Skip to content
From listening to matching — AI case study for HR services
Case study — Conversational AI for HR services

From listening to matching

How a trade association used conversational AI to gather the HR needs of 73 member companies, redesign its own service offering and build an agent that points each business toward the right solution.

0Companies interviewed
0Services redesigned
0Project phases
Phase 01 — Listening

73 AI interviews to map what companies actually need

The association served 240 member SMEs with a package of HR services built ten years earlier. It knew it had a usage problem. It didn't know what the problem was. Traditional research (an online questionnaire) had been tried twice. Response rate: 11%. Generic answers, unusable.

The decision was to use a conversational AI interviewer built with Tacita. 73 companies completed the interview. Average duration: 22 minutes. The agent followed a structured outline but adapted the flow to the answers. The result is a qualitative dataset with an information density impossible to obtain from a closed-field form.

0
name employment contracts as their first need
0
ask for support on disputes and absence management
0
identify employee benefits as an uncovered area
The data

Operational and strategic needs side by side

The 73 interviews return two distinct blocks. The first concerns the day-to-day management of the employment relationship. The second concerns HR policies companies would like to have but don't know how to build.

Three findings that changed decisions

01 — The labor consultant as an executor

74% of the companies have an active labor consultant. 68% of those describe them as a payroll administrator from whom they receive no strategic guidance. Companies pay for compliance. They would like to pay for advice.

"the payslips always arrive on time, but when we asked for an opinion on restructuring our shifts we were told it wasn't their remit"— Manufacturing company, 110 employees
02 — Union relations block decisions

40% of the companies have avoided organizational changes for fear of the union reaction. The problem isn't conflict. The problem is the lack of internal skills to handle the negotiation.

"we needed to introduce shift work but we didn't know how to raise it with the works council, in the end we didn't do it"— Food company, 42 employees
03 — The benefits plan exists but nobody uses it

47% name employee benefits as an uncovered area. Of those, 64% already have an active benefits plan. The problem is the distance between the plan on paper and actual take-up. None of the companies interviewed tracks the real participation rate.

What the member companies say

Filter the statements gathered by the AI interviewer by topic

Phase 02 — The redesign

From data to a new service offering

The findings from Phase 1 made it clear that the association's problem was not the quality of its services. It was the architecture. Companies were asking for guidance up front, segmentation by size and a single point of entry. The offering was rebuilt around those three requests.

The decisions

5 services redesigned on the basis of the interviews

Every decision is tied to a finding that emerged in Phase 1. None of the five changes had been planned before the research.

New service

Annual employment law check-up

A preventive audit of contracts, disciplinary procedures and litigation risk. Included in the membership fee for all companies with more than 15 employees.

Answers 74% of the sample
New service

Union relations desk

A package of pre-paid hours with a consultant specialized in collective bargaining. Activated on request, with no quotation needed.

Answers 40% of the sample
Redesigned service

Benefits with built-in measurement

A benefits platform with a dashboard that tracks actual take-up. Quarterly report to the owner. Providers pre-selected from the network.

Answers 47% of the sample
Redesigned service

Personalized regulatory alerts

A monthly notification filtered by the collective agreement applied and by company size. It explains what has changed, what to do and by when. It replaces the generic circular.

Answers 37% of the sample
New model

Segmentation by size band

Three service levels: micro (<15 emp.), SME (16-100), larger (100+). Each level has a dedicated contact, its own price list and a calibrated starter package.

Answers the 62% who didn't know the offering
The design criterion

Every service was built from a minimum threshold: if a need was mentioned by at least 35% of the sample, it became a candidate for a dedicated service. Below that threshold, the need was covered with information resources (guides, templates, webinars) rather than structured services. This criterion avoided creating an offering too broad to communicate.

Phase 03 — The matching

An AI agent that points each company to the right service

The redesigned offering solved the content problem. The communication problem remained. 62% of the members didn't know which services were available before the research. Sending a PDF with the new offering would have reproduced the same failure. The decision was to build a second AI agent with Tacita: this time not to listen, but to guide.

How it works

From passive listening to active matching

The matching agent was built starting from the knowledge produced in Phase 1. It knows the typical needs by sector and size. It knows the redesigned offering from Phase 2. Its job is to bridge the two.

01

The company opens the agent

Through a link in the membership renewal email or in the members' area of the website. An anonymous conversation, with no obligation.

02

The agent asks 4-6 targeted questions

Sector, size, collective agreement applied, most urgent pain point. In 5 minutes it has enough of a profile to give guidance. It adapts the flow to the answers.

03

It proposes 1-2 services with a rationale

It explains why that service answers that specific need. It cites anonymized data from other members as a benchmark. It offers a direct contact with the right person.

What the agent knows

The agent's knowledge contains three layers of information. The first is the map of needs that emerged from the 73 interviews, aggregated by sector and size. The second is the catalogue of the 5 redesigned services, with applicability criteria and explicit limits. The third is a set of frequent objections gathered during Phase 1 (cost, time, complexity) with the concrete answers the association has prepared.

What the agent doesn't do

Explicit limits

The agent doesn't sell, doesn't negotiate, doesn't discuss specific prices and doesn't give legal or tax advice. If a question falls outside its scope, it closes the conversation with a human contact. This constraint was designed before the prompt was written, not added afterwards. The agent's credibility depends on the clarity of its boundaries.

An example conversation

WhoMessage
AgentWhat sector does your company operate in and how many employees do you have?
CompanyWe're a food company, 48 employees.
AgentWhich aspect of managing your people is causing you the most difficulty right now?
CompanySickness absence is paralyzing our production and we never know when we can act.
AgentAmong the food companies interviewed in your size band, disputes over absence is the second most cited need. The association has launched an annual employment law check-up that includes a review of disciplinary procedures and the calculation of the protected sickness period for each employee. It's included in the fee for companies with more than 15 employees. Shall I put you in touch with the right contact?
Results

What the complete project produced

Figures measured 90 days after the launch of the matching agent.

0
Interviews completed in Phase 1
0
Increase in the service usage rate
0
Of the members interacted with the matching agent
0
Activated at least one new service after matching

The most significant figure isn't the increase in usage. It's the quality of the match. Under the old offering, 78% of activations concerned the payroll service. After matching, activations are spread across all the services, with the employment law check-up and the union relations desk together accounting for 44% of new requests.