---
name: loss-cause-differential
description: "Load when the user pastes a batch of closed deals (outcome, value, competitor, CRM loss reason, rep notes or interview excerpts) and wants an internal win/loss debrief. Returns an executive summary that carries the finding, a method and data-quality section, reason-frequency and revenue-weighted tables, competitor evidence kept within what the counts support, a root-cause narrative with verbatim quotes, owner-mapped recommendations, and a per-deal appendix. Not for battlecards, single-deal post-mortems, or forecasting."
---

# The Loss-Cause Differential

A CRM loss reason is a referral note. Someone picked it from a dropdown on the day the deal died,
usually the person standing closest to it, often to clear a required field. It tells you where to
start looking. It is never the finding.

The finding comes from the record — rep notes, the buyer's recorded words, call summaries,
post-mortem fragments — and answers one question: **which single circumstance, had it been
different, most plausibly changes this outcome, and whose desk does it sit on?**

That is why a rival's name is such a weak answer. "We lost to them" names the venue, not the cause.
Ask what the buyer got there that you could not supply, and the answer usually lands elsewhere — on
a missing capability, on a price the buyer could evidence, on a sponsor who left, on a decision that
never happened. Route the loss to the desk that owns the circumstance, not the one that owns the
rivalry.

Use the requester's taxonomy when supplied. Otherwise: Product Gap, Pricing, Competitor Win,
No-Decision / Status Quo, Champion Loss, Timing, Sales Execution. **One primary cause per deal**, at
most two contributing factors. Contributors shape narrative and enter no count and no dollar
subtotal — a deal appearing in two totals has been counted twice.

## What the record has to show before a cause is bound

Each cause has an evidentiary price. Run the differential in this order, because the cheap
categories are the ones that swallow deals they have not earned.

- **Product Gap** — a named capability, requirement or standard the buyer needed and you could not
  supply. This is the *upgrade rule*, the one most often missed: when a deal is tagged Competitor
  Win but the note names a specific missing capability, the primary is **Product Gap** and the rival
  is a contributing factor. The capability would have been decisive against any credible
  alternative; the rival merely shipped it first. Bind this in the frequency table, not only in the
  prose — if the narrative calls something a product gap, the table must count it as one.
- **Distinct gaps stay distinct.** Deployment model, a named integration, a compliance standard and
  a licensing structure are four different findings that happen to share a rival's name. Collapsing
  them into one "integration gap" destroys the only thing the report was for — telling Product what
  to build first. A licensing-model mismatch is not Pricing: the buyer objected to the shape of the
  contract, not the size of the number.
- **Pricing** — a specific price comparison, a quantified delta, or a stated budget ceiling the deal
  exceeded. A buyer saying the words "too expensive" is a report of a feeling; it becomes Pricing
  when the record shows what it was too expensive *against*. Absent that, Pricing is not available.
- **Competitor Win** — the buyer preferred the rival on grounds you can name that are not a
  capability you lack, a price you can evidence, or a pursuit you lost. "They went with the bigger
  brand" qualifies, weakly. "Demo went well, then we lost to them" does not — that is an outcome
  with no cause attached.
- **Champion Loss** — a documented sponsor departure or loss of standing that initiates the
  collapse. Later silence and failed re-engagement are contributing context, not a reclassification.
- **No-Decision / Status Quo** — the buyer chose nothing. This is where a stall belongs. **A deal
  that goes quiet and ends in a polite exit is a No-Decision even when the CRM says Pricing** — the
  courteous "it's a budget thing" at the end of a stall is an exit line, not a price objection.
- **Timing** — the buyer named a specific external event with a date or a revisit window: a freeze,
  a reorganisation, a funding round. A named event separates Timing from a generic stall.
- **Sales Execution** — the buyer indicates the products were comparable and something in the
  pursuit decided it: access, references, responsiveness, a proof-of-concept you did not staff.

## When nothing clears the bar

**Unclassified is a finding in good standing, not an apology.** A note that reads only `lost —
budget`, or `no reason given`, or that simply repeats its own tag in different words, supports
nothing. Segment, deal size, the rest of the batch and the analyst's intuition cannot upgrade it,
and neither can the CRM tag, which is the assertion under examination rather than evidence for it.

Two failures cost more than an honest blank. The first is smoothing: quietly distributing thin
records into the most plausible neighbouring category so the distribution looks tidy. The second is
inventing a cause from an absence — a missing follow-up, an empty activity log or a silent month is
a gap in your own record-keeping, and a record-keeping gap is a process finding, never a loss cause.

When the unreadable share of a batch is large, that share *is* the headline. A debrief that reports
half its deals as uninterpretable and says so plainly is worth more than one that reports a clean
taxonomy it cannot defend, because the first one tells the reader what to fix in the CRM.

## Grading certainty, and holding a deal open

Every binding carries a grade. **High** — the buyer said it, or the fact is undisputed. **Medium** —
the strongest reading, with one live alternative. **Low** — thin material, or two readings the
record genuinely cannot separate.

A Low grade is not a weaker version of a decision. When two or three causes survive the evidence,
**name them all and say what would distinguish them.** Write which reading you lean to and why, then
say what the record does not settle. Do not resolve an ambiguity by deferring to the CRM tag, and do
not present a forced choice in confident language — a clean-looking classification that the file
cannot support is worse than an open one, because the reader acts on it.

## Reading the batch before reading any deal

If the paste is a whole CRM export, use closed rows only and say how many open rows were set aside.

Count first: total records, won, lost, substantive notes, thin notes, tag-versus-note contradictions,
and every competitor appearance. Then apply three tests that change what the report is allowed to say.

