Gong + Momentum Forecast Augmentation (L4) Playbook

Why this matters

The traditional sales forecast is a work of fiction. Every Friday, reps stare at a spreadsheet and "gut-feel" their numbers based on how they think a deal is going, often ignoring the red flags buried in their own call recordings. This manual process costs a mid-market sales org hundreds of hours in "forecast prep" and results in quarter-end surprises that swing wildly—often ±20% of the actual landing spot.

When you miss your forecast by 20%, you stop hiring, you miss quarterly targets, and you lose board-level credibility. The cost of doing nothing is a permanent "guess-work tax" on your revenue operations. By moving to an L4 maturity level—augmenting conversation intelligence (Gong) with structured hygiene (Momentum)—you replace the rep-authored forecast with an AI-generated baseline that reps simply adjust. The result: forecast accuracy increases by 20–30 points and rep prep time drops by 60%.

How it works

1. Pipe Gong + Momentum signals into the warehouse

To build a predictive model, you need more than "Stage: Discovery." You need the raw signals that correlate with winning.

Using Fivetran or Airbyte, you must land two specific data streams into Snowflake or BigQuery:

• Gong Signals: Extract call sentiment, talk-to-listen ratios, and the frequency of competitor mentions.
• Momentum Signals: This is the bridge. Momentum structured data (like MEDDPICC fields autofilled from Slack/calls) provides the "why" behind the deal.

Pro Tip: You cannot train a model on "current state" data alone. You need a weekly snapshot table (e.g., fct_deal_history) that captures what a deal looked like at that exact moment in week 3 of last quarter. DoD: A weekly snapshot table (deal_id, week, signals) with 12+ weeks of history.

2. Build the augmented score in dbt

With the data landed, your Analytics Engineer uses dbt to build a weighted win-probability model. A standard Salesforce stage might give a deal a 25% chance of closing, but your dbt model should adjust that based on real-world behavior:

• The Penalty: Sentiment trend is down over the last 3 calls + no Decision Maker engaged in 14 days.
• The Boost: MEDDPICC completeness is >80% + "Pricing" was mentioned in the last Gong call.

The Math: Validate this model against the last four quarters of closed-won/lost history. If your dbt model's F1 score (a measure of accuracy) doesn't beat the baseline SFDC stage-only forecast by at least 15%, your weights are wrong. Timeline: 2 weeks for v1.

3. Surface the AI forecast in the rep’s SFDC view

The biggest failure point in RevOps is building a "shadow forecast" in a BI tool that reps never see. Use Salesforce custom fields to push the AI Win Probability and the Top 2 Reasons (e.g., "High sentiment, but no Economic Buyer identified") directly onto the Opportunity record.

When the rep opens the deal, the AI has already told them the score. If the rep disagrees, they can "override" the number—but they must provide a written justification. This creates a data loop: if a rep consistently overrides the AI and loses, you have a coaching opportunity.

4. Run forecast as “AI-first, rep adjusts”

The Friday ritual changes. Instead of asking "What are you calling for the month?", the VP of Sales asks, "The AI says you're at $1.2M, but you've manually adjusted to $1.5M. Why do you believe in these three deals more than the data does?"

This shifts the manager's role from "data investigator" to "deal coach." By the second quarter of running this "AI-first" motion, your week 6 forecast should land within 8% of the actual end-of-quarter number.

Tools you need

• Gong: For conversation intelligence (sentiment, talk-ratio, topics).
• Momentum.io: For structured MEDDPICC capture and automated field updates.
• Snowflake/BigQuery: To house the historical snapshot data.
• dbt: To transform raw signals into the weighted "Win Probability" score.
• Salesforce: As the system of record and primary UI for the reps.

KPIs to track

• Forecast Accuracy: The delta between your Week 6 "Commit" and the final actuals. Aim for <10% variance.
• Rep Prep Time: Track time spent in "Forecast" meetings. Expect a 50%+ reduction as the AI does the heavy lifting.
• Override Rate: The % of deals where a rep manually changes the AI score. High override rates suggest either a broken model or "happy ears" in the field.

Common pitfalls

• Landing current state only: If you don't snapshot your data, you're training on "survivorship bias." You need to know what a losing deal looked like in Week 4.
• Hiding the score: If the score is in a Tableau dashboard 5 clicks away, it doesn't exist. Put it in Salesforce.
• Ignoring the "Why": Never show a score without the 1-2 reasons behind it. Without the "why," reps won't trust the "what."

When to graduate to the next level

Once your dbt model is consistently outperforming your managers' gut feel, you are ready for L5: Automated Pipeline Generation Modeling. At that stage, you aren't just forecasting what's in the pipe; you're forecasting what will be created and closed within the same quarter based on top-of-funnel activity signals from tools like Clay or 6sense.