Keeping prospect research fresh
Freshness is the record of when each claim about a prospect was last proved, and prospect_freshness uses it to decide what is worth checking again. Not everything ages: a company's founding year does not, its opening hours do. Re-research is routed through the same cost control as first research, so a deeper pass runs only where a stale claim could change a decision.
Things age at different speeds#
| Claim | Ages | Why it matters |
|---|---|---|
| Legal name, founding, registration | Barely | Re-checking costs money and almost never changes an answer |
| Address, opening hours, locations | Slowly | Wrong here makes a call or a visit fail, not a message |
| Named people and their roles | Quickly | The commonest way a personalised opening becomes embarrassing |
| Contact addresses | Quickly | A dead address produces a bounce, and a bounce produces suppression |
| Recent activity — an opening, a hire, a launch | Fastest | This is usually the angle; a stale angle reads worse than no angle |
| Your own supplied fields | Not at all, by Connect | They are attributed to you and are yours to correct |
That table is why freshness is per claim rather than per prospect. A record is rarely wholly fresh or wholly stale, and treating it as one thing means either re-proving facts that cannot have changed or trusting an eighteen-month-old job title.
What a re-check does#
Freshness marks the claims whose age has passed what their class allows.
Result The prospect is flagged as partly stale rather than removed from the list. A stale prospect is still a prospect.
The cost router decides whether re-proving those particular claims could change a decision you are about to make.
Result Where it could not, nothing is spent. This is the whole point of routing research spend rather than scheduling it.
The pass runs and replaces each claim together with its source.
Result A claim and its evidence move as one. You never end up with a new assertion pointing at last year's page.
Everything derived is recalculated.
Result The fit reason, the score, contactability and the next action all reflect the new claims, and the change is visible in the activity trail rather than appearing overnight with no explanation.
The cost control#
Left alone, freshness is an infinite job — every prospect ever researched becomes stale eventually, and a workspace with a long list would pay for a permanent research pass over organisations nobody intends to contact. prospect_cost_router is what stops that. Re-research is prioritised by whether the answer can change something: a prospect you are about to write to, one on the pipeline, one a person just opened.
- A prospect nobody has touched since it was discovered is refreshed last, or not at all.
- A prospect about to receive first outreach is refreshed first, because the message is built from claims that must still be true.
- A converted prospect is not refreshed by this mechanism at all — a live deal is kept current by the people working it.
- Deep research is never triggered by ageing alone. It is a decision, and it is described on its own page.
The spend that does occur lands in the same usage record as any other AI work, so it is visible where the rest of the bill is rather than hidden inside prospecting.
When freshness is not the problem#
- Nothing has been refreshed for weeks
- Usually correct behaviour: nothing on the list was near a decision. Opening a prospect or queuing outreach moves it up.
- A claim you know is wrong is still marked fresh
- Freshness measures age, not truth. Correct the claim; the correction is attributed to you and outranks the stale source.
- Everything went stale at once
- The list was imported or discovered in one run, so it ages in one cohort. Refresh the part you intend to use rather than the whole cohort.
- Re-research produced fewer claims than before
- Sources moved or removed pages. The record now says less and what it says is still proved, which is the trade this design makes on purpose.
Questions#
Can I force a refresh on one prospect?
Yes — opening a prospect and asking for research again is a decision, and a decision always beats the router's ordering. What you cannot do is schedule a standing refresh of an entire list, because that is the cost the routing exists to prevent.
Does stale research block outreach?
It does not block it outright, but the readiness path treats a message built on stale claims as worth a second look. A personalised opening that cites something that stopped being true is the most expensive kind of small mistake.
How long is a claim considered fresh?
It depends on the claim class, and the intervals are configuration rather than a published promise, so this page does not quote a number that a release could change. What is stable is the ordering: activity and people age fastest, registration facts effectively never.