What this session asks you to sign
You are still the VP of Sales at Cadence, and the leader track has given you the evidence (a1, a2), a forecast people believe (a3), an enablement model (a4) and the compliance rails (a5). None of it matters until something ships. Tonight is the commitment session: a 90-day plan with written kill criteria, a paired metric set your team cannot game and AI cannot inflate, and a one-paragraph statement of where the saved hours go that you read out to the team. Three artefacts. If you leave with only one, make it the third.
A stack you can undo 15 min live
Most bad sales-AI purchases are not bad products. They are good products bought in a shape that cannot be reversed. Four rules carry almost all the weight here: buy for one workflow at a time, insist on data export before you sign, never buy anything that owns your record of truth, and pilot with your two best reps and your two worst. That last one is the honest test. If a tool only helps the people who were already good, it is not enablement, it is amplification - and amplification widens the gap on your team while the average stays exactly where it was.
LiveThe four buying rules, and why each one is really an exit rule5 min▶
Every rule below is about the day you want to leave, not the day you arrive. Vendors sell you the arrival. You have to underwrite the departure yourself.
- One workflow at a time. A tool that does prospecting, sequencing, call recording, forecasting and coaching cannot be evaluated, because you can never tell which part is working. Buy the narrow thing, prove it, then widen. It also means removing it later costs you one workflow, not five.
- Insist on data export. Written into the contract, tested during the pilot, in a format you can actually load somewhere else. "We have an API" is not the same as "we have an export", and you find out which you bought at the worst possible moment.
- Never buy anything that owns your record of truth. Notes, next steps, stage, close date, contact history - those live in a system you control. A tool that becomes the only place a fact exists has quietly acquired your ability to switch.
- Pilot with two best and two worst. Four people, thirty days. The gate is whether the two who struggle improve. Top reps make every tool look good, because top reps make everything look good.
Self-studyEnablement or amplification: how to tell which one you bought3 min read▶
Run the numbers on the four pilot reps separately, never as an average. The average will look fine in almost every case, because your two best reps will pull it up on their own.
- Enablement - the two struggling reps close a visible part of the gap. Their prep gets better, their recaps go out, their pipeline hygiene stops being a monthly rescue. The tool supplied something they were missing.
- Amplification - the top two get faster and the bottom two produce more of the same low-quality work, now at speed. The gap widens. You have bought a multiplier, and multipliers work on negative numbers too.
- Neither - everyone reports it is "fine" and nothing measurable moved. This is the most common result and the hardest to act on, which is exactly what the kill criterion is for.
Amplification is not automatically a bad purchase. It is a bad purchase if you told the board it would lift the team, because it will not, and in nine months someone will ask why the average is flat.
Adoption without a mandate 12 min live
Mandate a tool and you get malicious compliance: logins on Monday, the dashboard satisfied, the work done the old way in another tab. Nobody is being difficult - they are protecting a number they are paid on from a tool they do not yet trust. The fastest adoption I have ever seen in a sales team came from one rep saying at standup that she got an hour back on Tuesday and here is exactly how. Gartner's numbers point the same direction: organisations providing sellers with AI-enabled next best actions are 2.6x more likely to achieve commercial growth, and those emphasising upskilling are 2.4x more likely to see good revenue growth. Both findings are about what the organisation does around the tool, not about the tool.
LiveHow to run adoption when you refuse to mandate4 min▶
Refusing to mandate is not the same as doing nothing. It means you spend your effort on demand rather than on enforcement.
- Make one win visible every week. Two minutes at standup, by the rep, not by you or the vendor. Specific: what she did, what it replaced, how long it took. Vague enthusiasm converts nobody.
- Pick the demonstrator carefully. Not your top performer - the team discounts them. The credible one is a solid middle rep who is known for being sceptical.
- Remove the old path only after the new one is preferred. Turning off the spreadsheet before people want the tool is a mandate wearing a different hat.
- Fund the upskilling, not just the licence. This is where the 2.4x lives. A tool with no teaching attached is a subscription, not a capability.
- Give people next best actions, not dashboards. The 2.6x finding is about the system telling a seller what to do next, in context, not about giving them another place to go and look.
