Candidate rediscovery means finding people already in your recruitment database who may suit a new vacancy, checking whether their information and interest are current, and bringing relevant candidates back into an active recruitment process.
AI can help search across different wording and extract evidence from CVs. The useful result is a recruiter-reviewed shortlist with clear next actions. A list of similarity scores is only an intermediate output.
This guide describes a proposed implementation approach. The example is synthetic, and no time saving or placement result is presented as a Binarify client outcome. For help assessing your own process, see our AI consultancy for recruitment agencies.
Start with the reason your database is hard to use
A recruiter receives a new job specification and opens a job board. The agency may already have suitable candidates, but finding and verifying them feels slower than sourcing again.
Before buying a matching tool, work out which obstacle applies:
- The CV text is inaccessible. Documents are attached to records, but the search only reads structured fields.
- The wording differs. A candidate describes relevant work using a different title or industry term.
- Records are stale. Skills, contact preferences, location or availability may have changed.
- The history is fragmented. Calls, submissions and feedback live in separate tools or individual inboxes.
- There is no clear next step. A recruiter can find a record but cannot see who owns it or whether someone has already contacted the candidate.
These are different problems. Search quality will not fix missing ownership, and a reminder sequence will not make an unreadable CV searchable.
Check what your existing software already provides
Inspect your current ATS or recruitment CRM before specifying a custom system. Test its document search, filters, duplicate handling, saved searches and candidate activity history using a real workflow that you are authorised to review.
Established products already offer AI assistance. LinkedIn’s Hiring Assistant documentation describes sourcing and applicant-review features, including ATS-connected functionality whose availability depends on supported integrations and access. That does not establish what is available in your agency’s subscription or ATS.
There are three possible recommendations: configure an existing capability, connect the missing systems, or build something for an unmet need. The case for custom development should explain why the first two are insufficient.
A practical rediscovery workflow
1. Agree the vacancy criteria with a recruiter
Turn the specification into an editable set of requirements. Separate essential qualifications from preferences and decide which claims need verification.
“Experience planning manufacturing production” is different from “has held the exact title Production Planner.” A client may require a specific qualification, but a preferred software package should not silently become a mandatory condition.
Keep the source specification and record changes to the criteria. Otherwise a recruiter cannot tell whether the search has improved or the target has moved.
2. Establish which records can be used
Identify the records your agency is permitted to process for this purpose. Review retention, contact preferences, access restrictions and deletion requests with the person responsible for candidate data.
Then check the practical foundations: duplicate identities, document formats, extraction quality, latest version and the link between a CV and its candidate record. A scanned page, a two-column CV and a missing attachment may each need different handling.
Keep extracted statements connected to their source. If an employment date cannot be read confidently, flag it instead of guessing.
3. Retrieve evidence, not just similar words
Combine reliable structured information with document search. Use explicit requirements where appropriate, and semantic retrieval to find related wording. Show the passage or record supporting each proposed match.
Missing evidence should remain visible as an unknown. The absence of a certification from an old CV does not establish that the candidate lacks it; it establishes that the record does not answer the question.
4. Have a recruiter review the suggestions
Give the reviewer the role criteria, supporting passages, unresolved questions and recent contact history. They need enough information and authority to disagree with the system.
A reviewer should be able to distinguish “worth a conversation” from “confirmed qualified.” The first stage need not produce a final hiring recommendation. For a closer look at evaluating candidates against one vacancy, read our AI CV matching and shortlisting guide.
5. Refresh information and confirm interest
Use an appropriate, authorised contact channel to ask whether the person is interested and to verify relevant gaps. Check recent activity first so two consultants do not send conflicting messages.
Record the reply and update the profile with its source and date. Respect opt-outs and contact preferences. Being present in an old database is not enough to assume that a candidate is currently available or wants outreach.
6. Connect the candidate to the vacancy and outcome
Record the reviewed shortlist, contact, reply, submission, interview and placement against the relevant vacancy. Define ownership and the next action.
If these connections are missing, the agency can count suggestions but cannot establish whether rediscovery led to useful conversations or placements. That pipeline foundation may need to come before a more sophisticated search.
A worked example: a different title, relevant experience
Synthetic example—not a real candidate or client result.
A vacancy requires manufacturing production-planning experience. Experience with a particular planning package is preferred.
