When we launched the new OutcomeCatalyst platform, my first calls went great. Buyers leaned in and sounded ready for next steps. Then every thread went quiet. Nobody said no, but nobody replied, either. My pipeline was full and none of it was moving.
I assumed the usual suspects: budget, timing, a competitor I hadn’t heard about.
But instead of guessing, I asked my data. I had AI compare my stalled deals to my closed ones and reason over the differences between the conversations that died and the ones that actually went somewhere.
What came back was a stark insight: the deals were stalling on my pitch.
Buyers were leaving impressed but with nowhere to actually start. The conversations that closed ended with a tangible AI workflow tied to a specific business impact. The conversations that stalled sold the grand vision of OutcomeCatalyst but gave nothing concrete.
That insight changed the way I run my sales process. I refused to end a first call without identifying a workflow that we could start with, and if they couldn’t choose one, I knew they weren’t ready to work with us yet.
Conversions went up, the sales cycle shrank, and leads started giving me definitive answers instead of leaving me waiting on a no.
I call this a deal autopsy. Here is the framework to run the same autopsy on your own pipeline to improve your pitch, increase conversions, and shorten timelines. Plus, how to tailor it to each industry.
The framework
A deal autopsy takes four steps:
Pull: Take your last 20 to 40 deal records from your CRM. Get a variety of closed-won, closed-lost, and still open.
Enrich: Attach every call transcript and email thread to its deal, then record a voice note on each one with the signals and context that never made it into the conversations.
Reason: Connect Claude or ChatGPT to that data and ask simple questions like, “what was landing in my pitches on the deals that closed versus the ones that stalled or lost?” Keep going back-and-forth on how each call went.
Improve: Codify the insights into your new sales strategy: how you open, what you show, and what you send after.
Let’s take a look at exactly how I did each step below:
The walkthrough
Step 1: Pull
Take your last 20 to 40 deal records straight from the CRM, with a variety of closed-won, closed-lost, and still open. Don’t sort them into piles yourself. Half the point of putting AI on this data is that it does the sorting for you, so pull the records and let it work out which conversations moved and which went quiet.
The one requirement is that a real sales conversation happened on each deal. A contact you emailed twice gives the AI nothing to work with.
Make sure you’re recording your sales calls and writing notes within the hour after meetings. A conversation that was never captured is invisible to this analysis.
Because I’m a customer of my own platform, I already have all this data centralized and tagged in my company brain. I simply go to Claude and reason over my entire company context.
Step 2: Enrich
Attach the full record of each buying conversation to its deal:
Call recordings and transcripts: Every discovery and demo call, plus the calls after.
Every email thread on the deal: Yours and anyone else who touched the buyer.
Notes from in-person meetings: Record them with Granola or any transcription tool, or type notes immediately after the meeting.
A voice note from you on the deal: The signals you picked up along the way and the context that never made it into the conversations.
The voice note is data that most people skip, and yet it can carry the most weight. AI doesn’t understand interpersonal relationships like you do. It will tell me a deal is riskier than it is, or that I should handle a buyer a certain way, and I have to explain the relationship before it corrects itself. Record a few minutes on each deal so the analysis starts with that context instead of waiting for you to supply it.
The enrichment work for me was the voice notes. I recorded one per deal and gave it to OutcomeCatalyst to attach to the right record.
Step 3: Reason
Treat the AI like a sharp revenue operations analyst and have it reason over your data. Give it context, encourage back-and-forth, and don’t expect every answer from a single question. The useful insights usually show up three or four questions in.
Here’s how mine went, in Claude, connected to my deals:
Start wide: “What was different about the deals that closed versus the ones that stalled or lost?” Budget, timing, and competitors weren’t the issue.
Then zero in on the pitch: “What was landing in my pitches on the deals that closed versus the ones that stalled or lost?” This is the question that led me to my answer.
Then dig into the closed-won: “On the deals that closed, what did the buyer ask for in their own words?” They all named a specific workflow.
Then look at the stalled and closed-lost: “On these deals, what was the last thing said on the call?” Nothing but enthusiasm and no tangible next step.
Then break it out: “Does that split by industry or company size?” That helped me understand the best industry-specific workflows.
Keep going until the answers stop surprising you.
My own answer was blunt. The calls that closed ended on one specific AI workflow with a business impact attached that I could map to a purpose-built prototype. Those deals closed in one to three calls.
1–3 calls
How long deals took when the first conversation ended on one named workflow with a business impact attached.
The calls that stalled ended on everything we could theoretically do. I would present the platform in the abstract and boast about how versatile it was, which left buyers with analysis paralysis. They were expecting me to tell them where to start, and I handed them a vision.
All my sales and customer data is centralized and organized via the OutcomeCatalyst platform, so all I do is consistently reason over it.
Step 4: Improve
Insights only matter if they improve how you sell. Take what came out of the questions and codify it into your sales strategy: how you open, what you show on a demo, and what you send after.
I made three changes off my own analysis:
First, I come to every call with AI workflows mapped by industry and business type. I open with the platform to set the stage, then zero in on the top three workflows for that prospect, each with its business impact.
Second, after the call I send a purpose-built prototype of the workflow they cared about most, so they can see what working together looks like.
Third, I re-engineered my entire sales funnel. My stages now match how deals actually move, and I used AI to assign each one its own probability of closing. That changed how I track my cycles and motions, giving me a realistic state for every deal, and letting me analyze where deals get stuck so I can keep improving the process.
The outcome
Buyers who left with a named workflow booked the next step. Buyers who left with a vision went quiet.
The new pitch shape: platform to set the stage, three workflows for a starting point, a purpose-built prototype after the call.
A re-engineered sales funnel that changed how I track my cycles and move through my motions. Every deal now shows a realistic state, and I can analyze where deals get stuck to keep improving the process.
■ The industry audit
A deal autopsy is pattern matching. I ran this on my sales conversations because that was my bottleneck. The same analysis works wherever your bottleneck is:
If you're in commercial real estate: Pull the last offering memo you passed on. Count every source someone opened to make that call: transaction history, ownership, comps, market data, and your own past deals. If it was more than two places and took more than a day, you're disqualifying deals slower than your competition is closing them.
If you're in industrials: Pull your last ten quotes and ask who priced them and how. If the answer is one estimator matching against jobs they remember, write down what fails the day that person leaves. That list is the deal flow you can't manage without them.
If you're in healthcare: Look at how your last ten intakes got routed and ask what history informed each call. If it lived in one person's head or across systems that don't talk, count how many you'd route differently if you could see the full picture. That gap is revenue or risk leaking out the door.
Same autopsy, same pattern matching, different use case.
Where it breaks
The autopsy tells you what’s correlated with closing or stalling, not exactly what caused it. Buyers who left with a workflow may simply be further along in their own process. Worth validating before you rebuild your whole pitch around it.
It also only knows what was said out loud. A buyer who goes quiet almost never tells you why on the call, so the record shows you the stall without the reason behind it. That’s what the voice notes are for, and they are only as good as what you remember about the deal.
You can avoid these issues by ensuring that as much conversation data as possible makes it onto the record, which is why I use the OutcomeCatalyst platform to automatically structure and add my conversation data to the right HubSpot record.
Zach
Founder & CEO, OutcomeCatalyst
Book 30 minutes with me →
