AI win-loss analysis: a decision guide for CROs and CMOs

When a lost-deal report says “price,” a CRO and CMO still have a decision to make. Should they change the offer, explain its value differently, improve the buying process, or find out more before acting?

AI win-loss analysis uses AI to organize evidence from completed buying decisions, compare recurring themes, and prepare commercial choices for human review. It should preserve the distinction between what the CRM records, what sellers believe, and what buyers report. A generated explanation cannot resolve disagreement among those sources by sounding confident.

For executives, the useful output is a short recommendation with its evidence, limits, and next test. Start with one question, such as whether implementation concerns justify changing the enterprise sales narrative. Keep prices, customer contact, and employee assessments within existing approval processes.

Define the decision before collecting more deal data

A closed-deal review looks backward to improve future commercial choices. It differs from investigating pipeline risk before a forecast meeting, which asks what threatens opportunities that are still live.

It also has a narrower evidence base than competitive intelligence. A buyer’s description of a competitor explains that person’s perception; it does not verify the competitor’s current capabilities.

Choose one executive decision for the review:

Positioning: Should the CMO test a different explanation of value for one buyer segment?
Sales execution: Should the CRO change when implementation or security specialists join an evaluation?
Offer design: Is there enough evidence to investigate a packaging problem with finance and product leadership?
Research: Which unanswered buyer question deserves the next round of interviews?

Assign a commercial sponsor and a research owner. Product marketing or revenue operations can prepare the work, but the executive sponsor should state what decision the evidence must support. Otherwise the team can produce an extensive report that nobody has authority to use.

Keep outcomes, observations, and explanations separate

Win-loss research asks buyers about a completed choice and turns the findings into decisions. SurveyMonkey’s win-loss framework describes gathering feedback from both won and lost buyers and coding responses into themes. Including wins helps reveal what persuaded buyers despite concerns that also appear in losses.

For an AI-assisted review, use three distinct records:

Record	What it establishes	What it cannot establish alone
CRM outcome and reason code	The recorded disposition, date, and seller-entered explanation	The buyer’s decisive reason
Deal correspondence and approved call records	Statements and events captured during the evaluation	Everything discussed inside the buying organization
Post-decision buyer feedback	A participant’s account of the decision and its drivers	Every stakeholder’s view or a causal estimate of a proposed change

A seller may record price while a buyer describes uncertainty about implementation. Preserve both, with their authors and dates. The disagreement is a research question, not a reason for AI to overwrite the CRM.

Keep the original text beside any assigned theme. “Implementation risk” and “missing integration” may require different responses even when a model groups them together. If the source does not support a reason, use unknown. Missing evidence should remain visible in the executive brief.

Lock the cohort and denominator

Before interpreting themes, agree on the population: sales motion, customer segment, geography, product version, close-date window, and reporting cutoff. Separate new business from renewals and expansions unless the decision genuinely spans all three.

Document how the CRM treats no-decision, disqualified, duplicate, reopened, and still-open opportunities. A prospect that selected another supplier differs from one whose project was canceled. Report those dispositions separately even if both sit inside a closed-lost CRM stage.

For a conventional closed-opportunity win rate, calculate:

Closed-won count ÷ (closed-won count + closed-lost count).

Show counts alongside percentages and state which dispositions enter the denominator. Keep open opportunities outside this closed-outcome calculation. If you also show a competitive-decision rate that excludes no-decisions, label it separately and reconcile it to the full population. Avoid letting a narrower denominator masquerade as improved performance.

Choose whether the decision needs a deal-count view, a value view, or both. Five small wins and one large loss tell different stories under those measures. Record the amount basis for lost deals; a seller’s estimated opportunity value is not realized revenue.

Use reviewed queries or calculations for these totals. Ask AI to explain the calculation and expose inconsistent records, while leaving arithmetic and inclusion rules reproducible outside the generated narrative.

Make the missing buyers visible

An interview sample can be informative without representing every buyer. The executive needs to see who was eligible, invited, and heard from, broken down by the segments relevant to the decision.

Record exclusions and nonresponses. If account owners exclude sensitive losses, preserve the reason and show how that changes coverage. Do not ask sellers to nominate only memorable or cooperative buyers. Clozd’s guide to win-loss interviews recommends starting from the full list rather than relying on volunteered contacts, and using open questions that do not steer the answer.

A practical interview guide should explore the buyer’s sequence of decisions. For example: what initiated the evaluation, which alternatives survived, what changed during the process, and what ultimately determined the choice? Ask for concrete events behind broad labels. A budget constraint, an unclear business case, and an unexpected service charge might all be described as price concerns.

Seek perspectives from the roles that matter to your question. An end user can explain usability concerns; the budget owner may describe a different approval constraint. Record participant roles without assuming that one person’s account represents the entire buying committee.