**Concentration.** Find the largest loss and its share of lost value. When one deal materially
controls the money view — roughly a third or more — that deal is its own event, not a pattern. Say
so where the money is first mentioned, remove the full amount from both the numerator and the
denominator, and state what leads once it is gone. A single $1,268,000 loss making its cause 61% of
lost revenue is one story about one deal, and a reader who is not told that will fund a roadmap
against it.

**Sample size.** Below roughly fifteen deals a batch is directional; in single digits it is a
structured deal review and not statistics at all. Percentages may describe the deals in front of you
and may not be extrapolated to pipeline, targets or headcount. Scale every recommendation to that.

**Competitor counts.** Rates come from outcomes, not from cause labels. Report wins, losses and `n`.
Below about ten observations a fraction is descriptive only — say that it cannot characterise a
stable rate, and give the count at which it should be revisited. One win and one loss against the
same rival is two structurally different situations, not a 50% win rate; if they reflect different
buyer contexts, that difference is the finding and the fraction is noise. **If the batch contains no
wins at all, do not print an empty matrix and do not silently omit it — state that no win rate is
producible from a batch with no wins, and say it where the reader will see it.**

## Where the report is actually read

A revenue leader reads the top of the document and delegates the rest. Whatever you have decided,
the first three hundred words have to carry it, and everything below has to survive being checked.

So the executive summary is not a warm-up. It is the report. Before the first table, in prose, it
has to have already said:

- what the batch is, and whether its size supports pattern claims or only deal-level review — if
  that limit exists, it belongs in the opening sentence, not in a caveat further down;
- the one finding you would act on, with the mechanism named rather than the category label;
- any deal large enough to control the money view, flagged as a single event, with what leads once
  it is set aside;
- how much of the batch was not interpretable and why, by ID, framed as a process finding;
- any CRM reason you are overriding, and what the record says instead;
- what the competitor evidence can and cannot support at the counts observed — including an
  explicit refusal when there are no wins or too few observations;
- whatever the requester singled out, answered directly, with the competing readings named if it
  is genuinely undecided.

Write it as judgement in continuous prose, not as a list of headings with values after them. Arithmetic
belongs in the tables below it; the opening carries decisions, and a number appears there only when
it *is* the decision.

## The brief

Begin the reply with the report title as a level-one heading. No preamble, no narration of your
process, nothing before the title.

1. **Executive Summary** — as above.
2. **Method & Data Quality** — batch boundary, note census, taxonomy used, the confidence standard,
   contradictions found, exclusions, and the sample-size limit stated plainly.
3. **Reason-Frequency Table** — counts with percentages, then a revenue-weighted table, with the
   rebased view alongside it whenever concentration was material. Unclassified appears as its own
   row in both.
4. **Competitor Evidence** — wins, losses and `n` per rival, what the contrast supports, and the
   explicit limit or refusal. Where the same rival was both beaten and lost to, compare the buyer
   conditions rather than reporting a rate.
5. **Root-Cause Narrative** — ranked by evidence strength and decision impact, not by count. Each
   finding is anchored to short quotes reproduced **exactly** as they appear in the record; a
   paraphrase inside quotation marks is a fabrication. Give an ambiguous or requester-flagged deal
   its own subsection with the surviving readings set out.
6. **Recommendations** — a table of finding, owner team, the specific next action, the weight of
   evidence behind it, and how you would know it worked. Confidence in the row must match confidence
   in the finding it came from.
7. **Appendix: Deal Rulings** — every input ID exactly once, with outcome, amount, primary cause or a
   dash for wins, contributors, grade, and the excerpt that decided it.

Adapt the depth to the material. A thin batch earns a shorter, more hedged report, and a section
with nothing to say should say why it is empty rather than be padded or dropped in silence. Finish
every section and every table; a report that stops mid-appendix is unusable.

## Figures that reconcile

Build one row per unique ID before writing anything, then hold the totals to it: wins plus losses
equal the record count; primary counts including Unclassified equal the losses; primary dollar
subtotals equal total lost value; contributors enter no subtotal; every percentage has a visible
denominator; and a rebase subtracts the same full amount from both sides. Publish one reconciled set
of numbers — never a correction trail or two competing totals.

## Four files, worked

Invented throughout. Carry the reasoning, not the particulars.

**Northvale Systems — a heavy deal beside unreadable ones.** Fourteen losses; one at $1,268,000
against a rest averaging $88,600. Its cause takes 61% of lost value on the money view and 7% on the
count view. The opening names it as a single event and gives the rebased leader; four other deals
carry only the word `budget` and are Unclassified by ID, reported as a CRM-hygiene finding.

**Sable Peak — one rival, six desks.** Every deal tagged Competitor Win against the same name. The
notes separate into two capability gaps on different dimensions, one Pricing with a quantified delta,
one Champion Loss, and one that says only that the buyer preferred them. Five findings, five owners.
The tag described the venue; nothing in it described a cause.

**Voltrix — seen twice.** One win, one loss. The win was a technical evaluator, the loss a
procurement-led committee. Reported as two cases with the contrast drawn, explicitly not as a 50%
rate, with a revisit point named once the count reaches double digits.

**Marnhold — a departure and a silence.** The sponsor left in month two; the record then shows five
months of nothing, then a decline with no reason. Champion Loss leads at Low grade because the
departure is the only dated event in the file, with No-Decision and a pursuit failure both named as
live alternatives, and the five-month hole reported as the process finding it is.

## The bar

Ship when every deal is bound once at a stated grade, every quote appears verbatim in the source,
every total reconciles, no cause is carrying a deal it did not earn, and the opening three hundred
words would let the reader act without scrolling.