Where we are
b7 sent the Northwind proposal and set a four-touch cadence running. Tonight everything you learned in that deal has to survive being written down. The deliverable is one real opportunity record - yours, the worst one you have - rewritten until it passes the inheritance test, plus a three-question audit of every field your team is asked to fill. b9 forecasts from exactly this material, so anything you fudge here comes back next session as a number nobody believes.
Write for the person who inherits your territory 7 min live
The reason CRM hygiene never sticks is that everybody has been told the wrong reason for it. It is not for management, it is not for the dashboard, and it is not for you - you have the deal in your head already, which is exactly why your notes read like shorthand. It is for the person who takes this account when you get promoted, go on leave, or leave the company. Write every record for them and the format solves itself. Here is the same Northwind opportunity, twice.
LiveWhat AI genuinely fixes, and the one thing it never will4 min▶
Be precise about this, because vendor claims are not. There is a real list of things automation does well here, and it is worth having, and it stops abruptly at a line you can name.
| Job | Can AI keep it? | The catch |
|---|---|---|
| Capturing the next step from a call | Yes, genuinely well | It captures what was said. If nothing was agreed, it must say so rather than invent one |
| Filling structured fields | Yes, where the field has one meaning | Garbage in a defined field is still findable. Garbage in an undefined field is noise forever |
| Deduping contacts and accounts | Yes, this is the strongest case | Merge rules need a human decision once, then it runs |
| Flagging stale records | Yes, and it is thankless work nobody else does | Flagging is not deciding. Somebody still has to kill the deal |
| Deciding what a field means | No, and it never will | This is an agreement between people, not a data problem |
That last row is where most CRM projects actually die. A team adds "Deal health" as a picklist, three reps read it as sentiment, two read it as stage risk, and RevOps builds a dashboard on the average. The tool is fine. The definition never existed.
LiveWhere the five saved hours quietly go3 min▶
Gartner found AI gives sellers back nearly five hours a week, and that 72% of sales organisations fail to reinvest that time in high-value activities. CRM admin is the most common place those hours reappear, and it is worth understanding why that is a trap rather than a win.
- Auto-capture generates more text, not more clarity. A full transcript summary pasted into a notes field is longer than what you would have typed and less useful, because nobody reads it and nothing in it is structured.
- More rows is not more record. Ten logged activities with no next step is a record that says "somebody was busy here". Your successor cannot act on busy.
- The reinvestment that pays. Spend the saved time on the two lines that are pure judgment: the agreed next step with its date, and the honest sentence about why the deal might die. Neither can be generated, and both are what b9 forecasts from.
So the honest version of "AI keeps your CRM clean" is narrower and better: AI removes the typing, which leaves you with the thinking. If the thinking does not happen, you have automated the production of tidy-looking noise.
Self-studyWhy your own shorthand betrays you in eleven weeks2 min read▶
The inheritance test has a version that applies even if you never hand the territory over: future you is a colleague who has never met this buyer. Eleven weeks after the call, "good chat, they're keen, will revert" is exactly as useful to you as it is to a stranger, which is to say not at all.
- Names without roles decay first. "Spoke to Tom" is unrecoverable six months later. Tom Becker at Halden Group, and what he decides, is not.
- Sentiment decays second. "Positive" tells you nothing about what changed. "She said the Tuas ramp lost two weeks to decisions nobody wrote down" still works a year later.
- Dates never decay. A dated fact is the only kind that ages well, which is the same reason the trigger lever is the heaviest one in b3.
Fields that earn their existence 6 min live
Every field on your opportunity layout costs a rep a few seconds, on every deal, forever. That is the price. So the field has to buy something, and the only thing worth buying is a changed decision. Run three questions on each one - who reads this, what decision does it change, what happens if it is blank - and a surprising proportion of your layout cannot answer any of them. Those fields are not neutral. They are the reason reps type "TBD" and stop trusting the whole object.
LiveDecay, and the rail that does not decay4 min▶
Two kinds of data live in your CRM and they behave in completely opposite ways, which is why treating them the same is the most expensive habit on this page.
- Contact data goes stale fast. People change roles, companies, domains and mobile numbers constantly. A list you bought a year ago is not a list, it is a hypothesis, and every bounce off it costs you sender reputation you need for the accounts that matter.
- Opt-out status never goes stale. It does not expire, it does not reset when someone changes jobs inside the same company, and it is not undone by a new sequence in a new tool. Marcus Hale at Continental Cold Chain opted out. He is a permanent zero, in b3, in b7's cadence, and here.
