Where we are
This is stage 4 of the deal loop - Advance. Mira Chen has had two calls, Priya Raman has asked what it costs, and Jonas Weber wants to know where the recordings live. The deliverable tonight is a one-page Northwind proposal where every block is tagged with who is allowed to have written it, plus a four-touch cadence where each touch could arrive on its own and still make sense. b8 takes what you agreed here and makes it findable.
The promise and the price stay human 8 min live
In b1 the trust line had three parts: the read, the promise, the price. The read matters in outreach. The promise and the price both land in one document, and that document is the proposal. Here is the mechanic that catches people out: a language model is built to produce fluent, complete text. Given a proposal template with a gap in it, it will fill the gap confidently and plausibly, because a confident plausible sentence is exactly what it is for. The output is a commitment your company did not make, in writing, in your buyer's hands, with your name on it.
LiveAssemble versus generate, and why the difference is not pedantic4 min▶
Every argument about AI in proposals dissolves once you hold these two words apart.
- Assemble - take material that already exists in your notes, your recap, your price book, your security page, and put it in the right order in the right template. Nothing new enters the document. The risk is a clumsy sentence.
- Generate - produce text that was not there before because the template had a slot and the slot was empty. The risk is a commitment nobody made, written in the confident register of a company that intends to keep it.
The failure mode is specific and it is not exotic. You paste your call recap into a proposal generator. The recap does not mention implementation time, because Mira never asked. The template has an "Implementation" section. The model writes: "Typical deployment is completed within two weeks of kickoff." That sentence is now your company's position. Nobody in delivery has seen it. Priya will quote it back in the contract review, and she will be right to.
LiveWhy the buyer cares who wrote it3 min▶
Gartner predicts that by 2030, 75% of B2B buyers will prefer sales experiences that prioritise human interaction over AI. That prediction is usually read as a comment on personality. It is not. It is a comment on accountability. A buyer signing a six-figure commitment is asking a question that has nothing to do with tone: who is answerable for this sentence?
- A generated recap that is slightly wrong reads as carelessness. Recoverable, embarrassing, fixable in one email.
- A generated promise that is slightly wrong reads as either incompetence or bad faith, and the buyer cannot tell which. Both are fatal at contract stage.
- A generated price ends the conversation about value entirely, because you have signalled that the number was not considered - and a number nobody considered is a number that can be halved.
This is also why "AI wrote the first draft" is a fine thing to say about a recap and a terrible thing to say about a commercial term. Same technology, completely different exposure.
Self-studyThe one-page proposal, and what the extra thirty pages are actually for3 min read▶
Long proposals are usually a confidence problem wearing a formatting costume. If you needed thirty pages, discovery did not finish. The one-pager that works has six blocks and nothing else:
| Block | What goes in it | Length |
|---|---|---|
| What you told us | Their situation in their words, dated. Jurong DC live, three ops reqs open | 3 lines |
| What we would do | The scope you actually discussed. Sites, seats, integrations | 4 lines |
| What good looks like | The measure they named, not the one you wish they had named | 2 lines |
| What it costs | The number, the term, what changes it. Written by you | 3 lines |
| What happens next | Named owner, named date, for both sides | 2 lines |
| What we are not doing | Explicit exclusions. This block wins more deals than the pricing block | 2 lines |
Everything else - architecture diagrams, security packs, reference lists - is an appendix that answers a question someone asked. If nobody asked, it is padding, and padding tells a buyer you are filling space instead of understanding their problem.
The follow-up is where the work pays 7 min live
You did six minutes of trigger research, ran a discovery call, wrote a proposal, and then sent it once. The benchmark evidence says a 3-7-7 style cadence captures the large majority of replies by around day 10. Read that as an economic statement rather than a tactic: one-and-done sending throws away most of the value of work you have already paid for. And the reason reps stop is not laziness, it is that touch two has nothing in it, so sending it feels like begging. Fix the content and the cadence stops being uncomfortable.
LiveThe test that kills "just bumping this up"4 min▶
Here is the whole rule, and it takes one sentence: every follow-up must be sendable on its own. Imagine the recipient never saw your first email, or saw it and forgot completely. Read the follow-up as if it is the first thing that has ever arrived from you. If it collapses without the previous message, it is not a follow-up, it is a bump, and a bump is a request that the buyer do emotional labour on your behalf.
