Worked scenario · Staffing and recruitment
The right person is already in your database. Nobody can find them.
A group of four temporary staffing and placement agencies, 35 permanent staff, around 450 temps placed a month. A sector platform, an e-contract service, job-board multiposting, and a talent pool of tens of thousands of candidates nobody can query. Here is a full engagement.
The company
- Size
- 4 branches, 35 permanent staff, around 450 temps placed a month
- Business
- Temporary staffing and placement, in construction, logistics and manufacturing
- Volume
- A pool of tens of thousands of candidates, a large share of whom have already worked for the group
- Tools in place
- A staffing platform (records, contracts, payroll, invoicing), an e-contract service, job-board multiposting, a sales CRM, one tracking spreadsheet per branch
- What is missing
- The talent pool cannot be queried, and the four branches do not count the same way
The starting point
A client calls: they need four forklift operators with a valid licence tomorrow morning, in town. The consultant opens the staffing platform, filters what can be filtered, does not find enough people, and goes off to post an advert. Meanwhile, sixty people who have already worked for the group match the request exactly. They do not come up, because their skill is written in a PDF CV, not in a field anyone can query.
The fill rate is discussed from memory. Nobody can say, client by client, how many orders were filled last month or how long it took. When a client claims the agency is slipping, there is nothing to answer with.
The margin on an assignment is discovered at invoicing. The billed hourly rate is known, the real cost arrives later, and in between sit overtime, meal allowances, indemnities and sometimes a commercial discount nobody has reconciled with the rest.
Finally, certifications and medical check-ups expire without warning. The information is captured somewhere; it surfaces nowhere.
What the audit puts on the table
- 01
A large share of the pool has already worked for the group. Those people are the easiest to place again and the cheapest to source, and they are exactly the ones the current search does not surface.
- 02
Skills live in CVs, not in fields. Job titles are free text: three different wordings describe the same reality and never meet in a search.
- 03
The same temp exists several times over from one branch to the next. Their history is therefore split, and their seniority with the group invisible.
- 04
The four branches do not share a definition of a filled order: some count at placement, others at contract signature. Any consolidation is wrong until that is settled.
- 05
No availability history is kept: you know who is on assignment today, never who was six weeks ago. Without that history, neither fill rate nor redeployment rate can be computed.
- 06
Access to the staffing platform's data is the first thing to check: some vendors in this sector charge for extraction, or refuse it. That is what decides feasibility and budget.
What gets built, week by week
- Weeks 1-2
Connectors and ingestion
Staffing platform, e-contract service, multiposting, sales CRM. Incremental, idempotent ingestion with retry on failure: if the platform is unavailable one night, the sync resumes the next day without creating duplicates. From then on, nobody exports a file to build a report.
- Weeks 3-4
Warehouse, model and deduplication
A historised warehouse, and a model that names your business: candidate, temp, client order, assignment, contract, timesheet, certification, branch. Records duplicated across branches are matched on verifiable attributes, never on a name resemblance. Anything that stays doubtful is flagged as such rather than merged wrongly.
- Weeks 5-6
The skills reference data
The CVs in the pool are read, skills, certifications and experience are extracted, then mapped onto shared reference data: job titles, licences, electrical certifications, driving permits, languages. Anything uncertain goes to human validation rather than being guessed. This is the step that turns a stack of PDFs into a pool you can query.
- Week 7
The definitions, settled
One session with management and the four branch managers: what a filled order is, when it is counted, what goes into the margin on an assignment and what does not. Each rule is written down and implemented once, for all four branches.
- Weeks 8-9
Dashboards and alerts
Fill rate by branch and by client, average time to place, margin by assignment and by client, redeployment rate, billed hours. And alerts: a certification expiring within thirty days, a medical check-up to rebook, an order still unfilled the day before it starts, a client whose fill rate is slipping.
- Weeks 10-13
Search and assistant
“Who is available Monday, holds the forklift licence, lives within thirty kilometres, and has already worked for this client?” The answer comes back in seconds, with the source of each piece of information and the date of the person's last assignment. The assistant does not rank candidates and does not decide for you: it finds and it justifies.
What the technical terms mean, in your business
The words you will hear in any data proposal, translated into your trade. A supplier who cannot make that translation does not know your business.
- Connector
- Your records, contracts and timesheets arrive on their own, every night. Nobody exports a file to build a report any more, and nobody works off a three-week-old extract.
