I need to grade my own sales chats and find the leak

Ch. 01 What it is

You close some chats and lose others and you're not sure why. The seller in the moment can't see their own misses — they were inside the conversation, reading one message at a time. The fix is to grade the whole thread after the fact against a fixed checklist: where did the diagnosis skip, where was the gap left flat, where did the ask never land. Three ways to run that review, from a checklist you apply by hand to an agent that scores every chat.

Ch. 02 The three ways to build it

Simplest path first. Every tier carries its real setup time and its honest trade-off — the cost is the part most write-ups leave out.

1
Tier 1 · simplest path
A rubric you self-apply
Setup~30 min
- a one-page checklist
- your own chat exports
Paste a finished sales chat into a doc and walk it against a short, fixed checklist — the same handful of things a good close always does. Did you develop the real problem before you pitched, or lead with the offer? Did you make the cost of doing nothing actually felt, or just name it? Did you surface the real objection or argue with the stated one? Was there a clean ask at the end, or did the thread trail off into 'let me know'? Mark each one strong / needs work / missed, pull the exact line where it slipped, and write the line you'd send instead. Two of these are deal-breakers — no diagnosis means you sprayed a pitch; no ask means there was never a sale to lose — and a miss on either is the whole story. Run five lost chats this way and the leak stops being a mystery; it's almost always the same one or two boxes, every time.
The trade-off
You're grading your own work, so you'll grade it generously — the move that felt smart in the moment reads as smart on the page. Fix it by reading adversarially: your job in the review is to find the miss, not to confirm the win. Read the buyer's replies, not your own lines — a stall, a one-word answer, a question you talked past is where the truth is. And grade a batch, not a single chat; one conversation is an anecdote, five is a pattern you can actually act on.
2
Tier 2
A deterministic checker
Setup~half day
- a short script
- your chat exports
Some of the review is objective and doesn't need judgement at all — it can be checked by a script that reads the transcript and never guesses. Was a real next-step or buy-link actually sent, or did the thread end on talk? Is there an explicit ask anywhere, or only rapport? Did the conversation move through the full shape — open, diagnose, offer, close — or skip a phase? And the redline that matters most for a no-call motion: did you slip a 'hop on a quick call' or a calendar link in, when the whole promise is that there isn't one? The checker scrubs personal details out first, then flags each of these in seconds, so by the time you (or a reviewer) read the chat you're confirming a pre-marked transcript instead of reading cold. It's the cheap, fast, never-tired layer under the judgement call.
The trade-off
It catches what's present or absent, never how well you sold — a script can confirm an ask exists, not that it was the right ask at the right moment. So it's a floor, not a verdict: a chat can pass every deterministic check and still be a weak conversation. Keep its job narrow and honest — objective facts only — and let the quality call live one tier up. The value is that it removes the dumb misses (the absent link, the missing close, the accidental call-booking) before anyone spends attention on the hard part.
3
Tier 3
A grounded grader that scores every chat
Setup~1–2 days
- an LLM + a written rubric
- your sales canon
- a sampled human review
Hand the judgement call to a grader that holds what you can't hold mid-conversation: the full transcript read after the fact, a written rubric tied to real sales mechanics, and a brief to find the misses rather than confirm the win. That last part is the whole trick — a grader told to be agreeable will tell you every chat was fine. Set up right, it returns a verdict (clean / fixable / weak), the specific line where each criterion slipped, the named reason it's a miss, and a rewrite of the weak moments. It runs the deterministic checks first, then grades the conversation, so every chat you've ever had can be scored without you reading each one — and the pattern across a hundred chats surfaces the leak the eye would never catch. It coaches; it never touches a live conversation.
The trade-off
It grounds its grade in a real rubric, but it's only proven once you've checked its verdicts against your own on real chats — until then it's a sharp first draft of a review, not a trusted judge. Sample its calls, watch where it and you disagree, and only let it run unwatched on the criteria it's earned. And keep it pointed at finished conversations, not live ones: an agent that grades is a coach, an agent that intervenes mid-chat is a liability. The grade is advice. A human still decides what to do with the leak it found.

Ch. 03 The detail

You close some chats and lose others and you're not sure why. The seller in the moment can't see their own misses — they were inside the conversation, reading one message at a time. The fix is to grade the whole thread after the fact against a fixed checklist: where did the diagnosis skip, where was the gap left flat, where did the ask never land. Three ways to run that review, from a checklist you apply by hand to an agent that scores every chat.

Category: Sales · Conversation review
Format: System
Level: intermediate
Provenance: Own-packaged

salesconversation-reviewno-calldm-salesself-coaching

The miss you can’t see from inside the chat

You sold well in some conversations and badly in others, and if you’re honest you can’t fully say which was which. That’s not a discipline problem — it’s a vantage problem. The seller is inside the thread, answering one message at a time, deciding the next reply before the last one has fully landed. From in there, the conversation that drifted into “let me know” felt like patience, and the pitch you led with felt like confidence. The misses are invisible precisely because you were the one making them.

The review fixes the vantage. You read the whole finished chat at once, from outside it, against a fixed list of what a close actually does — and the things that were invisible in motion become obvious on the page. The line where you offered before you diagnosed. The objection you argued with instead of surfacing. The end that never asked for anything. None of that needed more sales talent in the moment; it needed a second pass with a checklist and an honest eye.

What the checklist actually checks

Strip a sales conversation down and the same few things decide it. Did you develop the real problem before reaching for the offer — or lead with features into a problem you assumed? Did you make the cost of staying stuck felt, or just mention it and move on? When the buyer pushed back, did you surface the real blocker under the stated one, or take “it’s the price” at face value? And at the end — was there a clean, specific ask onto a path the buyer could actually take, or a soft “happy to chat more” that closes nothing?

Two of those carry more weight than the rest. Lead with the pitch and skip the diagnosis, and you sprayed an offer at a problem you never confirmed — everything after is noise. Reach the end with no ask, and there was never a sale on the table to win or lose. A chat can do four things right and still be capped by a miss on either of those two. That’s not a scoring quirk; it’s the shape of how deals are actually lost.

Why a grader beats a re-read — and where it stops

The reason a tool can out-review you isn’t that it has better sales instincts. It’s that it holds three things you can’t hold from inside the conversation: the entire transcript at once, a written rubric that doesn’t get tired or generous, and a standing instruction to hunt for the miss instead of nodding along. A re-read by the person who ran the chat drifts toward self-justification; a grounded grader, briefed adversarially, doesn’t have that pull. That’s the asymmetry the third tier is built on — and the reason a grader told to be agreeable is worthless, because it’ll pass everything.

But the grade is the start of the work, not the end of it. The deterministic layer is honest about its ceiling — it confirms an ask exists, not that it was the right one. The grader is honest about its own — it’s a draft of a review until you’ve checked it against your own judgement on real chats. And neither ever touches a live conversation: a tool that grades a finished chat is a coach, a tool that steps into a running one is a risk you don’t need. Run the review on what’s already done, find the one or two boxes you keep missing, and fix those — that’s where the leak closes.