LiveThe honest forecast: plan for tools arriving whether they help or not4 min▶
Two Gartner findings sit uncomfortably next to each other and you should hold both. AI agents may outnumber sellers 10 to 1 by 2028, and fewer than 40% of sellers will say agents improved their productivity. Read together, that is not a prediction that AI fails. It is a prediction that volume of tooling and value of tooling come apart, and that most of the arriving software will be mediocre for the people using it.
- Assume the tools arrive anyway. Procurement, your CRM vendor's roadmap and your own board will keep adding agents whether or not you asked. Your job is the filter, not the gate.
- Budget for removal. Most teams budget to buy and never budget the hours to rip something out. Put the removal cost in the business case at purchase time.
- Protect attention as the scarce resource. Ten agents each saving four minutes and each demanding attention is a net loss. Count the tabs, not just the features.
The two-tab test. Watch a rep work for twenty minutes without commenting. Count how many tools they open to complete one task they do many times a day. If the answer is more than two, your problem is not that they need another tool, it is that the ones you bought do not talk to each other. That finding is worth more than any vendor demo you will sit through this quarter.
Metrics that resist gaming, and where the hours go ◐ 12 min · cards
Here is the sharpest thing in this session. Activity metrics stop being evidence the moment you deploy AI - sends, calls logged, tasks completed and notes written are precisely the outputs a model can produce for free and at volume. They were always a weak proxy; now they are a broken one. Replace them with paired metrics that require a real human on the other side, and be honest with yourself about comp: think very hard before paying anyone for something a model can produce without a customer being involved.
LivePaired metrics: never one number on its own4 min▶
A single metric is an instruction to maximise it. A pair is much harder to satisfy dishonestly, because the second half needs a buyer to do something.
| Stop using alone | Because AI inflates it | Use this pair instead |
|---|---|---|
| Emails sent | Free and unlimited from day one | Reply rate and meetings held and pipeline created from those meetings |
| Calls logged | Auto-logged by the dialler and the note-taker | Conversations over two minutes and next steps agreed on the call |
| Notes written per deal | Generated wholesale from a transcript | Deals where the next step is dated and the successor test: can another rep pick this up cold |
| Tasks completed | Completion is a click | Stage progression and cycle time, measured on the same cohort |
| Forecast accuracy alone | Rewards sandbagging, as covered in a3 | Accuracy and commit coverage and how often the rep overrode the model and was right |
The rule underneath: a metric is only evidence if a person outside your company had to do something for the number to move. Everything else is now free.
LiveWhere the hours go - the closing question of the whole track4 min▶
Gartner: AI returns sellers nearly five hours a week, and 72% of sales organisations fail to reinvest that time in high-value activities. Note what that sentence does not say. It does not say the hours are imaginary. They arrive. They simply get absorbed by more tabs, more admin and more sending, because nobody decided otherwise, and an undecided hour always goes to whatever is loudest.
- Name the destination in writing, in advance. Two more discovery conversations a week. One deep account plan. Ninety minutes of live coaching. Pick something and write it down where the team can see it.
- Do not let the answer be "more sends". That is the 72% outcome with a positive label on it, and a2 already showed you where more sending leads.
- Measure the destination, not the saving. Nobody can audit "we saved five hours". Everyone can see whether coaching happened on Thursday.
- Say it out loud to the team. A commitment made only in your head is not a commitment, it is an intention, and intentions lose to the inbox every single time.
The 72% is not a competence problem, it is a decision problem. The organisations that reinvest the time are not smarter or better resourced. They wrote the destination down before the hours arrived. That is the entire difference, and it costs one paragraph and one uncomfortable standup where you say out loud that the reclaimed time is not free capacity for more volume.
LiveComp: what you will and will not pay for4 min▶
Comp is the loudest thing you say. Whatever it rewards, that is your actual strategy, and reps will find the shortest path to it within about three weeks. Three positions worth holding:
- Do not pay on activity. You are now paying for output a model produces at zero marginal cost. This was always a weak plan; it is now an arbitrage you are funding.
- Do not pay on forecast accuracy alone. a3 covered the Goodhart trap in full: pay for accuracy on its own and you buy sandbagging, delivered reliably and forever.
- Think hard before paying for anything a model can produce without a customer. Notes, recaps, updated fields, meeting summaries. Pay for the outcome those things support, not the artefact.
- Do keep paying for the human half of the trust line. The read, the promise and the price. Deals closed, pipeline that survives a scrub, accounts that renew and refer - these still require a person, and they are still what you want more of.
