What is opportunity scoring?

Opportunity scoring works by ingesting structured and unstructured data from across the revenue stack, training a machine learning model on historical win/loss outcomes, and then scoring every open deal against the patterns that predict a close.

The input signals fall into several buckets: deal properties (size, product fit, industry, source), stage velocity (how fast the opportunity is progressing relative to average), stakeholder engagement (how many contacts are involved, whether an economic buyer has been touched), activity levels (emails sent, calls completed, meetings booked), and conversation intelligence (sentiment, competitor mentions, commitment language on calls). A deal progressing quickly with multiple engaged stakeholders and a recent executive meeting scores high; one stalled in the same stage for 45 days with only one low-level contact scores low.

Most platforms update scores daily or as new activity is logged. Some, like Salesforce Einstein, rebuild the underlying model monthly against the latest closed data, which means the scoring logic evolves as the company wins and loses deals over time.

The signals that carry the most weight cluster into four categories: activity recency and volume, stakeholder breadth, stage and time dynamics, and conversation quality.

On the activity side, research consistently shows that more touches — calls, emails, meetings — correlate with higher close rates, but recency matters most. A deal where no one has responded in 14 days is at risk regardless of how active it was 30 days ago. Clari specifically flags "no activity in 14 days" as a red-alert signal for deal risk, independent of what stage the deal occupies in the CRM.

Conversation intelligence adds a qualitative layer that rule-based CRM scoring cannot capture: Gong draws roughly half of its deal likelihood score from call and email data alone, detecting whether a champion is still engaged, whether objections have been raised and addressed, and whether the buyer is using buying language ("when we roll this out") versus hedging language ("if we move forward"). This distinction — between a rep who is logging activity and a buyer who is genuinely advancing — is the core reason ML-based scoring outperforms spreadsheet-based scoring.

Sales forecasts that rely on rep-submitted pipeline are systematically optimistic. Reps tend to hold onto deals they believe in, even when the data shows momentum stalling. A 2024 Gong survey of 2,015 business leaders found that more than 80% of companies missed their revenue forecast in at least one quarter over the previous two years — a figure consistent with broader industry research on forecast miss rates.

Opportunity scoring introduces an objective, model-driven view that sits alongside — not instead of — the rep's judgment. Managers can see the gap between what reps are calling and what the model predicts, and use that gap to coach, pull resources, or reforecast early. Gong's AI model claims to be 21% more precise than reps as early as week four of the quarter, which gives revenue leaders a meaningful head start on identifying shortfalls.

The practical benefit compounds at the portfolio level. When every open deal carries a probability score, managers can roll up a weighted pipeline number that is more reliable than a simple stage-based calculation, and can allocate coaching time to the highest-leverage at-risk deals rather than the most vocal reps. Bain & Company's 2025 Technology Report found that early AI deployments in sales have shown 30% or better improvement in win rates, though the research cautions that the gains require reimagining the sales process, not just automating it.

Lead scoring and opportunity scoring address different stages of the funnel and answer different questions. Lead scoring asks: is this person or account worth pursuing at all? It evaluates pre-pipeline prospects against firmographic fit, engagement with marketing content, and behavioral signals like website visits or content downloads, and outputs a qualification decision.

Opportunity scoring asks: given that this deal is already in the pipeline, how likely is it to close, and when? It draws on deal-specific data — CRM activity, conversation transcripts, stakeholder maps — that does not exist until a deal is created. The two systems are complementary: a well-scored lead becomes a qualified opportunity, and a well-scored opportunity closes on schedule.

A key structural difference is that lead scoring typically evaluates an individual contact, while opportunity scoring evaluates a deal that may involve an entire buying committee. In B2B sales, where five to ten stakeholders often participate in a purchase decision, opportunity scoring is better suited to capturing the multi-stakeholder reality that determines whether a deal actually closes.

The most frequently cited failure mode is data quality: a scoring model trained on incomplete or inaccurate CRM data produces scores that reps learn to distrust and eventually ignore. Research consistently shows that the majority of failed AI sales initiatives trace back to dirty data — duplicates, missing fields, and unverified contacts that create false patterns in the model. Prospeo.io's 2026 analysis of deal-scoring implementations emphasizes that when CRM data is clean, scores predict outcomes; when it is not, the model amplifies noise.

A second failure mode is static thresholds. Rule-based scoring systems — where a rep manually weights factors in a spreadsheet — go stale as product mix, market conditions, and buyer behavior change. ML-based models that retrain on recent closed data avoid this, but require minimum deal volume to work: Salesforce Einstein needs 200 closed-won and 200 closed-lost opportunities, and Gong's deal predictor requires at least 100 qualifying closed deals before the model stabilizes.

A third issue is over-indexing on activity. A deal that looks active in the CRM because the rep is logging calls can still be heading to a loss if the buyer side has gone quiet. Platforms that layer in conversation intelligence — detecting whether the champion is still engaged rather than just whether calls occurred — catch this failure mode that activity-only models miss.

Komo is built around the idea that the hardest part of signal-based selling is not identifying signals, but acting on them quickly and consistently — for every deal, every week, without burning out the rep. Opportunity scoring fits directly into that workflow as the triage layer: it tells Komo's engine which deals in the pipeline have shifted in health so that research, drafting, and follow-up can be queued automatically.

When a deal's score drops — a champion goes quiet, a close date slips, a competitor is mentioned on a call — Komo surfaces that signal, researches what changed (new stakeholder, recent news, role change), and drafts a recovery touch for the rep to review and send. The rep stays in the loop on every send that matters; the scoring infrastructure removes the burden of manually reviewing every deal in a 40-opportunity book.

Komo does not replace the rep's judgment about a deal — it replaces the administrative work of staying on top of every deal simultaneously, so human judgment is applied where it actually moves the needle: on the relationship and the conversation.