Approve outreach and recording practices with the relevant account, privacy, and legal owners. Restrict transcripts to the intended research audience and establish retention rules. An executive summary usually needs a relevant excerpt and role, rather than a copy of every private exchange.

A worked example: should leadership lower prices?

Consider a hypothetical B2B software company reviewing 80 closed new-business opportunities in one segment. It has 24 wins and 56 closed losses. Of those losses, 16 are recorded as no-decision; the remaining 40 identify another chosen supplier.

Its closed-opportunity win rate is 24 ÷ 80, or 30%. Excluding the 16 no-decisions produces a competitive-decision rate of 24 ÷ 64, or 37.5%. Neither number has improved the business. They answer different questions about the same outcomes.

The CRM marks 28 of the 56 losses as price-related. Leadership is considering a broad discount change. Before approving it, the research owner invites buyers from 40 opportunities: 12 wins, 20 competitive losses, and eight no-decisions. Twelve participate: four wins, six competitive losses, and two no-decisions.

Of the six interviewed competitive losses, four have a CRM price label. Three of those four buyers describe implementation effort as a material concern. Two interviewed winners also mention implementation effort, but say an agreed rollout plan made it manageable.

This evidence does not establish that implementation explains most losses. The interviews cover only part of the invited group, and participation may be related to buyers’ experiences. Nor does it prove that a rollout plan caused the wins.

It does identify a testable commercial hypothesis: earlier implementation planning may help buyers assess the total commitment. Leadership can choose among three responses:

Change discounts broadly. This requires a stronger pricing case, including margin implications and evidence that the proposed concession addresses the obstacle. The executive pricing-strategy guide covers that downstream decision.
Test earlier rollout planning in a bounded sales motion. The CRO names an owner and defines eligible opportunities, the intervention, and how results will be assessed. The CMO keeps messaging consistent with what delivery can support.
Commission more buyer research. Seek missing budget-owner perspectives and additional no-decision accounts if those gaps could reverse the choice.

A defensible recommendation could combine the second and third responses while leaving the broad discount policy unchanged. The review has narrowed uncertainty without pretending to settle the market’s willingness to pay.

Use a one-page win-loss review template

Keep the decision on one page, with links to supporting records. The following template is an editorial recommendation, not a statistical standard:

Decision required: What commercial choice is proposed, for which segment, and by when?
Population: Cohort dates, inclusion rules, outcome counts, no-decisions, and amount basis.
Research coverage: Eligible, invited, participating, and excluded accounts; relevant participant roles and gaps.
Finding: The theme, number of distinct opportunities supporting it, source types, and evidence links.
Counterevidence: Wins with the same concern, conflicting buyer accounts, and plausible alternative explanations.
Options: Act, run a bounded test, or investigate; include the consequences and unresolved assumptions.
Accountability: Recommending owner, approving executive, execution owner, and review date.

Count distinct opportunities separately from excerpts. Ten mentions in one deal do not equal ten independent buyers. If several themes can apply to each opportunity, explain that their percentages can exceed 100% in total.

Clozd’s decision-driver documentation illustrates tagging interview themes and linking them to supporting quotations. For your own brief, retain the path from theme to source and distinguish how frequently something was mentioned from how decisive a buyer said it was.

Test the AI review before executives rely on it

Replay one completed review with the same source cutoff a human researcher used. Have that researcher inspect theme assignments and consequential summaries. Include a duplicate transcript, an unknown loss reason, a buyer who contradicts the seller, and a winner who mentions the leading loss theme.

The workflow should preserve those conditions rather than smoothing them into one story. Check whether it:

reproduces the agreed counts and cohort rules;
traces each substantive finding to an authorized source;
distinguishes buyer statements from seller interpretations;
exposes unsupported themes, missing feedback, and counterevidence;
avoids identifying a competitor winner when the outcome is unknown;
keeps outreach, CRM edits, and commercial changes behind approval.

The broader guide to evaluating AI agents on business data provides a framework for assessing grounding, permissions, and usefulness across the complete job.

Measure preparation effort, reviewer corrections, evidence coverage, and whether the review produces an owned decision or test. A subsequent win-rate change needs separate analysis of deal mix, timing, and other commercial changes before anyone attributes it to the AI workflow.

Carry the learning into the next decision

A future Chief of Staff experience could remember which hypothesis leadership accepted, who owns its test, and when new buyer evidence should reopen it. That memory should retain the segment, source dates, and expiry conditions. An old finding about one product version should not become a permanent instruction about all buyers.

This is a design direction, not a claim that Jovis currently conducts buyer interviews, operates a win-loss program, or changes commercial policies. Jovis’s current product positioning centers on finding relevant context across business systems and preparing leaders for decisions. The appropriate evaluation here is whether it can help prepare one review from your approved evidence while commercial leadership retains the choice and its consequences.