Win-Rate Forensics

Win-rate forensics is the practice of explaining a company's win rate from the evidence its own systems already hold. It joins every opportunity record with the calls, emails, chat threads and intent signals attached to that same opportunity, then counts the patterns across all of them. Every finding it produces carries a denominator, a piece of supporting evidence and a named owner.

The word forensics is doing real work in that sentence. A forensic account of an event is reconstructed after the fact, from physical traces, by someone who was not there. It does not ask the participants what they remember. It is reproducible, it is counted, and it survives disagreement because anyone can go back to the same traces and check.

That is the difference between win-rate forensics and most of what passes for win-loss work. The question is not what your sellers think happened. It is what the record shows happened, in how many deals, and to whom it belongs.

Win-rate forensics, defined

A win-rate forensics engagement starts from a question a revenue leader cannot answer with the tools already installed. Something in the shape of: we believe we lose on price, but do we? Or: our enterprise win rate dropped eleven points last year and nobody can say why.

To answer a question like that you need four things, in order.

1. A census, not a sample. Every opportunity in the period, not the ones a seller remembers or a buyer agreed to be interviewed about.
2. A join at the opportunity level. The call from March, the procurement email from May and the Slack thread where the discount was agreed all have to resolve to the same deal record, or they cannot be counted together.
3. A counted pattern. Not an anecdote that sounds true. A number over a denominator: of the 62 deals that qualified, 14 did this, and of those 14, none closed won.
4. An owner. A finding that belongs to nobody changes nothing. Each one has to land on a function that can act on it.

Work that produces all four is forensics. Work that produces a narrative, however well written, is research.

The four tests a finding has to pass

In practice we hold every candidate finding against the same four tests before it goes in front of a leadership team. Most do not survive. That is the point of having them.

The counterfactual test is the one teams skip most often, and it is the one that does the most damage when skipped. A pattern that appears in your losses is only interesting if it appears less often in your wins. Reviewing lost deals on their own, which is how most loss reviews are run, makes that comparison impossible by construction.

How it differs from win-loss analysis

Win-loss analysis, as the category is normally practiced, means commissioning interviews with buyers after a deal closes. A research firm calls a sample of won and lost accounts, asks a consistent set of questions, and returns a qualitative report, often quarterly. Firms such as Clozd, DoubleCheck Research, Anova Consulting and Primary Intelligence built a profession around doing this well.

It produces something win-rate forensics cannot: the buyer's stated reasoning, in the buyer's own words, including factors that never touched your systems because they happened in a room you were not in. If you want to know how your brand was perceived against a competitor you never met, you have to ask.

But the method has three structural limits.

• It is a sample, and a self-selecting one. Buyers who agree to a post-mortem interview are not a random draw from your pipeline. The ones who ignored your seller for four months are the least likely to take the call, and they are often the ones you most need to understand.
• It relies on recall. You are asking someone to reconstruct a decision made months ago, under pressure, often by a committee. People reliably supply a tidier reason than the one that operated at the time, and price is the tidiest reason available.
• It does not scale to a denominator. Thirty interviews a quarter is a substantial research programme. It is also thirty out of perhaps four hundred deals, which is not enough to say how often anything happens.

The two methods answer different halves of the same question. Interviews tell you what the buyer believes. Forensics tells you what the organization did. When they disagree, that gap is usually the most valuable finding of the year.

We are not a replacement for a win-loss research programme, and we say so on the first call. If you run one already, keep it. The engagement gets better when both exist, because the interview findings become hypotheses that can be counted.

How it differs from conversation intelligence

Conversation intelligence platforms record, transcribe and analyze sales calls. Gong is the best known. They are genuinely good at what they do: surfacing what was said on a call, coaching a rep on talk-time or discovery quality, flagging a deal that has gone quiet, and now, through the Model Context Protocol, pulling in context from systems outside their own walls.

The distinction is the unit of analysis. A conversation intelligence agent works on the record in front of it: this call, this deal, this account. Ask it why your enterprise win rate fell and it will retrieve what it can at the moment you ask, summarize it, and give you a well-cited narrative. What it will not give you is a count across every deal in the period, because that is not the shape of the question it was built for.

Our engagement uses the conversation platform's export as one input among five. You keep the platform. Nothing about this displaces it. The longer argument is in the white paper Reach Is Not Analysis.

How it differs from revenue intelligence and forecasting

Revenue intelligence tools, Clari among them, score pipeline and predict what will close. They are forward-looking by design and they are measured on forecast accuracy.

