What prospecting costs
Prospecting spends on model calls: reading sources, extracting claims, scoring against your brief, and writing an opening. prospect_cost_router decides how much of that a given prospect is worth, and its test is not importance but consequence — would a deeper pass change what happens next. The spend appears in the same usage record as every other AI call.
What costs and what does not#
- Reading and extracting
- The bulk of it. Public sources are fetched and read, and claims are extracted with the source attached to each one.
- Scoring
- Small per prospect, large in aggregate, because it runs on everything discovery returns rather than on what survives.
- A deeper pass
- The expensive item, which is why it is routed rather than scheduled. It reads more, from more places, about one organisation.
- Writing an opening
- One call per prospect at the point of outreach, not at discovery. A prospect you never write to never pays this.
- Storing and counting
- Nothing. Records, evidence and the headline numbers are database work.
- Re-research
- Whatever the first pass cost, again — which is the reason freshness is routed through the same control.
How the router decides#
The router asks one question about each candidate pass: could the answer change a decision that is actually about to be made? A claim that would move a prospect across the qualification line is worth proving. A fourth corroborating source for a prospect already comfortably qualified is not, however interesting it would be to read.
| Situation | Depth | Reason |
|---|---|---|
| Prospect about to receive first outreach | Deepest available for that prospect | The message is built from these claims and will be read by a person |
| Prospect sitting near the qualification threshold | Deeper | One more proved claim decides in or out |
| Prospect clearly qualified on strong evidence | Stop | More reading cannot change the outcome |
| Prospect clearly outside the brief | Stop early | Spending to confirm a no is the commonest waste in prospecting |
| Prospect nobody has opened since discovery | Minimum | No decision is pending, so nothing is at stake |
| Converted prospect | None | A live deal is worked by people, not refreshed by research |
The consequence is that two prospects in the same list can carry visibly different amounts of research, and that asymmetry is the design working. A list where every row has been researched to the same depth has overspent on the rows that were never going to matter.
Seeing what was spent#
Open Plan & Usage at
#/billing.Result AI usage for the workspace is there, in the same ledger as assistant work, drafting and voice — prospecting is not billed in a private corner.
Use the activity trail for a specific run.
Result Discovery and research appear as recorded work with their outcomes, so a run that produced few prospects can be told apart from a run that produced many cheaply.
Compare against the counts above the list.
Result Cost per qualified prospect is the number worth watching. Cost per row discovered rewards a brief that returns noise.
Spending less without getting less#
- Narrow the brief before widening the run. A precise description spends the same per candidate and wastes far fewer candidates.
- Set a real location scope. Distance is the cheapest filter there is, because it removes candidates before anything reads them.
- Do not ask for a standing refresh of the whole list; refresh the part you are about to use.
- Run deep research on the handful of organisations where the answer decides something, and treat it as a decision rather than a setting.
- Fix a brief that keeps returning your own customers — identity resolution will discard them, but only after research has been paid for.
Questions#
Does a prospect with no address still cost money?
Yes, and it is not wasted. The research proved something about a real organisation, which is why the prospect is kept rather than discarded. What it does not do is go on to pay for an opening that could never be sent.
Can I cap prospecting spend on its own?
The controls are the workspace's AI budget and its plan allowances rather than a prospecting-only dial. In practice the brief and the location scope are the stronger levers, because they act before any model is called.
Why did one prospect cost far more than the rest?
It was near a decision — about to be written to, or sitting on the qualification line — so the router bought more evidence for it. That is the intended shape of the bill, not an anomaly.