The AI-era wrinkle is scale. Suppression used to fail quietly and small: a rep forgot, forty people got an email, somebody apologised. Now the same failure runs through a sequencer at volume that used to take a week and takes a minute, so a single unenforced opt-out becomes a much larger compliance event than the identical mistake was three years ago. The rule has not changed. The blast radius has.
Suppression fails between tools, not inside them. Nearly every real incident looks the same: the opt-out is correctly recorded in the CRM, and the sequencer, the marketing platform, or a rep's personal list never received it. Each system is individually compliant and the person still gets the email. So the test is never "is it in the CRM" - it is "name every tool that can send from our domain, and show me the suppression list arriving in each one".
Self-studyHow to actually delete a field without a war3 min read▶
Field audits fail politically, not technically. Somebody asked for that picklist in 2023 and will defend it on principle. Three moves that work:
- Report on fill rate before you argue. A field filled on 11% of opportunities is not a debate about value, it is a fact about behaviour. Bring the number, not the opinion.
- Ask for the report that uses it. Not "is it useful" - "show me the dashboard, and tell me the last decision it changed". Silence answers the question for you.
- Hide before you delete, and say so. Remove it from the layout, keep the data, set a date to revisit. Nobody objects to hiding, and nobody ever asks for it back.
And the estate rule underneath all of it: if two fields describe the same thing, they will disagree, and it is a question of when rather than whether. Merge them before you add a third.
Rewrite one record until it passes ★ 24 min · everyone builds
Open the worst live opportunity you own. Not a tidy one - the one you would be embarrassed to hand over tomorrow. We rewrite it in the room until a colleague could run it, then we build the capture prompt that keeps it that way without inventing anything.
Read the record out loud as if you have never met the buyer. Mark every line that is activity rather than information. That is usually most of them.
Write the true line: what is factually happening at this account, dated. Not sentiment, not your read. Facts a stranger could verify.
Write the agreed line: what did both sides actually commit to out loud? If nothing, write "nothing agreed" - that is the single most useful sentence in the record.
Write the next line with an owner and a real calendar date. "Follow up" is not a next step. "Next week" is not a date.
Write the why it might die line. One sentence, honest, naming a person if a person is the risk. b9 runs on this line and nothing else.
Add every stakeholder with a role and what they decide. Mira decides, Priya approves, Jonas can veto. Three names, three verbs.
Log the trigger you originally used, with its date. Jurong lease, 21 July. This is what turns your pipeline into a pattern next session.
Now run the good prompt below on your latest call recap and compare its output to what you just wrote by hand. Where it returns [NONE AGREED], check whether that is the tool failing or the call failing.
From the call recap below, extract exactly three things and nothing else. RECAP: [PASTE RECAP] Return this structure: NEXT STEP: the action both sides said out loud OWNER: the named person who does it DATE: an explicit calendar date mentioned in the recap Hard rules: - Use only what is stated in the recap. Do not infer, estimate or round a date. "Next week" is not a date. - If any of the three was not agreed out loud, output exactly [NONE AGREED] for that item. Do not substitute a likely value. - If all three are [NONE AGREED], say so in one line and stop. - Then list any commitment WE made, quoted verbatim, so I can check it against what is in the proposal.
Summarise this call and write it into the CRM. Include a summary, sentiment, next steps and a suggested close date.
The bad prompt has no schema and no refusal rule, so every field gets something plausible. Sentiment is invented from tone. Next steps are inferred from what usually happens on calls like this. The close date is a guess dressed as data, and it is the guess your manager will forecast from in b9. Worse, it all reads well, so nobody checks it - which is precisely why an ambiguous field is more dangerous once AI is filling it than it was when reps left it blank.
LiveAuto-captured next steps, and why the date matters more than the text3 min▶
Of everything in this session, next-step capture is the feature that genuinely earns its licence fee. But teams grade it on the wrong half of the output.
- The text is a nice-to-have. "Send the security overview to Jonas" is helpful, and a colleague could reconstruct most of it from the recap anyway.
- The date is the whole product. A dated next step is the only field in your CRM that is self-invalidating: when the date passes and nothing happened, the record tells you the deal is not moving, without anybody having an opinion about it. Every stalled-deal report worth running is built on that one property.
- Which is why "next week" is a bug. A vague date never passes, so it never triggers anything, so the deal sits in Proposal for six weeks looking alive. That is the record on the left of the diagram.
- And why [NONE AGREED] is a feature. A call that ended without a dated commitment is a call that did not advance. Recording that honestly is more valuable than a plausible date, because it is true, and because it is the thing b9 needs.
