When I first set up my CRM, I created five pipeline stages in an afternoon and assigned each one a probability of closing. Every proposal I sent moved to a stage I’d told the CRM to score at 80% to close.
On paper, a monster month was always four weeks away.
But the month never came. The proposals weren’t closing, and they weren’t dying, either. Clearly I was missing something.
So I had AI reverse-engineer my stages from the deal records themselves. Mine aren’t normal deal records. Every call transcript, email, and text on a deal is attached to it, so the AI could read the whole history of each deal, the fields, and everything that happened around them.
I told it to ignore the five labels I’d created on day one and work only from what buyers actually said and did between the first call and the close.
It came back with ten stages. My deals moved through twice as many states as my pipeline could see, and the “80% to close” proposals were concentrated in one I’d never named, where the buyer had heard the pitch but hadn’t picked a place to start. The evidence weighted that new stage at 10%.
I rebuilt my pipeline on the ten real stages, weighted each one from my own history, and changed the rules for what moves a deal. My forecast became a weighted number that holds up month to month. Deals drop out, new ones come in, and the projection moves with the evidence.
I call this a pipeline teardown. Here is the framework to rebuild your stages from evidence, weight each one so your forecast is accurate, and find where your deals are sitting. Plus, how to run it in your industry.
The framework
A pipeline teardown takes four steps:
Export: Take every open deal your forecast is currently counting, plus your last 20 closed deals, won and lost, with their full stage history and every conversation attached.
Trace: Have AI reconstruct each deal’s actual journey from the record, ignoring the stage labels you gave it, so every deal gets a week-by-week account of what state it was really in.
Map: Have AI pattern-match over the traced journeys by asking simple questions, like which stages actually exist, what close percentage each deserves based on history, and what event marks a deal entering each one.
Implement: Install the evidence-based stages with their triggers and weights, set the rules for what moves a deal, then reconcile every open deal to where the evidence says it actually sits.
Let’s take a look at exactly how I did each step below:
The walkthrough
Step 1: Export
Pull every open deal your forecast is counting on, plus your last 20 closed deals, won and lost, with their complete stage history and activity dates. The open deals are where the forecast is likely wrong. The closed deals show how deals really move.
The export needs two layers.
The first is the field data: stages, dates, and amounts, exactly as the CRM recorded them. The second is everything that happened around those fields: the call transcripts, email threads, texts, and notes on each deal. The teardown works by comparing the two, so a deal with one layer missing can’t tell you where the CRM was wrong. A stage name with no conversation behind it is just a label.
My stage history and every buying conversation already live on the same deal record in the OutcomeCatalyst platform, so this step was done before I started.
Step 2: Trace
Before pattern-hunting, have the AI reconstruct the actual journey of each deal. You can simply ask it: “Ignore my pipeline stages. Trace the buyer journey of each deal and walk me through what happened from first call to close, stall, or loss.”
This trace will help surface blind spots and hidden data in your sales process.
Mine showed proposals going out, then week after week where the buyer was still deciding which AI workflow to start with, what data it needed, and who would own it. HubSpot showed that entire stretch as one state, proposal sent, at 80% to close, from the first week to the last.
Step 3: Map
Now put questions to the traced journeys and pattern-match across them. Ask simple questions and follow your curiosity until you have all open questions answered. Mine, for example:
The real map: “Across these journeys, what stages do my deals actually move through?”
The weights: “Using my closed deals, won and lost, assign each real stage a percentage likelihood of closing.”
The gap: “Which of my five stages never actually appear in the journey, and which real stages am I not tracking?”
The parking lot: “Where did my open deals spend the most time, and what was the buyer doing during it?”
The triggers: “For each real stage, what event marks a deal entering and exiting it?” The stage boundaries were already sitting in the traces in the form of a named workflow on a transcript, a delivered prototype, a redlined contract.
Mine came back with ten stages, each with a trigger and a weight, all of it verified against the traced journey of every deal.
The proposals showing 80% to close were concentrated in a stage I’d never named. The buyer had heard the pitch and hadn’t picked a workflow to start with. I now call that stage Discovery Held, and the evidence weights it at 10%. Once we align on a workflow we can start with, the deal moves to Scoping in Progress at 35%.
80% → 10%
What my CRM scored a sent proposal, versus what the evidence said it was worth.
My old CRM had no field for either state, so it scored every one of those deals at 80% from the day the proposal went out, and kept scoring them that way for as long as they sat.
Step 4: Implement
Install the new stages in your CRM with their triggers and weighted percentages, so the forecast is computed from your verified data instead of typed in from optimism.
Then change the rules. A deal advances when its trigger shows up on the record, and only then. I used to drag deals to “proposal sent” and type in an end-of-month close date. Now my AI sets both. After every conversation on a deal, whether it’s a two-minute call, a text, or an email exchange, it re-reads the record and adjusts the stage and the close date if the evidence moved. My hands never touch either field.
Then do a one-time reconciliation. I had AI re-trace every open deal against the new stages and place each one where the evidence said it was. That’s the moment the pipeline told the truth, and most of my 80% proposals moved backward to 10-35%. My forecast for the month dropped that afternoon, and for the first time it was a number I believed.
The new pipeline changed how I sell, too. Aligning on a workflow is now the trigger that moves a deal out of Discovery Held, and the prototype I build after that conversation is the trigger that moves it toward a proposal. A deal can’t reach the proposal stage until the buyer has seen their workflow running, so the state my old pipeline couldn’t see now has a name, a weight, and a way out.
The OutcomeCatalyst platform is what makes the rules hold. Every conversation lands on the deal record as it happens, the stage and close date get set from it, and the forecast re-weights the same day.
The outcome
My forecast is a weighted number computed from my own history, so the pipeline total is a projection I can defend month to month.
Every conversation re-weights the deal it belongs to, so the forecast is only ever as old as my last call, text, or email. Nothing in my pipeline gets set by hand.
Every open deal sits in its true state, which tells me exactly what has to happen to move it.
■ The industry audit
A pipeline teardown is a check on your system of record. I ran mine on my sales stages because that’s where my forecast lived. Wherever your system reports a state your money depends on, the same check applies:
If you’re in commercial real estate: Take the last ten OMs you’re actively working and check the date each one entered its current stage against how long your last three closed deals spent there. Count the ones at double that age. Each one is sitting in a stage your pipeline has no name for, and your forecast is counting it anyway.
If you’re in industrials: Pull your ten oldest open quotes and compare each stage to the last real buyer activity on it. Count the ones marked “pending” with 30+ days of silence, then check what the buyer was actually waiting on. Odds are it’s a spec question nobody logged, and the quote is sitting in stages you don’t track.
If you’re in healthcare: Take your last ten new patient intakes still sitting between “received” and “scheduled” and check how long each has waited against how long your last ten took to actually get scheduled. Count the ones past double that age. Each is parked in a stage your system doesn’t track, and it’s where the revenue leaks.
Same teardown, same business impact, different industry.
Where it breaks
Updating stages from just a few deals can overfit. A pattern that held across past history might be an artifact of how you sold last year, so run the updated stages against a month of new deals before you retrain a team or rebuild dashboards around them.
And the forecast weights also break down in stages few deals pass through. If only four deals have ever passed through a stage, its percentage isn’t reliable yet, so treat those stages as provisional until more deals move through them.
That’s why I run all of my sales data through the OutcomeCatalyst platform, which has complete context of my business and helps spot real patterns and signals instead of unreliable noise.
Zach
Founder & CEO, OutcomeCatalyst
Book 30 minutes with me →
