Next step agreed, and what the number on the dashboard is actually counting

How to hear an agreed next step on the call is covered in the piece on deals going quiet after the demo, and this one does not repeat it. The question here is what happens after the marker becomes a column.

The evidence for prediction is thinner than the claim

Almost every article on this metric rests on two figures: that a quarter of first meetings never touch the next step, and that close rates drop by about seventy percent when it is skipped. Both come from the same place, a 2018 post on the blog of a conversation-analytics vendor. The post does name a sample, 28,833 closed deals analyzed over web conferencing, but it names it at the top of a different argument. The two figures everyone quotes arrive at the end under the words "in our research", with no period, no industry, no geography and no statement that they come from that same set. The paragraph after them invites the reader to try the product.

A second figure from the same vendor, that the fastest deals spent about half again as long discussing next steps, does name a sample. It also appears on two of that vendor's own pages with two different comparison groups, once against slow deals and once against average ones. Both groups were in the original analysis, which is precisely why one number should not be printed against either without saying which.

That is enough to stop quoting them, and not enough to conclude the metric is worthless. What the evidence supports is narrower: the marker is a fact about the recording. Whether it causes the close or reports it is the circular question, and the piece on quiet deals settles that. Treat the marker as a gate you can act on, not a predictor you can forecast with. We can show you what share of your own calls clear that gate.

A CRM field and a recording are counting different things

The same metric name sits on two very different measurements, and teams compare them as though they were one. The pipeline field is filled in after the call by the person who ran it, and it records intent: what the rep plans to do next. The recording records what was actually said out loud, by whom, and whether the other side responded to it.

Neither is wrong. They answer different questions, and the gap between them is itself the interesting number. We built on the second deliberately, and the product says so in plain words: deal status comes from what was said in the calls, not from pipeline stages. That choice costs us something. It means we cannot tell you about a commitment that was made over email after the call, and we say so rather than pretending the recording is the whole record.

A no-next-step flag supports exactly one conclusion on its own

A flag that says "no next step" is a fact about the recording. Everything a manager wants to do with it depends on why, and the three reasons below look identical in a column of red cells.

Closing talk lives at the end of a conversation, which is exactly the part most likely to be missing when a recording is truncated. The third row therefore reports an absence of data as an absence of a next step, and the two are not the same finding.

There is a fourth case the table leaves out, where the flag reads green and should not: the rep named a date and the customer never took it. Who spoke the date is settled in the piece on quiet deals, and this piece does not reopen it.

All three need a reason attached before anyone acts, and attaching reasons is the whole job. Any such queue, including the one deal recovery works from, is only as good as the reasons attached to it. A flag with nothing behind it is work that has to be redone by whoever opens the row, which is why the reason belongs in the row and not in a manager's head.

Agreed is not the same as happened

There is a measurement nobody takes, and it would be more useful than the one everybody takes. A next step agreed on a call is a promise about the future. Whether that meeting actually occurred, with the people it was supposed to include, is a separate fact that lives in a calendar.

We do not check it. There is no connection to any calendar or pipeline system in our product, so our own number counts promises rather than kept promises. You are not under that constraint. Take last quarter's calls that ended with a date and open the calendar for the fortnight after each one: count how many of those meetings exist as events at all, and of the ones that exist, how many list the person who was named out loud on the call. Two counts, one afternoon, and nobody we have looked at publishes them, ourselves included.

The size of the gap is suggested by how buying actually works now. Forrester's survey of nearly eighteen thousand business buyers worldwide, run in 2025 and reported on its blog in January 2026, counts an average of thirteen internal and nine external people influencing a purchase, and reports that more than sixty percent of buyers pay for some kind of trial while just over a third said they planned to convert to a full paid version with the same supplier. The methodology is not published on the open page, so treat these as the firm's stated findings. The direction is the useful part: a commitment from one person among twenty-two who influence the decision is a smaller event than a dashboard implies. And notice what the trial figure actually says. It reports what buyers planned to do, not what they did, which is the same gap this whole piece is about. Which two of the four a recording can settle, and which two it cannot.

Report this number weekly and it will go up without anything else changing

Measuring a behavior changes the behavior, and in enterprise sales there is a real study of it rather than a parable. Ian Larkin's analysis of a single enterprise software vendor covers 7,912 deals closed by 412 salespeople across 28 fiscal quarters between 1997 and 2003 in North America. Nearly seventy percent of deals closed on the last day of a quarter. Separately, the discounting salespeople used to land deals in the quarters that paid them best cost the vendor an estimated six to eight percent of revenue, because the prices they accepted were lower than they needed to be.

The limits are worth stating. It is one vendor, the data is more than twenty years old, it is North American enterprise software, and more than nine in ten of the salespeople were men. The dataset was supplied by the vendor being studied, and the research was funded by the Ford and Truman foundations. The study measures gaming of close timing, not of next steps. Carrying it across to this metric is an argument by mechanism, not a measurement, and we are calling it that.

The mechanism carries cleanly enough. Producing one costs a sentence at the end of a call and a field afterwards. If the number is reviewed weekly and nothing checks whether the meeting happened, the number will go up and nothing else will. That is one practical reason to keep the check at the level of the recording, where producing a passing result requires actually having the conversation. It is also the reason this number belongs in a sales leader's weekly read rather than in a target.

Our flag answered the question of fact and broke the advice on top of it

Our own check answered the question of fact and nothing else: was a next step set, yes or no. That turned out to break the advice built on top of it. On a call where the rep had run a genuinely good conversation and simply forgot to say the closing sentence, the block that recommended what to fix first told them to fix the structure and the summary of the consultation. The advice was confidently about the wrong thing. We opened a random call at a second client and landed on the same case.

We turned that block off for everyone rather than tuning it. In its place there is now a plain screen of calls with no next step, where each one carries the reason and the quote that produced it, and a manager decides. The lesson we took is specific: a flag with no reason attached does not become useful advice by being phrased more helpfully.

The second thing is narrower and took longer. When a recording is incomplete, we used to report the next step with the same confidence as everything else. The argument about whether to keep doing that ended when somebody acted on one of those numbers.

What to do next

Export last month's deals that your pipeline marks as having a next step. Open the recordings for twenty of them and count how many contain a next step at all, in anyone's voice. The number of deals where the field says yes and the recording says nothing is this metric's error rate, and it is the figure to put in front of anyone about to set a target on it.

Then pick the most recent calls flagged as having no next step and sort them into three buckets: the closing sentence was never said, the account does not work that way, or the recording is cut short. If more than a handful land in the third bucket, the problem is not the sales conversation.

Start with the calls, not the field. Send us a month of recordings and we will mark which deals ended without a spoken next step, and which of the four reasons the recording can actually settle. We will name the ones it cannot.