- "Just floating this back to the top of your inbox" - contains nothing. It is a reminder that you want something. Fails the test.
- "Following up on my email below" - the content is literally elsewhere. Fails the test.
- "You asked what happens to the recordings if a manager leaves. Answer: the record stays with the site, not the person. Here is the two-line version for Jonas." - stands alone, answers a debt you owed, and gives them something to forward. Passes.
Notice that passing the test costs you something real: you have to have something to say. That is why touch two is designed on the call, not three days later. Leave the call with one question you deliberately did not answer in the room, and you have written touch two before you get to the car park.
The proof point that earns touch three at Northwind. Pacific Freight's Tuas ramp took action items closed within a week from 41% to 88% in one quarter. That is one customer, one number, one time period, and it is comparable: a ramp, a new site, an ops team. Sent on day 10 with a single line - "closest thing we have to your Jurong situation" - it stands entirely on its own, and it is the only touch in the cadence that changes what Priya thinks the thing is worth.
Self-studyThe polite break-up that is not a manipulation tactic3 min read▶
Touch four is the one reps get wrong in both directions. Some never send it and let deals rot in the pipeline for two quarters, which is a b9 problem in the making. Others send the version that has been passed around sales floors for a decade: "Since I have not heard back, I will assume this is no longer a priority and close your file." That is a threat dressed as courtesy, and buyers who have received it forty times read it exactly as what it is.
- What an honest close-out contains - a plain statement that you are stopping, a one-line summary of what you understood their situation to be, and an open door with no expiry attached to it.
- What it never contains - a fake deadline, an invented price increase, a guilt clause, or a question engineered to be answered with a single letter because a reply of any kind restarts your sequence.
- Why the honest version outperforms - the manipulative one gets replies from people who feel cornered, and cornered people do not buy. The honest one gets replies from people whose timing was genuinely wrong, which is the population you actually want.
And the compliance rail from b3 sits underneath all of it: if anyone in this cadence opts out, the suppression is permanent and it applies across every tool you own. Marcus Hale at Continental Cold Chain is still a zero, and he is a zero in the cadence too.
Self-studyWhat to do when procurement enters3 min read▶
Procurement arriving is good news wearing an intimidating hat: nobody sends a deal to procurement that they have decided not to do. But the conversation changes shape completely, and reps who keep selling get squeezed.
- Their job is not to like you. It is to reduce risk and cost against a comparable. Asking them to feel the value is asking them to do someone else's job.
- Get the comparable named. "What are you comparing this to?" is the highest-value question available to you, and it is usually answered.
- Concessions go through your desk, always. This is the same rail as the proposal: a discount is a commitment, and commitments do not get drafted by a tool, an assistant, or a helpful reply written at 11pm.
- Trade, never give. Every movement on price buys something: term length, a case study, a reference call, a faster signature date. A discount with nothing on the other side teaches them the first number was fiction.
- Bring Mira back in. The economic buyer set the value; procurement sets the terms. If value is being relitigated in a terms conversation, the wrong person is in the room.
Self-study"We need to think about it" is not an objection2 min read▶
b6 gave you the objection handling. This one is different, because it is not an objection at all - it is the absence of information, and treating it as an objection makes it worse. Somebody in that building has a concern with a name, and you are not being told what it is.
- Ask what specifically. "Totally fair. What is the part you want to think about most?" It is a real question and it usually gets a real answer.
- Ask who else needs to think about it. This is where you learn Jonas has not seen the security page and Priya has not seen a number she believes.
- Ask what would make it a yes. If nothing would, you have learned something worth more than the deal - go to b9 and mark it honestly instead of forecasting it.
- Never fill the silence with a discount. Price is the loudest thing you can change, so changing it first tells them the real problem was price even when it was not.
Build the Northwind proposal and its cadence ★ 22 min · everyone builds
Open your call recap from b5 and a blank document beside it. We are building the one-pager for Mira Chen with every block tagged, then the four touches. The rule for the next twenty minutes: you may not type a sentence you cannot point to a source for. Where there is no source, write [NEEDS HUMAN] and keep going. The tags stay in the working file and come out only at the end.
