AI coaching only works for SDR teams when it has a time slot. A scorecard nobody opens and a roleplay bot nobody uses are the normal outcome of buying the software without a routine. Here are the seven plays we see work, placed on a real day with two call blocks, and the single metric to watch for each. Everything below assumes calls are scored automatically; if they are not, start there.

The rhythm first: blocks, not days

Our benchmark of 1,000,000 B2B cold calls in France says when to call: 10am connects at 14% and 4pm at 15.8%, against 9.8% at 8am. Founders pick up around 10am, marketing around 4pm, HR around 1pm. So the day is two blocks, 10am to 12pm and 4pm to 6pm, and the coaching plays sit before, inside and between them.

The north-star is 20 conversations per SDR per day. Every play below exists to turn more of those 20 into meetings, not to make reps dial more.

The seven plays

Play 1, 9:30: build the list by signal and slot

The first coaching decision is who to call. Reps use about 20% of the signals already in their CRM, and 50% of Sonik's own pipeline comes from signals that were already there. The morning play is a five-minute review of the day's list: job changes, website visits, closed-lost from last year, dormant contacts, matched to the slot. Founders in block 1, marketing in block 2.

Metric: share of the day's dials with a named trigger. Target above 70%.

Play 2, 9:50: ten minutes of roleplay on yesterday's objections

Not a library persona. The three objections the team scored lowest on yesterday, replayed by a simulated buyer. Ten minutes, before the first dial, with the opener included. 65% of reps still open by explaining what they sell; trigger-based openers reach 7-10% meeting rate versus 2-3% pitch-first, and the opener is the easiest thing to drill.

Metric: opener score on block 1 calls, versus the week before.

Play 3, block 1: live prompts on three triggers only

During the block, the model watches each call for three triggers: a known objection, a competitor mention, a pricing question. One line on screen within two seconds, drawn from a teammate's answer that booked. Nothing else fires. A rep who gets a prompt every ten seconds closes the panel by Tuesday.

Metric: prompt dismissal rate. Above 20% means the triggers are badly calibrated.

Play 4, 12:00: fifteen-minute score review, one criterion per rep

After block 1, each rep opens their scores. Not all of them: the lowest criterion and the two calls where it cost the meeting. The manager does the same at team level and messages two reps with one ask each for the afternoon. One behaviour per rep per week; reps told to fix five things fix none.

Metric: share of reps who opened their score before 1pm.

Play 5, block 2: discovery before pitch

Block 2 is marketing and sales personas, who connect above 15% in our data and are more likely to talk. The coaching focus switches from the opener to discovery. Questions starting with "how" lift conversion by 190%. Asking "What do you mean?" before answering an objection lifts it by 197%. The live prompt set for block 2 reflects that: fewer objection prompts, one "ask how" nudge after the pitch.

Metric: share of connected calls with at least one open question the prospect answered for more than ten seconds.

Play 6, end of each block: the close

The most common way to lose a meeting that was already won is "I'll send you an email". The score flags every call that ended on a next step without a date. The rep's end-of-block job is to call back the two best of those within the hour, while the prospect still remembers the conversation.

Metric: share of positive calls that end with a dated next step. Target above 80%.

Play 7, Friday 4:30: clips and next week's prompts

Thirty minutes, whole team. The three best-scored calls of the week, as 30-second clips, with the rep who made them explaining what they did. Then the manager promotes the winning objection answers into next week's live prompts, and the lowest-scored objections into Monday's roleplay. This is the loop closing: the top rep's call becomes everyone's prompt.

Metric: meetings per conversation, team, week on week. This is the only number that justifies the whole routine.

What this routine does not fix

Connect rate. If reps are getting eight conversations a day instead of twenty, the problem is the list, the caller ID reputation or the slots, not the coaching. Parallel dialing gives reps roughly 5x more conversations per hour; a better list and a clean number give them better ones. Fix that first, then run the plays. Coaching a rep who gets six conversations a day is coaching on a sample too small to learn from.

How Sonik approaches the daily coaching rhythm

Sonik's dialer is built around the two blocks. Sessions are assembled from signals in the CRM, matched to persona and slot, so Play 1 is the default rather than a discipline. Every call is transcribed in French or English and scored against a rubric that defaults to the four behaviours above: trigger opener, question before the objection, open discovery, dated next step.

Live prompts fire on a small set of triggers you choose, with text taken from your own team's best-scoring answers. Reps see their lowest criterion and the two calls behind it; managers get the weekly list and one-click clips for the Friday review. Scores, summaries and next steps are written to the CRM on call end, so Play 6's callbacks are already in the queue.

Conclusion

AI coaching is a routine with software in it, not software with a routine attached. Two blocks, a ten-minute drill before the first, one criterion per rep at noon, a Friday clip review that feeds next week's prompts. Start with Play 4 this week: ask every rep to open their lowest-scoring criterion at noon and pick one call to discuss. If that one habit sticks, the other six follow.