Why Our Competitors Field Has Fewer Entries Than Yours

A design decision that looks wasteful from the outside, and what it does to your win-loss analysis.

Our model checks every name on your competitor list against every call. Not the ones a rep flagged, and not the ones that surfaced in a summary. All of them, one at a time, every time.

When I describe this to engineers they tend to point out that it sounds inefficient, and from the outside they have a point. It is the opposite of how most teams wire this up. Here is the problem it solves.

The answer you have no way to check

Hand any capable model a stack of transcripts and ask who you are competing against on a deal. You will get names back, and the names will frequently be right, which is the part that makes this worth writing about rather than merely a bad idea.

What you will not get is a way to check. The model read everything, formed an impression, and returned a conclusion. It does not report how sure it was, and it does not hand back the sentence that convinced it. A vendor the buyer mentioned once, in passing, as somebody they had already ruled out two quarters ago can end up sitting in the field looking identical to the vendor you are genuinely in a bake-off against.

Your win-loss analysis then runs on that field. So does battle card assignment, and so does the forecast.

None of this is an argument against the models. We ship a deal inspection skill that runs inside Claude and an MCP server that agents can query directly, because those are good tools and reps already live in them. The failure happens further along, when an open question gets asked and the resulting prose is written into a field that people make decisions from.

What CDAM asks instead

Our Calibrated-Decision AI-Model, CDAM, never asks who the competitors are.

It walks your list. For each name on it, against each transcript, it asks whether you are competing with that particular company in this particular opportunity, what the probability is, and where the evidence sits.

A name gets marked as an inspected competitor only when the probability is high and there is a specific, openable piece of evidence behind it. Anything less than that stays unmarked.

Which is why the checking looks wasteful and isn't. A large number of narrow questions with defined answers turns out to be more reliable than one broad question with an open one, and far easier to audit, because every answer points back at the words that produced it.

The names that aren't on your list

Buyers bring up vendors nobody has been tracking. Most systems either lose that entirely or squeeze it into a category where it does not belong.

CDAM puts those into a separate bucket, labeled other competitors, and keeps the real name in the evidence instead of forcing it into a picklist with no row for it.

Then it goes back to them. On a regular cycle it reads through what has collected in that bucket looking for two things: names that keep recurring, which usually means a competitor that has earned a place on the real list, and names already on the list that have stopped appearing in any deal, which usually means you can retire them.

Most competitor lists I have looked at were last edited by somebody who has since changed jobs.

Why this matters for more than one field

The field was never really the point.

Once competitor mentions are structured and inspected, the same treatment applies to the pain the buyer described, the objections raised, who was actually in the room, and the evidence behind each stage. At that point you can ask questions that were previously unanswerable, not because nobody thought of them but because the data to answer them did not exist in a form anything could query.

Questions shaped like this one: when that competitor is in the deal, and the pain is this, and the objection raised is that, what tends to happen? If the answer is that you lose most of the time, you have found a losing pattern, and it is now a property of your business rather than a theory somebody floated in a pipeline review.

Once the pattern is known the system can act on it while the deal is still alive. Raise the risk on the opportunity. Change a MEDDPICC letter from green to something less flattering. Move the forecast score. Trigger whatever comes next, which might be a set of qualifying questions the rep has not asked, or pulling in a domain expert while there is still time for that to matter.

That chain is what sits underneath the 64% reduction in deal slippage our customers report. Every tool in this category can fire an alert. Making the pattern detectable in the first place is the part that requires the data underneath to be real.

Tulip ran this across a 65-person sales team and went from 6.3% to 15% conversion over twelve months.

What it costs you in tokens

Nothing, and it is worth a paragraph because finance always asks.

Turning transcripts into structured, labeled, inspected data and then analyzing it for patterns happens inside our model. It is not a chain of API calls billed to you by the word. So when your team asks the pipeline a thousand questions in a quarter, nobody has to work out afterwards whether the questions were affordable.

A build assembled from a transcript tool and a general model behaves differently. Every new question re-reads the deal, and you pay for the reading and for the answer.

The part you won't enjoy

Your competitor's field is going to look emptier than it does today.

A general model fills every row it is asked to fill. CDAM leaves a row alone when the evidence is not there. Since field completeness is the metric most RevOps teams have been optimizing against for years, the first fortnight reads as a regression, and there is no way around that except going through it. The feeling goes when you run your first competitive win-loss cut that you actually trust.

The initial list is also yours. The sweep will tell you what to add and what to drop, but it can only work with what you hand it, so a list that was already neglected will take a month of inspection before that becomes obvious.

One question for anyone selling you this

Not accuracy rates. Every vendor in this category has a number for that and none of them are comparable.

Ask them to show you the sentence.

See how Spotlight.ai inspects deals from evidence at Spotlight.ai.