An older CV uses the title “Operations Coordinator.” It describes scheduling production lines and coordinating materials. Its record does not establish current availability or experience with the preferred software.
| Question | What the record supports | Next action |
|---|---|---|
| Relevant planning work? | A CV passage describes production scheduling | Recruiter reads the passage and checks the scope of responsibility |
| Exact job title? | Different title | Assess duties against the requirements rather than title alone |
| Preferred software? | Not stated | Ask; do not invent proficiency or assume its absence |
| Current availability? | Unknown | Confirm interest and availability before progressing |
The useful output is an explanation of why this record merits review and what still needs checking. A headline “92% match” would conceal those distinctions.
Evaluate the search before scaling it
Start with a representative sample of vacancies and candidate records that your team is authorised to use. Agree the evaluation criteria with experienced recruiters before comparing approaches.
Include straightforward matches, related job titles, incomplete documents and cases where a similar-sounding candidate does not meet an essential requirement. Historical placements can supply examples, but they are not a complete list of everyone who could have been suitable.
Compare the new process against the current one:
- How many suggested candidates are relevant after review?
- Which independently identified suitable candidates were missed?
- How much time does review and correction take?
- Can each factual explanation be traced to a source?
- Are certain document formats or groups being systematically overlooked?
Reserve a separate set of examples for evaluation after changes. Repeatedly adjusting the system against the same few CVs can make the demonstration look better without showing whether it generalises.
Do not adopt a universal accuracy target from a sales page. Set acceptance criteria for your role types, consequences of errors and human review process. Begin with a limited rollout and review actual exceptions before expanding it.
Measure the funnel, not the size of the shortlist
Useful measures cover effort, quality and downstream outcomes.
| Measure | Definition to agree |
|---|---|
| Time to reviewed shortlist | Recruiter time from agreed vacancy criteria to an approved initial shortlist |
| Review relevance | Candidates judged relevant divided by candidates reviewed |
| Qualified candidates missed | Suitable candidates identified independently but absent from the retrieved set |
| Candidate response | Replies divided by successfully contacted candidates, with channel and period recorded |
| Submission-to-interview conversion | Candidates invited to interview divided by submitted candidates |
| Placements from rediscovery | Completed placements traced to existing records, using a consistent attribution rule |
| Cost per outcome | Search, communication, review and operating costs attributed to the defined outcome |
Compare similar role types and record changes in demand, team size and sourcing policy. A stronger hiring market can improve placements even when the workflow has not improved.
Rediscovery is not free sourcing. There are still costs to maintaining records, finding candidates, confirming their details and handling conversations. Its value depends on those costs and on the opportunities the process actually creates.
Build a business case without counting the same benefit twice
For preparation work, potential annual capacity can be estimated as:
Annual task volume × verified minutes saved per task ÷ 60 × loaded hourly cost.
Label this capacity value. It becomes a financial benefit only through a credible mechanism, such as serving more clients, reducing overtime or avoiding a planned hire.
For additional placements, use attributable contribution after variable costs and expected reversals, rather than gross placement revenue. Subtract implementation and continuing costs, including software, hosting, messaging, support and review.
Do not count the same recovered hours once as salary savings and again as extra placement profit without explaining how each benefit is realised. Use conservative and optimistic scenarios with explicit assumptions; neither is a promised result.
Candidate information and human responsibility
The UK ICO’s 2024 audit findings on recruitment AI tools raised concerns about fairness, excessive data collection and transparency. These are relevant questions for a recruitment workflow, not a complete statement of the current rules in every country.
Before implementing, establish the applicable requirements and document which information is processed, why it is needed, where it goes, who can see it and how it can be corrected or removed. Review cross-brand access rather than assuming every consultant should see every record.
Our proposed starting scope keeps retrieval and factual assistance separate from final selection. It does not infer sensitive characteristics or use automatic rejection as the default. Human review needs the supporting evidence, time and authority to change the proposed next step.
A useful first exercise for your agency
Choose a small set of recent vacancies. Ask a recruiter to demonstrate how they would find suitable existing candidates today. Record the steps, time spent, uncertain information and systems they switch between.
Then ask:
- Was the main obstacle search, stale data, missing records or follow-up?
- Could a feature we already pay for resolve it?
- Could we track the suggested candidate through to an interview or placement?
- Which single improvement could we test without rebuilding the entire platform?
That gives you a starting scope grounded in your agency’s work. Binarify’s AI Impact Diagnostic formalises that assessment, establishes the baseline and evaluates whether implementation is justified. Read about our recruitment AI consulting approach and current pricing.