Forecasting tells you what is likely to happen given how things currently work. Forensics tells you why things currently work that way. A model can tell you that deals in this segment close at 19 percent without being able to tell you what the 19 percent is caused by, and a model tuned for prediction has no reason to care, because accuracy and explanation are different objectives.

In practice the two are complementary and the sequence matters. A forecast that keeps missing in the same direction is a symptom. Forensics is how you find what it is a symptom of.

What it needs from your systems

The method is only as good as the join. These are the sources that make the join possible, in rough order of how much they contribute.

• CRM opportunity history, including field-level change history. Not the current state of the record. The current state has already been tidied by whoever closed it. The change history is where the original close date, the original amount and the first-entered loss reason still live.
• Recorded calls and their transcripts, with timestamps and participants, so a moment can be located rather than summarized.
• Email threads with external participants, which is usually where procurement, legal and security enter the deal, and where the real objection first appears.
• Internal chat on the deal, which is where sellers say what they actually think is going on, in a register they would never use in a CRM field.
• Intent and marketing engagement, which establishes what the account was doing before it ever spoke to you.

Two of those five carry most of the weight, and both are commonly overlooked. Field-level change history is the single highest-value source in the set, because the gap between the first loss reason entered and the final one is itself a measurable phenomenon. Internal chat is second, for the same reason: it is the only source written without an audience in mind.

All of it is read inside the client's own infrastructure. The joined dataset stays with the client when the engagement ends.

A worked example

Here is what one finding looks like when it has passed all four tests. This is from a live engagement, with the account names removed.

Note the last row. The CRM was not lying. The seller who wrote "price" was describing the conversation they remembered having. The forensic account is that cutting the scope to reach a number removed the part of the deal that justified the project internally, and the project then lost its sponsor. Those are different problems with different owners, and only one of them is fixable by discounting less.

When it is not the right tool

It is worth being direct about the cases where this does not work, because recognizing them on the first call saves everyone a quarter.

• Too few deals. Below roughly 80 closed opportunities in the period, there is no denominator worth having. The answer is interviews, not counting.
• Calls were never recorded. If the richest source is absent for the period in question, the join thins out and the findings get weaker. We will say so before any access is granted.
• The question is about a market you have not sold into. Forensics reads traces your own systems left. It has nothing to say about buyers who never became opportunities. That is a research question.
• The decision has already been made. If the restructure is announced and the engagement is intended to justify it, the work will produce findings nobody wants and the exercise will have been expensive theatre.

Questions we get asked

• What is win-rate forensics?

  Win-rate forensics is the practice of explaining a company's win rate from the evidence its own systems already hold. It joins every opportunity record with the calls, emails, chat threads and intent signals attached to that same opportunity, then counts the patterns across all of them. Every finding carries a denominator, supporting evidence and a named owner.

• How is win-rate forensics different from win-loss analysis?

  Win-loss analysis interviews a sample of buyers after a deal closes and returns their stated reasoning. Win-rate forensics reads a census of every deal from the company's own systems and counts what actually happened. Interviews capture what the buyer believes, including factors that never touched your systems. Forensics captures what the organization did, across all deals rather than a sample. They answer different halves of the same question and work well together.

• Does this replace Gong or Clari?

  No. Conversation intelligence platforms analyze calls and coach reps at the deal level, and revenue intelligence tools forecast what will close. Win-rate forensics counts patterns across every deal to explain why the win rate is what it is. The engagement uses exports from those platforms as inputs. Clients keep them.

• How many deals do you need for this to work?

  Roughly 80 closed opportunities in the period, as a practical floor. Below that there is no denominator worth having and buyer interviews are the better method.

• What does a finding have to contain?

  Four things. A denominator, meaning a count over a defined population. Supporting evidence that can be put on screen with its source and date. A counterfactual, showing the pattern appears less often in won deals. And a named owner, meaning a function that can act on it.

• Where does the analysis run, and who keeps the data?

  Inside the client's own cloud infrastructure, using the client's licensed model. There is no pooling and no training on client data. The client keeps the joined dataset and the agents when the engagement ends.

• Why is CRM field history more useful than the CRM record itself?

  The current state of a closed opportunity has already been tidied by whoever closed it. The field-level change history still holds the original close date, the original amount and the first loss reason entered. The gap between the first loss reason and the final one is itself a measurable pattern.