How does AI help sellers win, not just send more?
Point AI at what your seller knows before a first meeting and what every draft says, and keep a named seller on every send. Then test it on stage conversion before win rate, because win rate reads late.
Sending more won't fix any of it
Product marketing approved a new competitive play two weeks ago, and one rep has pasted it out of Slack. A rep spent Tuesday morning researching an account before a first call, and the research lives in her notes. The SDR tool sends more than it did last year, and win rate sits where it sat.
AI can't win the deal. It can change what your seller knows walking in and what your seller drafts from.
Which accounts earn that work is decided in Insights & Targeting. This page starts where the ranking ends.
Give every seller the account's own facts before a first meeting
Your sellers need one page per account, built from what the account has said in public and to you, with a source under every line. Keep it where the whole team can read it, not in one rep's notes. Check any claim about someone's current title within a week of anything going out.
| Line | Where it comes from | Check it again |
|---|---|---|
| What they said they're betting on | Their earnings calls, letters, keynotes | When they publish again |
| Who they're hiring, and for what | Their open job posts | Before each meeting |
| Who you're meeting, and their current title | Their public profile | Within seven days of anything you send |
| What they told you | Your recorded calls, found by speaker, with a link to the moment | After every call |
We expect this to pay where deals are large, buyers are named and first meetings are scarce, and not where a seller has two minutes per account. If it helps, you'll see it first in the share of first meetings that lead to a second.
As of
Ours is written for the buyer to read. Before a first meeting we build the account a page from what its leaders have said in public, with a source under every line and a place to comment on each section, so they can tell us what we got wrong. We check titles because our own notes once called someone the CEO of a company they had left five months earlier, and a one-pager built on that note nearly said so in print. The check covers a claim about someone’s current title, not every fact about them, which keeps it cheap enough to run on everything.
Put this week's approved play where the seller's AI drafts
Today the play lives in an enablement deck and a Slack post, and the AI your sellers draft with reads neither. Give approved plays one home the AI can read, and give each play a record like this one.
| Field | What it answers | Who sets it |
|---|---|---|
| The play | What the seller says, and the proof behind it | Product marketing |
| Fits | Which accounts, and which deal stages | Product marketing, with sales leadership |
| Version and date | Which wording a draft used | The owner, when approving it |
| Owner | Who answers a seller's question about it | Product marketing |
| Status | Live, or pulled and when | The owner |
Then give the AI its instructions. A connection alone leaves it guessing among every action the tool offers, and Shared Context shows what that costs.
- Before drafting, find the live plays that fit this account and this stage.
- Show the play and its version beside the draft, and keep that record with the draft and the send.
- If no play fits, say so.
- If the home can't be reached, say that too. Never fall back on an old copy.
As of
We run part of this today. Every brief we write produces a packet beside it that carries only the proof points cleared to travel. A number we haven't cleared isn't in the packet, so the draft can't take it from there.
Pulling a play reaches forward only, as Shared Context explains. Plan for what it misses.
| Where the play is | When the owner pulls it |
|---|---|
| The home the AI reads | It's gone from what new drafts are offered |
| A seller's saved example or old conversation | It can still turn up, until the seller clears it |
| Drafts sitting unsent | Flag the ones that recorded the pulled version, if the tool keeps drafts |
| Text a seller copied into an email or a deck | No record follows it. Only a check before send catches it |
| Email already sent | Nothing recalls it. Your send records show which accounts got it; ask sellers about the rest |
Keep a named seller on every send
Let the system research, assemble and draft. The seller reads the draft, cuts what doesn't fit, and can drop a play with a reason. The reason goes to product marketing, who decide whether the play changes or goes. We'd have them read those reasons every week.
The buyer reads every email as the seller's word. The bottom rows of that table are why a person stays on the send.
The steps can run without a person. The send can't.
That's what a system of action means here: the steps run on their own, and a named person owns what reaches a customer. Our drafting skill stops at a review link, and our founder sends.
Count the drafts that carry your latest play
You can run this count with no new tools.
- Pick the play product marketing approved most recently, and list the target accounts and stages it fits.
- Pull the drafts reps wrote on those accounts since it was approved, with the deal stage each was written at.
- Ask each rep to mark each draft one way: used it where it fit, used it where it didn't fit, left it out on purpose (and why), knew about it but missed it, or didn't know it existed.
- Where your records have the dates, note the weeks between approval and the first draft that used it.
Count drafts, not reps, and only drafts at an account and stage the play fits. Keep uses where it didn't fit as their own number. If most say they didn't know it existed, fix how plays reach the draft before you buy anything. If most say they knew and missed it, the draft isn't putting the play in front of them. If they left it out on purpose, read their reasons before you push it harder. The count tells you whether plays arrive. Whether they win is the next test.
Assign the accounts before the quarter starts
The usual read compares accounts where sellers used a play with accounts where they didn't. Sellers chose those accounts, often for reasons that also predict the outcome, so the gap can't be credited to the play. Test the change instead: the account brief and the play in the draft, against today's workflow, with the same approved plays open to both groups.
| Step | What to do | Why |
|---|---|---|
| Pick the unit | Assign by seller when the same sellers would otherwise work both groups. Assign by account only if you can keep the briefs and plays out of the other group's hands | A seller who learns on one account carries it to the next, and the gap shrinks |
| Assign | Before the quarter, pick which accounts get the new setup and which keep today's workflow | Takes the seller's choice out of the result |
| Compare | A random split where you can; otherwise accounts matched on segment, size, stage and seller tenure | Matching leaves some bias; say so when you report |
| Count everyone | Read accounts as assigned, whether or not the seller used the setup | Dropping non-use flatters the change |
| Read first | Conversion from one stage you name to the next | It can move inside the quarter |
| Read win rate | After at least one full sales cycle of your own, the same window for both groups | Report wins, losses and deals still open; a short window favors whichever group closes faster |
| Hold steady | Account mix, seller tenure, pricing, and which plays are live | Any of them can move the result on its own |
We can't run this on ourselves. Three of us run Fresh Context, and we don't have enough pursuits to split. We keep a ledger per pursuit instead: the date the buyer did something, what they said word for word, and who said it. It can't prove lift. It keeps the record dated and in the buyer's words.
What leaders ask
- What about an SDR team that sends at volume?
- Split the messages. A sequence step written once and sent to many is a template: a named owner approves it and answers for every send. A first touch written for one named buyer gets a seller's read before it goes.
- Should the buyer see the brief?
- Decide on purpose. A brief built only from what the account has said in public can be shown to the buyer, and showing it invites corrections. A brief that holds what they told you on calls stays inside.
Draft copy, agent's words, not yet reviewed.