How to build a weekly sales coaching plan from last week's calls

Most advice on sales coaching is advice about cadence. Book the one-to-one, protect the slot, hold it weekly, do not cancel it. All of that is reasonable. None of it answers the question a sales leader actually has on Sunday evening: what to say once the meeting starts.

Frameworks step in at that point, and they answer it with a list of things a seller should be good at in general. Run one twice and it returns the same plan, because nothing inside it changed when the week changed. You could have written the output before the week started. Nothing in it came out of the week.

Gartner put the consequence bluntly in December 2024: sellers are overwhelmed by the skills and tools they are expected to carry, and "traditional strategies of more training, resources and coaching are not leading to improved performance". Read that as a position rather than a measurement: Gartner publishes no sample behind that particular sentence. The direction still matches what most sales leaders already suspect. The amount of coaching was never the problem.

Where the weekly plan actually comes from

A weekly coaching plan is one decision per person: what this rep will do differently on their next call, in words specific enough to repeat. That decision comes out of a conversation that happened, because the exact wording only exists there. General feedback is agreed to and then not acted on, because there is nothing in it specific enough to use on Thursday.

The arithmetic works against you before anything else does. A team of eight running five conversations each is forty calls a week. A leader who listens to four of them is doing more than most, and is still choosing what to coach from a tenth of the evidence, picked mostly by which calls were easy to find.

Sellers report the same gap from their side. In the seventh edition of Salesforce's State of Sales, fielded across 22 countries in August and September 2025, 46% of sales reps said they rarely get feedback on their sales conversations, and 40% named their manager's lack of time as an obstacle to enablement. Those answers come from the 1,032 reps inside the survey rather than from all 4,050 respondents, and the figures sit in the downloadable report rather than on the overview page.

A head of sales at an education company described the same thing to us from the other end. Manual review covered about a tenth of their calls, and the picture it produced never moved: everybody seemed to be calling well. That is one conversation and not a statistic, but it recurs often enough that reviewing everything instead of a sample is usually the first thing that changes what a coaching plan contains.

Which behaviour deserves the hour on Monday

Which behaviour do you give the hour to when six of them went wrong last week? Take the one where two people in the same role got different results. Rank behaviours by how much they vary across your team, not by how often they go wrong. A check that nearly every call fails cannot tell you who needs the hour, because it separates nobody from anybody.

We learned this by shipping the wrong version of it. One of our early weekly numbers was the share of calls containing at least one serious mistake. On a team that had been running for a while, that share sat so high that every rep looked identical, and the number was arithmetically correct the whole time. It was replaced by the average count of those mistakes per call, which spreads a team out properly, and the fix took about a day of work after several weeks of nobody noticing there was anything to fix.

The failure mode here is specific and it is not laziness. A number everybody fails looks like the most urgent thing on the dashboard, so it takes the top slot in the weekly review, and the behaviours that actually differ between people get pushed down to where nobody scrolls. The meeting then runs on the one topic that was already settled. If you want to know which behaviours separate your own team, that is the one thing we can run against a month of your recordings.

Two kinds of behaviour to cross out before you rank them

Before you rank anything, cross out every behaviour that forms part of how you define a good call, and every behaviour nobody has failed often enough to conclude anything about. What is left is a short list, and the shortness is what makes it usable on a Monday morning.

The first kind tops every ranking by construction, whatever the data says, because it is guaranteed by the wording rather than discovered. That is the trap we wrote about in the piece on deals going quiet after the demo, where the agreed next step is both the thing you want and the thing you are measuring.

The second kind is quieter and costs more than it looks. A behaviour missed twice in a quarter can produce a spectacular influence figure, because two calls is not a sample. Leave those in and your Monday runs on a behaviour two calls wide, while the behaviour forty calls wide waits until March. In Spellit Actions both kinds are stripped out before the ranking is drawn, and the screen says so where the ranking appears, because a filter nobody can see reads as a missing row.

Our second version of the skills map had a related flaw, and it was worse because a manager carried the chart into the one-to-one. The map averaged pass rates across behaviours, so a behaviour a rep had been scored on exactly once came out at 100% and pulled the whole shape outward. The rep looked strong at something they had demonstrated a single time. It is now weighted by the number of calls behind each behaviour, which is duller and correct.

One behaviour a week, and why adding sessions does not help

Give each rep one behaviour for the week. Not three, and not a theme. Adding sessions will not do the work that naming the behaviour does, and that is the part of this the evidence is clearest about.

