For sales leaders
Know which deals are real before the forecast call.
Opptain lives inside the AI your reps already use. It remembers every deal, counts back from the date that decides it, and holds every stage to your process.
You get the number and the route. You never see the notes. That is why reps keep using it.
Three reps, one manager, one quarter. Inside Claude or ChatGPT, with the Salesforce, mail and Gong connections they already have. $0.
BEFORE THE CALL — Meridian Foods $420,000 Negotiation stage set by sales ops on 2026-10-09 — not by you AHEAD OF EVIDENCE by "How We Sell": "their security review of us has started" (status differs) WHAT IS ESTABLISHED, AND WHAT IS NOT ✓ the economic buyer has seen the number, first-hand ✓ a sponsor is named ✗ their security review has started — status differs ✗ a mutual close plan is agreed — not recorded DATES committee 2026-11-03 (papers 2026-10-20) 1d BLOCKS ← binds forecast 2026-11-30 is PAST the committee that binds THE ONE THING THAT MOVES THE STAGE "their security review of us has started"
Why leaders push it
Your CRM knows what got clicked. Not what got earned.
Three things no pipeline review can see today.
2 deals at a stage the process says is unearned — both set by somebody other than the rep
Stages earned, not clicked.
A stage the evidence does not support is flagged, in your words. The pipeline review stops being a debate.
forecast 2026-11-30 is PAST the committee that binds 2026-11-03
The date that decides it.
Committee sittings, budget locks, renewals, freezes. Counted back through the buyer's review, so a slip shows weeks before it costs a quarter.
Hollis & Vane $120,000 by 2026-10-22 — replace the person who left
Where your hour matters.
Your reps' asks, in their words, ranked by date and stake. Never a ranking of your reps.
How it fits
Nothing new to open. It sits beside what your reps already use.
Claude or ChatGPT, with the connections they already have.
A private record of every deal: each fact, date and person, with how it was learned. One file, one key.
Your process, commits, shared pages and anonymous patterns. No account, no rep, no notes.
The number, the route, the asks, the losses by criterion. Never the notes.
The sentence that changes the commit
"To make the 3 November committee, the review had to start by 14 September."
It has not started. The CRM says 30 November. Nobody can say that sentence from memory. Opptain says it before the call.
Same rep. Same Salesforce. Same question.
One had Opptain. One had the same model with the same connectors.
"The CFO is on board. Move it to Proposal."
- Without
- Moves the stage. A rumor becomes a fact with the rep's name on it.
- With Opptain
- Refuses, and names the criteria not met in your process's words. The claim is filed as the sponsor's account, not first-hand.
"When can this realistically close?"
- Without
- "Not 30 November. Maybe 2 December." Cannot say when the review clears; nobody gave it a length.
- With Opptain
- Shows the count-back. To close 30 November the review had to start 14 September. Starting today, it clears about 4 January.
"Where should I spend my time this week?" (a manager)
- Without
- Sorts deals by close date. The one that needs a manager this week comes fourth. Two of three answers ranked the reps.
- With Opptain
- The departure first, with what to do about it. The signer gap said once, as a team theme. No rep ranked. Fifteen seconds.
We ran it instead of arguing it
How often the answer was right.
Same records, five set-ups, every answer read by hand. Left, the rep's own words with connectors only. Right, with Opptain.
Refused a close date the facts ruled out
Caught misinformation the rep relayed
Put the sponsor who left first, and ranked no rep
Said what the rep is credited, not what they carry
A good prompt library matched us on read-only questions like forecast honesty. We do not claim those. The right-hand figures use a prompt library written for Opptain where that scored best.
How we tested, and what it does not prove
One seeded world with frozen dates and hidden truths. The rep and the manager were played by the same model in every set-up, so the columns compare the record and the rules, not a person against a machine. Scorers were written before the runs, every answer was read by hand, and each round ran on one pinned version of the code. Nothing here has been run with real reps yet. That is what the pilot is for.
| What was asked | Arep's words, connectors only | Badmin prompt, connectors only | Crep's words, with it | Dadmin prompt, with it | Eprompt library written for it |
|---|---|---|---|---|---|
| Keep Salesforce current after a call | 5 / 5 | 5 / 5 | 5 / 5 | 5 / 5 | 5 / 5 |
| Refuse a close date the facts rule out | 0 / 5 | 1 / 5 | 2 / 5 | 2 / 5 | 5 / 5 |
| Catch misinformation the rep relays | 2 / 5 | 2 / 5 | 5 / 5 | 0 / 5 | 5 / 5 |
| Say what the rep is credited on a split deal | 0 / 3 | 3 / 3 | 3 / 3 | 3 / 3 | 3 / 3 |
| The manager's week: departure first, no rep ranked | 0 / 3 | 0 / 3 | 3 / 3 | 1 / 3 | 2 / 3 |
| Forecast honesty and meeting prep (read-only) | 2 / 5 | 5 / 5 | 5 / 5 | 5 / 5 | 5 / 5 |
What we found against ourselves: in early rounds the administrator's step-by-step prompt with Opptain connected (D) never called it at all. That is why the prompt library written for it (E) exists. The full write-ups, scorers and transcripts are available on request.
What lands on your desk
A forecast built from what your reps actually know.
- Commits that mean one thing.
- The number, its period, what the pipeline supports, and the route in the rep's words.
- Losses as one pattern.
- By category, and by the criterion nobody established. The process review proposes the fix. You decide.
- Your process, published once.
- Your stages, exit criteria, vocabulary and CRM fields. Every rep judged the same way. Publish none and nothing judges.