- Historised warehouse
- Last week's state is kept. You can compare this March's fill rate with last March's, which your staffing platform overwrites today by keeping only the current situation.
- Data model
- A temp is one record, even if they have been through three of your branches. A client order is an object you can count, compare and track, not a line in somebody's spreadsheet.
- Skills reference data
- Three different wordings can describe the same licence and the same job. The reference data ties them together, so a search does not miss a profile because of the words used in their CV.
- Semantic search
- You search by what the person can do, not by the exact words written in their CV. That is what surfaces the profile already in your database, the cheapest one to place again.
- Tool call
- The assistant does not invent an availability: it queries your staffing platform and shows, under its answer, the query it ran. You can check every figure instead of trusting it.
- Alert
- A certification expiring within thirty days, an order still unfilled the day before it starts: the information travels to the right person on its own, instead of waiting for somebody to think of looking.
- Written scope
- The list, approved by you, of what the assistant is allowed to do on its own. Outside that list it stops and hands over, along with what it has already checked.
What it changes
Answering an urgent order
A manual search, then an advert
The pool first, in seconds, and an advert only if needed
Skills in the pool
Inside PDF CVs
Extracted, mapped to reference data, queryable
Fill rate
Discussed from memory
Measured by branch and by client, usable in a client meeting
Margin on an assignment
Discovered at invoicing
Estimated at order time, tracked during the assignment
Certifications and medical check-ups
Found expired
Flagged thirty days before they lapse
A temp who worked for three branches
Three separate histories
One record, with their real seniority
These orders of magnitude are what this kind of engagement produces when the conditions below are met. They are not contractual commitments: the pricing for your own situation comes out of the audit.
How long it takes
The breakdown in working days and calendar weeks. The price itself is quoted to you before you commit to each step.
| Audit2 weeks | 5 days | 2 weeks |
|---|---|---|
| Data foundation and skills reference data9 weeks | 26 days | 9 weeks |
| Search and assistant4 weeks | 16 days | 4 weeks |
| Total15 calendar weeks | 47 days | 15 calendar weeks |
The engagement is decided in three steps: you only commit to the foundation after reading the audit, and to the search after seeing the foundation run. Each step is sold at a fixed price, quoted before it starts: if it takes more days than planned, that is on us. The fifteen weeks are calendar time, not full time: the working days are spread across them. It is the longest of the four scenarios, because reading the talent pool is an engagement in itself. On top of that come the monthly infrastructure cost and an operating fee once in production. You own the code and the credentials.
What would make this scenario fail at your company
A locked staffing platform
Several vendors in this sector charge for access to your data, or refuse it outright. It is the very first thing the audit checks: if your data cannot be retrieved at a reasonable cost, you need to know before committing to anything, not after.
A pool that only exists as email attachments
If CVs are not attached to candidate records and sit in individual mailboxes, they have to be gathered first. That is doable, it takes time, and it is priced separately.
Four branches, four definitions
If the branch managers do not agree on what a filled order is, the dashboards will stay correct branch by branch and incomparable across them. That is not a technical question, and nobody but you can settle it.
Wanting the machine to pick the people
We do not build automated scoring or ranking of candidates. Sorting people by a score exposes them to indirect discrimination and exposes you to GDPR, which regulates automated decisions. The assistant finds, matches and justifies; the decision stays with your consultants.
The building blocks involved
The data audit
Two weeks at most, a fixed price, and a document that belongs to you. It is the only honest way to know what your data allows before committing a budget to it.
The data foundation
Four to eight weeks for your data to live in one place, historised and reliable. Delivered in usable stages, not in one go six months out.
The AI assistant
An assistant that answers your business questions in plain language, on figures that come out of your own database. It arrives once the foundation is reliable, never before.
The other scenarios
B2B trade and distribution
Monthly revenue lands twelve days after close, stockouts are discovered the day the customer calls, and four tools give four different definitions of revenue.
Manufacturing and after-sales
Four people in after-sales spend their days answering the same six questions: lead time, order status, part compatibility, warranty, documentation, scheduling a visit.
Hotels and restaurants
Occupancy, RevPAR and average spend live in three tools that don't talk to each other, and booking requests that arrive outside opening hours get lost.
Is your talent pool working for you?
Thirty minutes are enough to know what your staffing platform actually exposes and what your database holds. If it is already usable, we will tell you.