A 2023 meta-analysis of workplace coaching in Frontiers in Psychology, covering 11 studies and not one of them about sales, found the number of coaching sessions was not a significant predictor of the outcome (Z = 1.03, p = 0.30). Skill-related outcomes scored higher than affective ones, g = 0.72 across five studies against g = 0.41 across eight, though the authors state plainly that the two are not significantly different. So the honest reading is narrow: more sessions is the lever most teams reach for, and it is the one the evidence does not support. What to aim the session at is the more promising lever, and the research on it is thinner than the way it usually gets quoted.

What goes into the one-to-one: seven parts

The structure of the conversation matters more than its length. Each part answers the objection the previous one raises.

1. Why this behaviour got picked and not one of the other five, in terms of what it costs.
2. What it sounds like now, next to what it should sound like. Two versions, side by side.
3. How to actually do it, in enough mechanical detail that it can be followed under pressure.
4. The words themselves, close enough to spoken English that the rep can use them on Tuesday without translating first.
5. A teammate's recording where those words worked, so it is not a theory about these buyers.
6. The rep's own recording where it went wrong.
7. What will count as success next week, named before the meeting ends.

Part six is the one that changes the meeting. Coaching from memory turns into a disagreement about what was said. A recording settles that outright and moves the conversation to what to do instead, and that is the only part worth the hour. Part five is the step that quietly does not happen, because nobody has time to search a month of calls for the recording where a colleague got it right.

The check the week after is what makes it a plan

Monday's plan and the following week's calls usually never meet. The plan is finished when somebody confirms the behaviour turned up in a conversation that happened afterwards, not when the rep marks the task complete. Without that step, Monday produces a set of intentions, nobody finds out which survived contact with a live buyer, and the same plan gets written again seven days later. Finding the pair of recordings that makes the check worth having is the part that stalls by hand, and we can pull those out of a month of your own calls so you can judge the plan yourself.

Mechanically this is unglamorous. A task moves into a review state, the leader confirms it against a real call or returns it with a written reason, and the history is append-only so the record of what was agreed cannot be tidied up afterwards. We built that review step before a single customer asked for it, and building it turned out to be the easy half. Those decisions need a slice of somebody's attention every week, and it comes out of a week that is already full. No column in a tool creates that time.

One more piece of plumbing is really a management decision. Thresholds carry the date they came into force, so raising the bar in week six does not repaint weeks one to five, and a report for March keeps reading the way it read in March. Move a bar retroactively and the team learns that the scoreboard is negotiable, after which everything you measure is worth less than it was. So a weekly coaching cycle is worth building around dated thresholds before you spend much time arguing about which behaviours to track.

There is also a legal edge here that coaching write-ups rarely touch. The EU AI Act treats AI used to evaluate people at work as high-risk under Annex III, point 4, requires human oversight of such systems under Article 14, and gives the affected person a right to an explanation of a decision under Article 86. In practice a rep has to be able to challenge a score and a human has to be able to overturn it. One thing we added past what the law asks: count how often those challenges succeed. A system that never overturns itself is not being checked.

Four sales coaching statistics with no traceable source

Four figures turn up in most articles on this subject, and we could not reach a publication behind any of them. This matters for a weekly plan, because a target set from a number nobody can check is a target you cannot defend the first time somebody asks where it came from.

One figure we expected to add to that list survived, and the way it survived is worth a sentence. The claim that only 26% of sellers get weekly one-to-one coaching is real and traceable: it comes from the fifth edition of Salesforce's State of Sales, fielded in August and September 2022 across 7,775 respondents in 38 countries. What almost nobody quoting it mentions is the date. Our own first check looked for it in the current edition, did not find it, and nearly concluded it was invented.

The rule that falls out of this is small. Before a number goes into a plan your team gets measured against, find the sample size and the year behind it. If either one is missing, leave the number out of the plan.

What to do next

Take last week. Pick the rep you are least certain about, open two of their calls, and write one sentence: the behaviour, and the wording you want instead of it. That sentence is the plan.

Put it at the top of Monday's one-to-one, and put a note in your own calendar for the Monday after to open one call from the week in between and check whether the wording appeared. One rep, one behaviour, one check. Three weeks is enough to see whether the loop holds, and if you miss the check twice, cut back to one rep rather than the whole team and keep the check. The part that does not scale by hand is knowing which behaviour to pick, and that is what we can run against a month of your own recordings.