You Don't Have a Forecast Problem. You Have Qualification Theater.

By Lolita Trachtengerts, VP GTM Ops & Growth, Spotlight.ai

Your qualification fields are full. Here is why that has almost nothing to do with whether your deals are qualified.

Here's a test you can run before your next pipeline review.

Open a deal that closed lost last quarter. Go back sixty days, to before anyone knew it was dying. Look at the qualification fields.

They said yes. Economic buyer, identified. Pain, identified. Champion, confirmed. Decision criteria, captured. Every box green, right up until the deal went dark.

That opportunity was never qualified. It was populated.

There is a difference, and most of the AI being pointed at revenue teams right now cannot tell you what it is.

The Build Everyone Is Doing Right Now

A prospect described their setup to me recently, and I have heard some version of it a dozen times since.

They run Gong. They take the transcript, hand it to ChatGPT, have it work out the MEDDPICC answers, and push the result into Salesforce. Their question, more or less: there's a step in between, slightly more manual than I'd like, but aside from the auto-update, what else is there?

It's a fair question. And the build is genuinely clever. Cheap, fast, running this quarter, no procurement cycle, no vendor.

Here's what I notice about every version of it I've seen. Nobody removes the human.

Another prospect, same month, told me they wanted human intervention before anything reached Salesforce. Garbage in, garbage out, they said.

Both of them built an automation and then refused to let it run unsupervised. That isn't a workflow detail. That's two experienced operators telling you they don't trust the output, and being right.

How Did We Get Here?

It starts sensibly, the way these things do.

Your CRM data is bad. Everyone knows it. Reps fill seven of ninety-three fields and do it at 5:58 on a Friday. Managers inspect deals by reading what the rep typed about the rep's own deal, which is about as useful as it sounds.

Then AI arrives and can read a transcript. So you wire the transcript to the field.

Great. Smart move.

And it works, in the sense that the field is now full. Adoption goes up. Hygiene dashboards go green. The MEDDPICC completion rate in the QBR looks better than it has ever looked.

Then you lose the deal anyway, and nobody can tell you which part of the qualification was real.

Here's What That Actually Costs You

Evidence and opinion end up in the same field. The buyer said their renewal cycle starts in March. The rep believes the buyer has budget. One of those is a fact and one is a hope, and once both are sitting in the same picklist they look identical. Your forecast is now built on a column that can't distinguish between what was heard and what was assumed.

One call becomes the whole truth. A model reading a single transcript is answering from one moment. It has no idea whether the champion said the same thing three weeks ago, whether the economic buyer has shown up to a single meeting since, or whether the metric the buyer quoted in discovery was ever mentioned again. One call is a claim. A claim that holds up across meetings, emails and who actually attends is evidence. A one-shot read of one transcript structurally cannot tell those apart, because it only ever sees one.

Nothing re-checks the answer later. This is the one that gets me. The model writes “economic buyer: identified” in March, and that answer sits there through Q2, through a reorg on the buyer's side, through the champion going quiet. It was true once. Nobody asks it again.

Your reps aren't the problem here. Neither is the model, really. The problem is the question.

The Real Problem Is the Question

When you hand a conversation to a general-purpose model and ask it which MEDDPICC box this belongs in, you're asking an open-ended question of a system built to produce readable text for a human.

It will always produce an answer. That's what it's for. It has no mechanism for returning nothing, so when the evidence is thin, it gives you a plausible sentence instead of a gap.

That's where the hallucinated field comes from. Not from a bad model. From an open question.

It's also where the cost comes from. Every time someone asks a follow-up, the whole deal gets read again, and you pay for the reading and for every word written back. Survivable in a pilot with four deals. Across four hundred open opportunities it becomes a line item somebody starts asking about.

We built Spotlight the other way around, and it comes down to one design decision.

We never ask an open question.

Every question our engine asks has a closed set of possible answers, defined before a model is ever called. Is this true, yes or no, with a probability attached. Which of these contacts is the champion, choosing from this list. How much influence does this person have, on a scale we defined. The model cannot return anything outside that set, so there's nowhere for an invention to go.

Then the conclusions get assembled from those answers in code, rather than written by a model. Qualification status, deal score, gaps, risks, the forecast. Computed, not generated.

What Changes When the Question Is Closed

The answer stops moving. Ask the same thing of the same deal on Monday and on Friday and you get the same answer, because it was computed from stored evidence rather than regenerated on the spot. When the score does change, something in the deal changed. Sales teams will forgive a lot, but they won't forgive a number that moves because somebody re-ran it.

Missing data shows up as missing. A question the evidence can't answer comes back with low confidence and lands on the deal as a visible gap. Knowing what you don't know is worth more to a forecast than being right slightly more often with no warning attached.

The bill stops growing with curiosity. At the decision layer nothing is being written, so there's nothing to bill for. Your team can ask the pipeline a thousand questions this quarter without anyone doing arithmetic about whether it was worth it.

The Questions That Should Keep You Up At Night

If you're in RevOps: can you look at any qualification field in your CRM and tell whether it came from something the buyer said, something the rep assumed, or something a model inferred? If all three look the same in your schema, your data quality project is measuring the wrong thing.

If you're a sales leader: when a deal slips, can you find the moment the qualification stopped being true? Or does the record just show green until the day it shows closed lost?

If you've been asked to build this in-house: you can, and it will demo beautifully on one deal, because one deal is where an open-ended prompt still works. Before you scale it, ask what happens on the four hundredth. Ask who writes the questions. Ask what the thing does when it doesn't know.

A field being populated doesn't mean the deal is qualified. It doesn't mean it's progressing. And it doesn't mean risk has been reduced.

We spent a decade getting reps to fill in the boxes. Then we automated the filling in.

Nobody automated the proving.

To learn more about how Spotlight.ai qualifies deals from evidence rather than assertion, visit Spotlight.ai.