A staffing partner sent us an export last year: 186,000 candidate records going back to 2018, four ATS migrations deep, with resumes attached as PDFs and free-text notes in three languages. The team described it as an asset. At that moment it was closer to a storage bill.
A candidate database becomes an asset when three things are true. The records are indexed by what people can do rather than by what they once applied to, the contact details are verified, and the consent position is documented per person. Most databases fail all three at once.
Most databases are not dormant, they are unindexed
The usual objection is that old candidates are stale. Some are. The larger problem is retrieval. A recruiter types keywords into a system that stored the resume as a blob and the person as a row attached to one req from 2022. A warehouse supervisor who applied for a driver role three years ago is invisible to every warehouse supervisor search you will ever run.
TalentGraph indexes the person rather than the application. Skills, seniority, location history, languages, shift tolerance, and every prior interaction collapse into one record, so a search returns people who match the requirement instead of people who happened to click the matching posting.
The decay curve is real, and it is uneven
Contact data does go bad, but not at the same rate across channels. These are the verification results from the 186,000 record set, measured at reactivation time and bucketed by months since last activity.
- Email deliverable under 12 months since last activity: 96 percent
- Email at 12 to 24 months: 89 percent
- Email at 24 to 48 months: 71 percent
- Email beyond 48 months: 52 percent
- Mobile numbers still reachable at 36 months: 84 percent
Mobile numbers outlive work email addresses by years, which is why reactivation over SMS and WhatsApp beats email on older cohorts. It also means the verification pass belongs before the outreach pass. Sending 186,000 emails into a 30 percent bounce rate is how a sending domain gets burned in a single afternoon.
Silver medalists are a different population
Inside that set, 2,700 records were silver medalists: people who reached a final interview or an offer and did not start. That is 1.4 percent of the database, and it behaves nothing like the other 98.6 percent. They were assessed by a human, against a real req, at your client, and someone wrote down why they lost.
Reply rates bear that out. The general dormant pool replied to a reactivation sequence at 6.4 percent. Silver medalists replied at 31 percent. Hires per 100 people contacted came in at 0.4 for the general pool and 3.9 for silver medalists, roughly a tenfold difference in yield from identical outreach machinery.
The most qualified candidate for today's req is often the person who came second for last year's.
What reactivation actually costs
For that client, cost per hire through TalentGraph rediscovery landed at $1,180, against $4,900 for net-new sourcing through Beacon Sourcing and paid job boards on comparable roles. Time from req open to a submitted slate was 2 days versus 11. Most of the gap is explained by qualification work that was already done and simply never retrieved.
Rediscovery does not replace sourcing, because it runs out. A 186,000 record database holds a finite number of relevant, reachable, still-interested people, and pulling them forward borrows from next quarter's supply. Treat it as a cheap first pass on every req rather than as a channel with a fill rate you can forecast indefinitely.
Handle consent before anyone drafts a message
- Confirm a lawful basis for contact under GDPR for EU records, and honor storage limitation instead of keeping everything forever
- Maintain a suppression list that survives ATS migrations, since a deletion request from 2023 must still hold in 2026
- Respect CCPA deletion and opt-out requests at the person level, not the record level
- Re-permission any cohort that has been quiet for more than 24 months before a bulk sequence
- Log the source, date, and channel of consent on every record, because you will be asked for it
Two operational notes. Run verification and suppression before anyone writes a message, and have a human approve the first sequence for each cohort. Voice Outreach and Signal Screening will happily contact 200,000 people in a day, which is precisely why the gate belongs upstream of them.
The reframe worth keeping: a dormant database is not a list of people who said no. It is a list of people who said not now, about a different job, to a different manager, sometimes years ago. Index it properly and it is the cheapest supply you already own.
Wen ZhaoStaff Data Scientist, TalentGraph at Novexhire