Revenue forecasting stack — a commit number that survives board scrutiny

The Friday forecast call fails the same way in most companies. The CRO asks why a seven-figure deal is still in commit, the AE says the champion is engaged, and nobody in the room can check the claim. The number gets defended with adjectives, the board hears it, and the quarter lands 12% under.

This stack replaces the adjective with a record. Salesforce holds the deal. Gong holds what the buyer actually said. Clari holds the commitment and measures it against quota. Rattle keeps the fields the other three depend on from going stale. The chain runs one direction: evidence → risk flag → forecast category → commit.

Two of these vendors sell each other’s job. Read the overlap section before you take either sales meeting.

How the pieces fit

• Salesforce is the record. Amount, close date, stage, forecast category, owner. Collaborative Forecasts rolls those up a hierarchy and does it adequately; what native forecasting does not do is tell you whether the roll-up is believable. Enterprise at $175 per user per month is the baseline edition for this stack, because the API, Sales Engagement and Conversation Intelligence entitlements live there. Since hosted MCP servers went GA on April 29, 2026, an external AI client can read the pipeline out of Salesforce over OAuth without a middle tier you maintain.

• Gong is the evidence layer. Calls, emails and meetings become structured signal — no next step scheduled, a single thread into a buying committee, a competitor named in week three. That signal is what makes a risk flag arguable instead of a vibe. Gong shipped MCP support on October 21, 2025 in two directions: the MCP Server lets an external agent in Salesforce Agentforce or Microsoft Copilot query Gong, and the MCP Gateway pulls partner data into AI Briefer and Ask Anything. On June 24, 2026 the Mission Big Dipper release added the Revenue Harness, an agentic execution layer, and Custom Agents that reason from the Gong Revenue Graph with scoped data access and approval gates rather than open access to the record.

• Clari is the commitment layer. Rep commit rolls to manager, manager to CRO, and the roll-up is measured against quota rather than a spreadsheet copy of it. Clari Inspect is where a specific deal gets challenged and the category change gets logged with a reason. Since the Salesloft merger closed on December 3, 2025, execution sits next to the forecast: the April 14, 2026 release lets a manager create a Salesloft task or an AI follow-up email from inside Clari Forecast and Clari Inspect, and shipped an MCP server exposing pipeline movement, deal activity and customer interactions to Claude, ChatGPT, Microsoft Copilot, Gemini and Agentforce.

• Rattle is the hygiene layer. Reps update Salesforce from Slack through interactive forms, and RevOps builds workflows on CRM events without writing Apex. The CRM Data Agent proposes field values with its reasoning shown so the rep confirms in Slack instead of RevOps chasing the update. Rattle holds no state of its own — it writes into Salesforce and stores nothing you would miss — which is why it is both the cheapest layer and the most swappable one.

Named handoffs

1. Call ends → risk flag. Gong transcribes, detects the missing next step or the single-threaded committee, and raises deal risk within hours. The flag reaches the forecast review that week, not the quarterly business review.
2. Risk flag → category change. A flagged deal sitting in commit gets challenged in Clari Inspect. Either the manager moves it to best case, or the AE writes the counter-evidence into the deal. Both outcomes are logged; “we discussed it” is not an outcome.
3. Missing field → Slack prompt → CRM write. A stage change without a close date, or an open opportunity with no activity for 30 days, fires a Rattle workflow into the rep’s DM. The value lands in Salesforce the same day, which is the only reason the two layers above have inputs at all.
4. Commit submitted → measured against quota. Clari rolls rep to manager to CRO against the quota record, so a segment committing under target is visible before the last week of the quarter rather than after it.
5. Forecast gap → execution. From Clari Forecast or Clari Inspect, a manager creates the Salesloft task or the AI follow-up email on the deals carrying the gap. Coverage action happens in the same session as the diagnosis.

The overlap you have to resolve first

Gong sells Forecast. Clari sells conversation intelligence — Clari Copilot, and since the June 2026 release its call data flows natively into Salesloft with recordings, transcripts, AI summaries and action items landing in the Activity Feed, Deal Summary Agent and meeting-prep plays. Each vendor can credibly quote you the other’s half of this stack, and each will.

Buying both without a written boundary produces two deal-risk scores, two sets of call data, and a quarterly argument about which one the CRO looks at. The boundary that holds:

• Gong owns the evidence. Recording, transcript, Revenue Graph, what the buyer said and when. It is the upstream publisher of signal.
• Clari owns the commitment. Categories, submission history, roll-up, quota measurement, and the audit trail of who changed what and why.
• Salesforce owns the record. Neither revenue-AI vendor is the arbiter of what closed.
• Rattle owns nothing. It is an input mechanism, and treating it as a system of record is how teams end up with workflow logic nobody can find.

If you cannot defend two revenue-AI contracts, run one. The rule for choosing: if the expensive problem is the forecast call — repeated misses, categories that mean different things to different managers — keep Clari and use Salesloft conversation intelligence for call capture. If the expensive problem is rep behavior — ramp time, discovery quality, a coaching program with no curriculum — keep Gong and forecast in Salesforce until the miss rate justifies a second vendor. Running both is defensible when the forecast is a board-committed number and call evidence is the primary input to it, which is a $100M+ ARR enterprise-sales pattern, not a Series B one.