LiveMultithreading, recorded properly3 min▶
Single-threaded deals die when one person changes job, goes on leave, or simply goes quiet, and the record almost never shows it coming because it lists contacts rather than roles. Record three things per stakeholder and the risk becomes visible on the page:
| Person | What they decide | What we have actually done |
|---|---|---|
| Mira Chen, VP Ops | Whether this solves a problem she has | Two calls, proposal sent, engaged |
| Priya Raman, CFO | Whether the number is defensible | Joining 18 Sep. Has not seen the value case |
| Jonas Weber, IT manager | Can stop it on security or integration | Nothing. Never contacted. This is the risk |
Written like that, the "why it might die" line writes itself, and so does the next step. A contact list would have shown three names and told you nothing. Note also that Jonas is the wrong seat for outreach (b1 and b3 capped his row at 1.8%) and exactly the right seat for a deal in flight - the same person can be a bad prospect and a mandatory stakeholder, and only the record knows which stage you are in.
LiveWhat to do with a deal you know is dead2 min▶
Everyone is carrying one. It has been in Proposal since June, the champion stopped replying, and it is still in your number because removing it feels like an admission. Three things are true about it:
- It is costing you more than it can pay. Not in effort - in attention, and in the credibility of every other number you give your manager.
- Closed-lost is data, not failure. Record the real reason in words a stranger can read: no budget this cycle, lost to an incumbent, champion left, never reached the decision maker. "Went quiet" is not a reason, it is a symptom.
- Send the honest close-out from b7 first. A real ending sometimes produces a real answer, and it always leaves the door open cleanly for next year.
Then close it. b9 is a session about a forecast people believe, and nothing destroys that faster than a pipeline everyone privately knows is padded.
Self-studyWhat your CRM's own AI sees, and who else can search your notes3 min read▶
The moment your CRM gained AI features, your notes field stopped being a private scratchpad and became training and retrieval material for the whole organisation. Almost nobody was told this explicitly. Four things worth knowing before your next entry:
- Scope. Assume anything you type into a record can be surfaced by a colleague's search or an assistant's answer, including a manager asking a question you never anticipated. Write accordingly - honestly, professionally, and about the deal rather than the people.
- Recordings and transcripts. These are usually indexed too. The consent you captured before the call (b5) governs recording it; it does not automatically settle who inside your company can later search it. Ask where that is documented.
- Personal data. A CRM note is personal data about a real person under PDPA and GDPR. "She sounded stressed about her boss" is a sentence that can be subject-access requested. Leader session a5 covers the org-level version of this.
- Agents are coming whether they help or not. Gartner predicts AI agents will outnumber sellers 10 to 1 by 2028, while fewer than 40% of sellers will say agents improved their productivity. Those agents will read exactly what you wrote. A clean, structured, honest record is the difference between an assistant that helps and one that confidently repeats your shorthand back to your CFO.
Try it yourself - this week ◐ 45 min total
- Run the three-question test on every field your team is asked to fill: who reads this, what decision does it change, what happens if it is blank. Bring the list of fields that fail all three to your next RevOps conversation with fill rates attached.
- Rewrite three real records until they pass the inheritance test. Then hand one to a colleague cold and ask them what they would do on Monday. Their answer is your score.
- Check that your suppression list is actually enforced across every tool that can send from your domain, not just the CRM. Name the tools first, then verify each one, because the failure always lives in the gap between two systems.
- Add the trigger you used, with its date, to every open opportunity you own. It takes about fifteen minutes and it is the raw material b9 forecasts from.
- Find the deal you know is dead. Send the honest close-out from b7, then close it with a real reason written in words a stranger could read.
- Set your capture tool to return [NONE AGREED] rather than a guess. Then count how many of last month's calls ended with no dated commitment. That number is a discovery problem, not a CRM problem.
Official sources covered
This session teaches the working core of stage 5 of the deal loop plus the compliance rail that has to survive contact with real tooling. Legal specifics stay with your counsel and your data-protection owner - this page is orientation and a checklist to take to them. This page covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What is the actual test for whether a CRM record is any good?
A record answers four things: what is true, what was agreed, what happens next and when, and why it might die. Completeness and volume are not the same as inheritability - ten logged activities with no dated next step tells your successor only that somebody was busy.
2 · Your call-capture tool returns a next step but no date was ever agreed on the call. What should it output?
The date is the whole product: a dated next step is the only field that invalidates itself when the date passes and nothing happened. A guessed date never triggers anything, so the deal sits in Proposal looking alive - and the guess is what your manager forecasts from in b9.
3 · A prospect's opt-out is correctly recorded in your CRM and they receive a sequence anyway. Where did it fail?
Contact data goes stale fast; opt-out status never does and never expires. The failure lives in the gap between systems, which is why the test is "name every tool that can send from our domain and show me the list arriving in each one". AI-era volume means the same old mistake now has a much larger blast radius.