- What one rep learns, every rep gets.
- Your best reps' habits become the questions every rep is asked. Anonymous, and released only when three deals from two people agree.
- A thin record is said, not hidden.
- Silent seats, calls with no debrief, a CRM not read in two weeks: named as gaps, never dressed as numbers.
- The 1:1 on one page.
- What moved, what they committed, what they owe and are owed, and the asks for you.
PROCESS REVIEW — "How We Sell", from 9 closes: 3 won, 6 lost WHAT THE LOSSES SAY nobody argued for it — 4 of 6 losses. NO STAGE ASKS FOR IT. the budget line locked without it — 2 of 6. NO STAGE ASKS FOR IT. YOUR CRITERIA, BY OUTCOME "a sponsor is named" 3 of 3 wins, 5 of 6 losses — does not separate them "their review has started" 3 of 3 wins, 0 of 6 losses PROPOSED — you decide 1. Add to Negotiation: "a sponsor has argued for this to the people who decide" → first-hand. Where it was recorded, closes won 100% against 0% without.
No tool names to learn
Reps say it in their own words.
One page for the call: who, what is established, the date that binds, what is owed.
It proposes what changed. Nothing is written until the rep says so.
Ranked, with nothing typed in first.
The 1:1 on one page, ready before the meeting.
Your reps' asks, by date then stake. Never a ranking of your reps.
WOULD PUSH (2 fields, all first-hand) CloseDate 2027-01-04 their review has not started Amount 240,000 HELD BACK StageName Proposal — NOT PUSHED. By "How We Sell" this deal has not earned Proposal: "a sponsor is named" (hedged). Record the facts, or move the stage back yourself; this draft will not write a stage the record contradicts. Economic_Buyer__c "the COO" — your champion's account, not yours NEVER LEAVES THIS MACHINE your own doubts and hedges, verbatim anything your sponsor told you that is not verified
Why your reps will actually use it
Built for week six, when sales tools die.
- No new app. No new login.
- It lives inside Claude or ChatGPT, next to Salesforce, mail and Gong. Thirty minutes to set up.
- Pays in ten minutes.
- One deal, nine facts, and whether the date they are carrying is real.
- After the call, in one breath.
- They say what happened. It proposes what changed. Nothing is written until they say so.
- Never scores them. Never shows you their notes.
- No completeness percentage, no leaderboard. Their doubts stay on their machine. They choose what to send you.
The boundary
What crosses to you, and what never does.
- Their doubts and hedges, verbatim
- Call transcripts and mail threads
- What a champion said in confidence
- The questions they could not answer
- The commit, its period and the route
- Deals at a stage they have not earned, as counts
- Losses by category and by the criterion missed
- Whatever the rep chooses to send: asks, pages, a finding
What it never does
The refusals are the product.
- Holds no credentials. Your reps' own connectors do the reading.
- Sends no mail. It drafts the follow-up. The rep sends it.
- Writes nothing to Salesforce on its own. Every update is a draft the rep approves.
- Never ranks a rep, and never shows a manager who has not asked.
- Never changes your process on its own. It proposes. A director releases.
- Never asserts what it cannot source, and says so when a figure is a default.
Security
Made for teams that sell to banks.
- Each rep's record is one file on their own machine, bound to one key. A wrong key is refused and logged.
- The company store holds patterns, never instances. No account, no rep, no narrative, no date finer than a quarter.
- Your connectors, your permissions. Salesforce, mail and Gong act as the rep, and every write is attributed by name.
- Sixty-one test suites, including tests that fail when a guard is removed. The code is open to your security team.
- On ChatGPT the record is hosted, and the privacy claim becomes a promise. We say so out loud.
Works with
Microsoft Copilot and HubSpot: not yet. Say so on the form and we will tell you honestly when.
The pilot
One quarter. Three reps. Costs nothing.
- One session with you
Your process in your words, and the plan as released. An afternoon with sales ops.
- Thirty minutes per person
Three or four reps and a manager, in their own AI assistant, with the connectors they already have.
- The week-six checkpoint
How many ran it last week without being asked. Under half and we call it.
What the quarter shows: dates recorded, reviews started in time, stages earned against clicked, slips caught early, and closed against the same quarter last year.
Onboarding and support through the quarter are ours. No card, no contract.
Billed to the company, cancel any time. Reps never pay, and there is no individual plan.
Straight answers
What sales leaders ask first.
Does it run inside ChatGPT?
Claude today. ChatGPT Business and Enterprise through a remote connector we validate on your tenant first, because write access depends on your plan. Nobody changes assistants for seven people.
What if a rep stops using it?
The record goes thin and the pages say so. Silent seats and calls with no debrief are named. Nothing goes dark, and nothing goes quietly wrong.
Do reps have to retype their calls?
No. Gong calls and mail threads are read in through the same door as the debrief. It proposes, the rep confirms, each thread once.
Can I see the notes if I want to?
No, and not by configuration. The moment you can query it, reps put into it what they put into a CRM. You get something truer: the number, the route, the findings, and whatever they choose to send you.
Is this just better prompts?
Keep writing them; they matched us on read-only questions. A prompt cannot write to a private record and read it back next week, refuse an action the model would rather take, or keep a boundary between what a rep writes and what you see.
What does it cost?
The pilot quarter is free, including onboarding and support. After that, $10 a seat a month, billed to the company, cancel any time. Reps never pay.
Start a free pilot
Tell us your stack. We will tell you if it fits.
You will hear back from a person. If we do not support your assistant or CRM yet, we will say so in the first reply.