Cost reality

Salesforce is the only published number here. Sales Cloud lists at $25 (Starter Suite), $100 (Pro Suite), $175 (Enterprise), $350 (Unlimited) and $550 (Agentforce 1 Sales) per user per month billed annually. A 100-seat Enterprise org is $210K/year list, and it is already in the budget — treat it as the platform you are extending, not a line this project adds.

The three layers on top are all quote-only:

• Gong publishes its shape but not its numbers: licences priced per user, plus a platform fee that scales with the number of users supported. Vendr reports a median annual contract of $54,900 across 1,127 analyzed purchases, a $11,184-$204,033 range, and average savings of 14.17% off the opening quote. Third-party buyer guides place effective per-seat cost at $1,200-$2,400 per year, with the low end reachable only at 100+ seats.
• Clari is custom-quoted with modular bundling across Forecast, Copilot and the Salesloft execution products. Vendr reports a median annual contract of $76,000 across 291 purchases and a $19,065-$415,001 range. Customer-side reports place mid-market deployments at $300-$700 per seat per year, negotiating into $200-$400 at 500 seats on multi-year commits.
• Rattle is custom-quoted. Per Vendr’s February 2026 data across 71 purchases, the median contract lands near $16K/year across a $6K-$71K range, against list pricing of roughly $30-40 per user per month on Professional.

For 100 quota carriers the three layers land at roughly $145K-$175K/year on the median anchors above, on top of Salesforce. A lean configuration — Gong for AEs only, Clari without Copilot, Rattle on a single team — reaches about $85K. An enterprise configuration with both agent layers licensed clears $400K.

The cost nobody budgets is definition work. Stage exit criteria and forecast category definitions have to be written and agreed before Clari is configured, because the roll-up inherits whatever ambiguity is in them. Budget a quarter of a RevOps lead’s time for that, and do it before the implementation kickoff, not during.

Common variations

• Drop Clari; forecast in Gong. One vendor across evidence and commitment, one contract, no boundary document. The swap rule: take it under about 50 quota carriers with a single forecast hierarchy. What you give up is submission history and the audit trail — Gong Forecast will tell you what the number should be; it is weaker at showing who changed the commit, when, and on what argument.

• Drop Gong; run Clari Copilot with Salesloft conversation intelligence. Take it when calls feed the forecast rather than a coaching program, and when Salesloft is already the execution layer, since the July 2026 conversation-intelligence release makes call signal a native input to cadences and plays. What you give up is the Revenue Graph and the Custom Agents built on it.

• Swap Rattle for Slack Sales Elevate or Salesforce Flow. Take it when the hygiene problem is notification-shaped rather than workflow-shaped — reps who need a reminder, not an approval chain. Native is cheaper and improving. Keep Rattle when the fix requires multi-step workflows, approvals, and a HubSpot or Microsoft Teams surface, which sits on its Enterprise tier.

• HubSpot instead of Salesforce. Gong, Clari and Rattle all support it, but the depth argument in this stack — custom objects, opportunity-level write-back, MCP servers against the record graph — is a Salesforce argument. On HubSpot, evaluate native forecasting plus Gong before adding a third vendor.

• Cheaper forecasting specialists: Aviso or BoostUp, now Terret. Evaluate them when the six-figure Clari line is the blocker rather than the capability. The rule: run the same three deals through each vendor’s inspection view during the trial and compare what each one flags against what actually happened last quarter.

What this stack does NOT replace

• Stage exit criteria. Every risk flag and category rule reads the definitions you wrote. If Stage 4 means five different things across five managers, the forecast is wrong at higher cost. See deal stage definitions and forecast categories explained.
• The CRO’s judgment. These tools produce a defensible number and the argument behind it. Committing to it is still a person’s decision.
• Sales methodology. Nothing here improves discovery. Gong shows you that discovery was thin; the fix is a program, not a licence.
• Quota and capacity planning. The target the roll-up is measured against comes from somewhere else — see quota coverage and the RevOps planning stack.
• Finance’s revenue recognition. A Clari commit is a sales number. It is not ASC 606 revenue, and treating the two as interchangeable in a board deck is a mistake auditors catch.
• Data ownership. Rattle enforces a hygiene standard; somebody still has to write it. Start from CRM hygiene and name the fields that are load-bearing for the forecast before you build a single workflow.

Match rules

Use this stack when:

• You carry 25+ quota carriers and the quarterly commit is a number the board holds you to. Both halves matter — headcount alone justifies conversation intelligence, not a forecasting layer.
• You have missed the commit by more than 10% twice in a row. That is the trigger event to buy on, and it is also the business case that gets the budget approved.
• Deals run 3+ months with multi-threaded buying committees. Call evidence is worth paying for when the deal is long enough for the story to change mid-cycle; in a 30-day cycle it is not.
• Someone owns the forecast definition. A stack with two revenue-AI vendors and no named owner regresses to a spreadsheet within two quarters, and now it is an expensive spreadsheet.

Do not use this stack when:

• You are under $10M ARR with fewer than 15 reps. Salesforce reports plus a weekly pipeline review get you inside 10% — see forecast accuracy for what to measure first.
• Deals close in under 30 days at high volume. Your forecast is a conversion-math problem, and pipeline velocity instrumentation beats deal-level inspection.
• Revenue is majority self-serve. Product usage, not call signal, is the leading indicator, and the money belongs in a product-led motion instead.
• Stage exit criteria are unwritten. Buy nothing until they exist. Configuring Clari on ambiguous stages produces a confident number built on a definition nobody agreed to, which is worse than the honest guess it replaced.