Self-studyThe 90-day plan, on one page3 min read▶
Days 0-15. Pick one workflow. Write the goal, the gate and the kill criterion in one document. Confirm data export in the contract and test it. Name the rails owner from a5.
Days 15-45. Pilot with four reps: two best, two struggling. Baseline their paired metrics before day one, because you cannot reconstruct a baseline afterwards.
Days 45-60. Read the four reps separately. If only the top two gained, say so in writing and hit the kill criterion. This is the meeting where the plan earns its keep.
Days 60-75. If it passed, widen to one full team with the demonstrated-win pattern: a sceptical mid-performer showing a real saving at standup. No mandate, no SPIFF.
Days 75-90. Decide: keep, kill, or renegotiate. Then write where the hours went, share it with the team, and put the next review in the calendar before you leave the room.
Self-studyWhat to say to the board, and what to say to the team3 min read▶
To the board. Lead with the workflow, not the technology. "We are cutting call-prep time and reinvesting it in discovery volume; here is the pilot, the gate, and the date we kill it if it does not clear." Give them the kill criterion unprompted - it is the single fastest way to be trusted with the next budget, because it shows you are measuring rather than believing. Do not promise a headcount saving you have not evidenced.
To the team. Say the quiet part first: this is not a prelude to cutting the team. Then be exact about what stays human - the read, the promise, the price - because that is what they are actually worried about. Then name where the hours go, and be honest that you are asking them to spend the saving on harder work, not easier work. Reps handle that fine when it is said plainly and badly when it is discovered later.
Self-study">Your own AI use: forecast call, pipeline review, one-on-ones3 min read▶
You are asking the team to change how they work. It is worth being specific about your own, because they will notice either way.
- The forecast call. Let the model produce the first pass and the anomaly list, then spend the meeting on the deals where you disagree with it. Keep the override log from a3 - your overrides are the most interesting data in the business.
- The pipeline review. Use AI to surface stalled deals, missing next steps and single-threaded accounts before the meeting, so the meeting is about decisions rather than a status recital.
- One-on-ones. This is the boundary. Bring call insights, never a generated assessment of a person. A rep who suspects their coaching was written by a model will stop telling you anything true, and you will have traded the single best signal you have for a saved twenty minutes.
- Your own writing to the team. Draft with it if you like, then rewrite it in your voice. Your team can tell, and the trust line applies to internal communication exactly as it applies to buyers.
The three things you leave with ◐ 60 min total
- Write the 90-day plan with kill criteria. One page, four phases, one gate and one written kill criterion each. Date it and send it to whoever will hold you to it.
- Pick your paired metrics. Choose three pairs, retire the activity metrics they replace, and tell the team which numbers stopped counting and why.
- Write the where-the-hours-go commitment and share it. One paragraph, a specific destination, read out loud to the team. This is the 72% question, and it is the whole leader track in one sentence.
- Baseline before you pilot. Capture the four pilot reps' current paired metrics this week. A baseline you did not take cannot be reconstructed later, no matter how much you want it.
- Audit your own AI use against what you asked of the team. Especially the one-on-one boundary. If you would not accept it from them, do not do it to them.
Official sources covered
The final leader session closes the loop on the Gartner findings the track opened with, and converts them into rollout, adoption, metric and comp decisions. Benchmarks and forecasts are directional - treat them as the shape of the curve and measure your own team against your own baseline.
Three questions before you go 🎯 ◐ 90 seconds
1 · Your pilot ends and the average across the four reps looks good. What do you check before deciding?
Averages hide the only question that matters. If it only helps your best reps it is amplification, not enablement: the gap widens, the team average stays flat, and in nine months someone asks why. Pilot with two best and two worst precisely so this is visible.
2 · Why do activity metrics stop being evidence the moment you deploy AI?
Activity was always a weak proxy for outcome; AI makes it free, which breaks the proxy entirely. Use paired metrics instead - reply rate and meetings held and pipeline created from those meetings. The test to hold: a metric is only evidence if someone outside your company had to do something for it to move.
3 · AI returns nearly five hours a week per seller and 72% of organisations reinvest none of it. What separates the ones that do?
The hours do arrive. Undecided, they get absorbed by more tabs, more admin and more sending, because an unassigned hour goes to whatever is loudest. The organisations that reinvest are not smarter, they simply named the destination in advance and made it visible. That is one paragraph and one honest standup.