Measure per source, not in aggregate. For each thing you watch, track people collected, people who fit, people contacted and replies, over 30 days. The ratio between the first two tells you whether the source or the filter is wrong, and the last number is the only one that is an outcome.
Disclosure: Signl is ours, and it reports some of these numbers for you. We publish no reply rates, conversion rates or ROI multiples of our own anywhere on this site, because we have not measured any across customers in a way we could reproduce. This article is a method, not a benchmark.
Key facts
- The unit of measurement is the source. Aggregate numbers hide the only decision you make.
- Four numbers are enough: collected, fitted, contacted, replied.
- The collected-to-fitted ratio diagnoses the source. The contacted-to-replied ratio diagnoses the message.
- Published benchmarks for this channel are vendor-authored and mutually contradictory.
- Attribution here is a stamp at collection, not a model.
Why aggregate numbers are useless here
"Social listening produced 14 conversations last quarter" is a sentence you can do nothing with. Add a source, remove a source, change the ICP, rewrite the openers: the aggregate moves and you cannot say why.
The decision you actually make, every month, is which sources to keep. That is a per-source question, so the measurement has to be per-source or it does not inform the only lever you have.
This is also why the first thing to build is not a dashboard. It is a stamp: every person collected carries which source surfaced them, from collection through to the CRM. Without it, nothing else in this article is possible, and retrofitting it later is painful.
The four numbers
Per source, over a rolling 30 days.
| Number | What it is | What a bad value means |
|---|---|---|
| Collected | People the source produced | Very high: a noisy source. Very low: a dead one |
| Fitted | How many passed your ICP check | Low against high collected: wrong source, or wrong ICP |
| Contacted | How many you actually wrote to | Much lower than fitted: a bottleneck in your own process |
| Replied | How many wrote back | The only outcome number in the table |
Two ratios carry the diagnosis.
Collected to fitted tells you about the source. A competitor profile producing 400 people and 4 who fit is the wrong competitor, or your ICP is wrong. The way to tell the difference is to read the rejected pile: if you recognise people you would sell to, the ICP is at fault; if it is job seekers and vendors, the source is.
Contacted to replied tells you about the message, and only once the first ratio is healthy. Diagnosing openers while the source is producing the wrong people is a waste of a month.
The gap between fitted and contacted is the one nobody tracks and everybody has. It is usually a person who meant to write and did not.
The scorecard
One table, updated monthly, one row per source. This is the artefact.
| Source | Type | Collected | Fitted | Contacted | Replied | Cost units | Verdict |
|---|---|---|---|---|---|---|---|
| Competitor founder profile | Competitor | 180 | 22 | 18 | 3 | 1 | Keep |
| Industry creator | Creator | 420 | 11 | 9 | 0 | 1 | Watch, one more month |
| Topic search, "month end close" | Topic | 60 | 19 | 19 | 4 | 2 | Keep, consider a second |
| Our own posts | Own | 95 | 31 | 25 | 7 | 1 | Keep. Best ratio |
| Company page of target account | Account | 12 | 1 | 1 | 0 | 1 | Drop |
The numbers above are illustrative, not ours and not measured. They are shaped to show the four patterns worth recognising: a healthy competitor source, a creator with reach and poor fit, a low-volume topic search with the best fit ratio in the table, your own audience outperforming everything, and a company page earning its place in the bin.
"Cost units" is whatever your tool charges in. In Signl it is sources, where a search counts as 2 against the plan's 25, which makes the topic-search row cost double what its collected number suggests. Put your own unit in that column so a source that costs twice as much has to earn twice as much.
Cost per conversation
One derived number, and it is the one to take to a budget conversation.
Monthly tool cost, plus the loaded hourly cost of the time spent reading and writing, divided by replies.
Both inputs are yours. If the tool is $99 and someone spends three hours a week on it at a loaded $60 an hour, that is about $819 a month. Fourteen replies makes it roughly $58 a conversation. Four replies makes it $205.
Two things that model exposes. Time dominates the tool cost, usually by a factor of five or more, which means the question is rarely "is the software worth it" and nearly always "is this person's time worth it here". And the number only improves by improving the fit ratio, since contacting more people costs more time in direct proportion.
Compare that figure against your other channels on the same definition. Comparing a cost-per-conversation here against a cost-per-lead from paid advertising is comparing different objects, and it is the usual way these arguments go wrong.
Metrics that mislead
Five, in order of how often they appear in reports.
People collected. Improves by adding noise. The purest vanity metric available here.
Engagement volume on your own posts. Measures reach, and reach and pipeline pull apart. A post with nine comments from the right people beats one with ninety from peers; LinkedIn posts that bring buyers into the comments goes into why.
Connection acceptance rate. Measures whether your headline is credible, which is not nothing, and is not pipeline.
Emails found. A count of lookups is a cost, not an outcome. If it is in your results section, it is in the wrong section.
Any vendor benchmark, including a comparison against one. The published figures for this channel ranged from 30% to 45% reply rates across the pages we read on 25 October 2026, against cold baselines quoted anywhere from under 1% to 3%, none with a methodology. Measuring yourself against a number nobody can reproduce tells you nothing.
Closing the loop
Replies have to come back automatically or the scorecard decays into guesswork within two months.
| Destination | How outcomes return |
|---|---|
| HubSpot | Polled daily. Deals and meetings associated with the contact come back |
| Instantly, Smartlead, lemlist | Reply webhooks |
| HeyReach, Expandi | The tool's own reporting path |
| Clay, generic webhook | Nothing returns. You are logging manually |
Signl reports what each source led to where the connected tool sends outcomes back, which is what makes the last two columns of the scorecard real rather than remembered. The field mapping for each destination is in sending engagement leads to HubSpot and Clay.
What you cannot measure, and should admit
Some of the value of this channel does not show up in the scorecard.
A public reply that someone reads and does not respond to. A comment thread where forty people in your market saw you be useful. A prospect who arrives inbound four months later because they recognised the name. These are real and they are not attributable, and the honest response is to say so in the report rather than to invent a model that claims them.
Keep one qualitative line in the monthly review: how many discovery calls this month mentioned something we wrote or said in public. It is not attribution, it is evidence, and it is the thing that stops a working channel being cut because the spreadsheet cannot see it.
Limits
- Small numbers are noisy. At a dozen replies a month, one good week moves every ratio. Judge on quarters.
- The same person appears via several sources. Decide first-touch and hold to it; apportioning is not worth the effort at this volume.
- Reply is a weak outcome. It is the earliest honest one. Meetings and pipeline are better and arrive too late to steer monthly source decisions, so track both at different cadences.
- Nothing here measures whether the message was any good in isolation, only in combination with the source that produced the person.
