
- Reliability. A classification that business teams do not trust will never be used.
The need: a measured, fast and reliable answer on the value of industrializing voice-of-the-customer analysis.
Solution
A team of experts augmented by AI agents, with a measurement requirement at every step.
We deployed a delivery model in which two experts steer AI agents. Three principles structure it.
- Expertise on structuring decisions. Scoping, the verbatim taxonomy, evaluation criteria and the decision to industrialize stay in the hands of the experts.
- AI agents on production. The agents produce labeling, evaluation sets, code, tests and specifications, under the experts' control.
- Measurement as the arbiter. We evaluate each deliverable on validated reference sets before any handover to business teams.
Set-up
Scope, structure, evaluate, decide.
- Scoping. Definition of the business questions the voice of the customer must answer, with the product and quality teams.
- Taxonomy. Construction of the verbatim categories and of the drivers of satisfaction and dissatisfaction.
- POC. Classification of verbatims by generative AI, measured on reference evaluation sets.
- MVP. Continuous reporting of satisfaction drivers to the product and quality teams.
- Decision. Go or no-go on industrialization, based on measured results.
A team of two experts, a product lead and a data scientist, supported by AI agents, over 1.5 months.
Results
A POC and an MVP delivered in 1.5 months, compared with six months for a traditional organization.
The system classifies customer verbatims automatically according to a taxonomy built with the business teams. Satisfaction and dissatisfaction drivers reach the product and quality teams, who now have a continuous reading of the voice of the customer.
The POC and the MVP were delivered in 1.5 months, where a traditional organization needed six.
Two effects extend the engagement.
- A decision grounded in measurement. Industrialization rests on evaluated results, not on a promise.
- A reproducible delivery model. The pair of experts supported by AI agents applies to other verbatim sources.
Expertises mobilisées
Savoir-faire
- Voice of the customer
- Verbatim analysis
- Generative AI
- AI agents
- Evaluation sets
Technologies
Signature
Four mechanisms that make the difference.
- AI agents at the heart of the team. They handle production, which divides the timeline by four.
- Human expertise in control. The experts own the taxonomy, quality criteria and decisions.
- Evaluation before use. We measure each result on validated reference sets.
- A taxonomy built with the business. Product and quality teams find their own categories in the analysis.
For you
The voice of the customer, leveraged in full.
This approach addresses four situations.
- Your contact center, online reviews or complaints produce volumes of verbatims that nobody reads.
- You want to cover all customer interactions, not a sample.
- You want to evaluate a generative AI use case without committing to a six-month project.
- You need to decide on industrialization based on measured results.
We assemble a lean team of experts supported by AI agents. We build the taxonomy with your business teams and deliver a measured answer within a few weeks.
