If you've read The dormant talent pool: the gold mine nobody exploits, the next question is a practical one: where do you start on Monday morning? Here's the method we've drawn from teams that genuinely turned their talent pool into a source of placements, as opposed to those who've been talking about it for three years without moving.
The idea isn't a six-month transformation project. It's a sequence of five steps, each actionable in a few days, where each one visibly pays off the previous.
Before anything else, stop debating in the abstract. Over the last twelve months:
- How many hires came out of the internal talent pool (a candidate already in the base before the need arose)?
- How many came from cold sourcing (LinkedIn, job board, referral)?
- What's the average time to fill when a placement comes from the pool, versus when it comes from external sourcing?
Most ATSs expose this raw data, badly: you often have to export it and cross-reference it by hand. That's fine, you're not after a perfect dashboard, you're after one number to show the team at kickoff.
We don't put forward an order of magnitude, because no reliable public statistic measures this share. Only your own measurement counts, repeated with the same method a few weeks later. For context, Apec reports in its study Executive recruitment practices 2026 (in French) that 54% of French companies consider posting a job offer the channel that most often found the executive they ended up hiring.
That measurement becomes your baseline, and it's the first of the recruitment metrics worth tracking over time. Without it, you'll never know whether the effort you're about to put in pays off.
This is the step that takes thirty minutes in a team meeting and changes everything.
An interview produces two kinds of data. The factual (skills, years of experience, mobility, salary expectations) and the qualitative (motivation, blockers, weak signals, market context).
Decide which 6 to 10 pieces of information each interview should produce as output, and which can be reused elsewhere. A few examples found in most teams:
- Key technical skills assessed, with level.
- Format preferences (fully remote, hybrid, on-site).
- Availability (possible start date).
- Salary expectation stated by the candidate.
- Type of role targeted (lead, hands-on, consulting, etc.).
- An explicit dealbreaker, for example "doesn't want to do management."
What sets this list apart from a plain free-text field is that it's shared. Sophie and Karim use the same grid, and the pool becomes consistent from one recruiter to the next.
This is the technical step. Most teams try to read poorly written interview notes better. That doesn't work, it's the writing you need to change.
Two options depending on your setup.
You keep your ATS, and each recruiter commits to filling in the 6 to 10 structured fields at the end of every interview.
Upside: zero cost. Downside: in the teams we meet, this discipline holds for a few weeks, then it drifts. The cognitive load at the end of an interview is too high for it to last, and the gap widens quickly between recruiters who keep filling in the grid and those who have given up.
The interview is recorded, after informing the candidate, and an AI tool whose every output links back to the interview automatically produces the 6 to 10 output fields, for the recruiter to validate in two clicks before pushing them into the ATS.
Upside: the structuring holds over time, because it's no longer an effort. Downside: you need a tool, and one whose workings you can document against the AI Act requirements, which apply to tools that screen or evaluate candidates from 2 December 2027 (see The AI Act and recruitment).
This is the layer Hirify adds on top of your existing ATS, with no migration and without changing the candidate flow. The Hub listens to the interview (after informing the candidate), structures the output, and pushes the fields into the candidate profile in your ATS. The discipline disappears because there's no longer any discipline to keep up.
Once the interview data is structured at write time, you need to be able to query it in some way other than a static filter.
The test you can put to your current tool:
Find me 5 senior SOC profiles available in January, OK with 2 days on-site, who mentioned an interest in ransomware topics.
No rigid filter can answer that. A natural-language search can, provided the interview data is present, structured, and linked to the candidate profile.
What changes for the team once this building block is in place:
- The starting point for a new need becomes "what do we already have in the base," not "where do we post."
- Managers ask their own questions of the pool instead of going through a recruiter.
- Candidates already met but not placed surface automatically when a role matches. It's often the first effect the team notices.
After 8 to 12 weeks, you redo the Step 1 measurement.
For the roles opened since the start:
- How many hires came out of the pool?
- How many short-lists were built faster?
- How many recalled candidates would have been lost if you'd started over from external sourcing?
These are the numbers to show leadership and the team. If the share of hires sourced from the pool rises clearly above the baseline, the discussion stops there and you keep going.
The effect on recruiters is qualitative too. The satisfaction of calling back someone you already know, instead of cold outreach, can't be quantified, but it retains teams better than a bonus.
Don't try to go back through five years of interview history to structure it. It doesn't pay off and it discourages the team.
Start structuring from now on, the useful base rebuilds itself with each new interview. For the old pool, check retention periods instead, since the CNIL recommends in principle not keeping an unsuccessful candidate more than two years after the last contact.
We say it in the pillar guide, and we'll say it again: a CV is not a talent pool. If you structure only the CVs and not the interviews, you'll stay capped at what a standard ATS already does.
If the team hasn't agreed among themselves on the 6 to 10 criteria they want to capture as interview output, any tool will be misused. The grid must come before the tool, never the other way around.
A tool that does the job but transfers candidate data outside the EU without solid safeguards often ends up blocked by the DPO, with months lost. That question falls under the GDPR and its rules on transfers outside the EU. From 2 December 2027, the AI Act will add its own obligations for tools that screen or evaluate candidates. See the guide The AI Act and recruitment.
With a baseline. Over the past twelve months, count the hires that came from the internal pool (a candidate already in the database before the need) and those that came from cold sourcing, then compare the time to placement for both. An ATS export cross-checked by hand is enough. This number is your benchmark for judging the effort, since there is no public statistic on the subject.
Six to ten reusable pieces of information, agreed as a team and shared by every recruiter, for example the key skills assessed with their level, the preferred work format, availability, salary expectations, the type of assignment sought and any deal-breaker. A short grid that gets filled in beats a long one filled in halfway.
Yes, if the candidate was told their file would be kept in a talent pool. The CNIL, France's data protection authority, recommends that this retention should in principle not exceed two years from the last contact with them. Beyond that, unless there has been new contact or you have documented and justified a different period, the file should be deleted or anonymized, even if the profile still interests you.
No. Reworking years of free-text notes is costly and discourages the team. It's better to structure every new interview from now on. For the old pool, the priority is to check retention periods and purge files whose last contact dates back more than two years.
- Measure the baseline before you move. With no starting number, there's no visible progress.
- Decide the 6 to 10 criteria to capture as interview output. Fewer fields, better filled, beats many empty ones.
- Structure at write time, not at read time. Manual discipline wears off, AI assistance takes over.
- Query in natural language. A talent pool becomes useful when you ask it real questions, not when you apply filters to it.
- Re-measure at twelve weeks to show the team the ROI. That's what anchors the practice.