Paste the b5 recap and the agreed scope into your working file. Nothing else. No product page, no boilerplate, no last quarter's proposal.
Run the good prompt below. Read what comes back looking only for the [NEEDS HUMAN] tags. Count them. Three or fewer means discovery went well; six or more means b4 and b5 are your rewatch.
Write the price block yourself, by hand, including the one sentence explaining what the number is anchored to. Nobody else touches this block.
Write the exclusions block. What are you explicitly not doing at Jurong in phase one? This is the block that survives the contract review intact.
Now the cadence. Touch 2 is the question you owed Jonas from the call. Write it as if Mira never received the proposal at all.
Touch 3 is Pacific Freight: Tuas ramp, action items closed within a week from 41% to 88% in one quarter. Two sentences maximum.
Touch 4 is the honest close-out. Read it back and delete anything that would only be there to make her feel bad about not replying.
Final pass: search the document for every [NEEDS HUMAN]. Each one is either something you answer from your notes, or a question for your next call. None of them get deleted quietly.
You are assembling a one-page proposal. You may ONLY use text from the two sources I paste below: (1) the call recap, (2) the agreed scope note. Sources: [PASTE RECAP] [PASTE AGREED SCOPE] Fill these blocks, each 2-4 lines: 1. What you told us 2. What we would do 3. What good looks like 4. What it costs 5. What happens next 6. What we are not doing Hard rules: - Do not add any fact, number, date, guarantee, security claim, service level, discount or timeline that is not in the sources. - If a block cannot be filled from the sources, output exactly [NEEDS HUMAN] and one line naming what is missing. - Block 4 is always [NEEDS HUMAN]. Never write a price. - Quote their wording where possible. Do not improve their phrasing. - At the end, list every [NEEDS HUMAN] as an open question.
Write a compelling proposal for Northwind Logistics for our AI meeting intelligence platform. Include pricing, an ROI case, implementation timeline and security details. Make it persuasive.
The second prompt has no sources, so every one of those four sections is generated from nothing. Pricing becomes a plausible number. The ROI case becomes an invented multiple. The timeline becomes a commitment your delivery team has never heard. The security details become a claim your company may not be able to substantiate. All four are fluent, all four read well, and all four are now in writing in Priya Raman's inbox.
Try it yourself - this week ◐ 40 min total
- Take the last proposal you actually sent. Rebuild it with every block tagged: AI assembles, you write, or never generated. Be honest about which blocks in the original had no agreed source behind them.
- Count the sentences in that proposal that you could not point to a call note for. That number is your real exposure, and it is usually higher than people expect.
- Write a four-touch cadence for one live deal where every single touch is independently sendable. Test each one by reading it as if the recipient has never heard of you.
- Write your honest close-out template once, properly, and never write it again in a hurry. Delete every fake deadline and every guilt clause from the version you are using now.
- Add the [NEEDS HUMAN] rule to whichever tool drafts your documents. Then run last month's proposals through it and see what it flags retrospectively.
- Ask your legal or security contact for the exact approved wording on your two most-asked security questions. Paste it somewhere your prompts can quote from, so nothing ever has to be generated.
Official sources covered
This session teaches the working core of the follow-up cadence evidence plus the trust-line constraint established in b1. Benchmarks are directional - vendor samples skew to their own users - so treat them as the shape of the curve and measure your own. Contract, security and pricing specifics stay with your legal, security and finance teams. This page covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · Which proposal block may AI legitimately put together for you?
Assembling is rearranging material that already exists; generating is inventing what is missing. The recap exists in your notes. A timeline the buyer never agreed to and a price nobody defended are both commitments, and commitments are written by a person who is answerable for them.
2 · Your proposal generator returns a confident sentence about your security posture that you have never seen before. Why is this the worst of the failure modes?
A wrong recap reads as carelessness and is fixable in an email. A wrong security or service-level claim is legal exposure that arrives in writing with your name on it. The correct output for that block is the exact approved wording, quoted, or [NEEDS HUMAN].
3 · What makes a follow-up a real touch rather than a bump?
A 3-7-7 style cadence captures the large majority of replies by around day 10, but only if each touch carries something: an answer you owed, a comparable example, a new dated trigger. "Floating this back to the top of your inbox" contains nothing except your own want.