Why Your Sales Forecast Can Still Miss With Clari or Gong
Every call on the deal was recorded, the commit was signed off on Thursday, and the deal still slipped out of the quarter in its last two weeks.
The deal that sat in commit until the last two weeks
Picture a $640K deal at a mid-size security company. Six calls are recorded in Gong and the summaries are accurate. The rep marks it commit in Clari, the manager accepts it in the roll-up, and it stays there for five weeks. Then, two weeks before quarter end, procurement mentions that the budget now sits with a new CFO nobody on the deal has met.
Go back through the recordings and the signal was there. On call three, the VP running the evaluation said the CFO would "want to see it before anything gets signed." The summary logged it as a next step. The economic buyer field in the CRM still named the VP.
Neither tool failed at its job. Gong captured the sentence and Clari rolled up what the rep believed. The forecast was wrong in the space between them, where someone has to turn a conversation into a judgment about whether the deal is qualified, and that judgment rested on summaries read late on a Thursday.
What Gong and Clari do well
Gong is very good at capture. Calls are recorded, transcribed, attributed to speakers and searchable. If a buyer said it on a call, you can find the sentence and the timestamp.
Clari built the forecast cadence that a lot of revenue teams still run on: the weekly roll-up, rep calls next to manager calls, one place where the number gets submitted. Clari is now merging into Salesloft, and the combined company goes to market as Salesloft. The cadence it made standard will outlast the brand.
Both tools make the forecast faster to run and easier to audit. Neither decides whether a deal meets your qualification bar. That judgment still comes from somewhere, and where it comes from explains most forecast misses.
📊 Two-thirds of RevOps teams have seen an AI use case stall or get rolled back. — Vasco, 2026 State of AI in RevOps |
What actually feeds the number
Trace any deal in your commit back to its inputs and you usually find four. Two of them are the rep's opinion, one is a compressed account of a single call, and one proves contact happened without saying who was in it.
Input | Who produces it | What it tells you | What it can't tell you |
|---|---|---|---|
CRM qualification fields | The rep, usually after the call | What the rep believes about the deal | Whether the buyer said it |
Forecast category | The rep, then the manager | How confident the team feels | Why, in the buyer's words |
Call summary | A general model, one call at a time | What was discussed on that call | Whether it held up over the next five |
Activity data | Email and calendar logs | That contact is happening | Which member of the buying group is missing |
Every row is useful. None of them is evidence that a qualification criterion is met. A forecast built on all four inherits the rep's optimism and the summary's blind spots at once.
Why summaries make a weak forecast input
A summary is written for a person to read, one call at a time. It drops the hedge in the buyer's voice. It turns "we'd probably need legal to look at it" into "next step: legal review," and it has no memory of what the same buyer said three weeks earlier.
Forecast accuracy depends on what a summary throws away: whether an answer holds up across calls, emails and attendance over time. A champion who calls the budget approved on one call and then goes quiet for two has told you something important. No single summary contains it.
📊 The typical buying group for a complex B2B solution involves six to ten decision makers. — Gartner |
With that many people in the buying group, a deal can read as healthy in every call summary and still be missing the person who signs. The gap only shows when someone checks the whole deal against the bar, week after week.
A filled field and a qualified deal
Plenty of teams have wired Gong to a general model that reads each transcript and writes MEDDPICC answers into the CRM. It saves typing. The meaning of the field stays the same, because the model is deciding which box a conversation belongs in, once, and nothing re-checks that answer as the deal moves.
A populated field tells you the box has text in it. Whether the deal still meets the bar in week six is a separate question, and it is the one the forecast depends on.
Watch for the manual step. Teams running this build often keep a person between the model and the CRM to review what it wrote. That review is the team saying it doesn't trust an unsupervised read, and it brings the same late-Thursday judgment straight back into the number.
Putting evidence under the forecast
Spotlight.ai runs on top of Gong. It reads the same recordings plus the buyer email threads on each deal, and turns them into qualification answers through Spotlight's Calibrated-Decision AI-Model (CDAM). Every stored value keeps its source: confirmed by the buyer in conversation, logged by a rep, or calculated from other signals.
CDAM splits each criterion into narrow questions with a closed set of possible answers, and the conclusion about the deal is calculated in code from those answers. Ask about the same deal twice with no new evidence and you get the same answer. When a deal's qualification changes, something in the deal changed. In the $640K example, the CFO's name on call three would have reopened the economic buyer question the week it was said.
The output is a bottom-up forecast where every number traces back to the evidence behind it. The commit call stays the rep's. Now it arrives with the buyer's words attached, so the rep can defend it.
The same evidence serves the rest of the revenue team. Sales gets every deal qualified with MEDDPICC and forecast on what the buyer said. RevOps gets complete deal data and the slippage and velocity patterns behind the number. Value teams get a business case on every deal, built from the buyer's own words. Customer success runs renewals with the rigor that won the deal, carrying what was promised into what was delivered.
Gary Olson, CRO at Sysdig, described the gap this way: "What I didn't have was an ability to integrate and access all the different data points. Spotlight brings it all together and spotlights what needs to happen next." Sysdig's 135-person team saw 3.8x pipeline conversion over 12 months.
An audit you can run on this quarter's commit
Take the five largest deals in commit. Dollar weight matters more than deal count here.
Find the buyer's words behind each qualification field. A call timestamp or an email counts. A rep note with no source does not.
Check the economic buyer by attendance. When did that person last join a meeting or answer an email?
Read the first summary on the deal next to the latest one. Look for a champion, a date or a budget claim that changed and never reached the CRM.
Flag fields written once and never revisited. A discovery-era MEDDPICC answer may not describe the deal in week six.
Count sourced fields per deal. The deals with the fewest are where next quarter's slippage is sitting.
Where to go from here
Run the audit on those five deals before your next forecast call and bring the gaps to the meeting along with the totals. If most fields trace back to something a buyer said or did, your Gong and Clari setup is doing what you need. If most trace back to rep notes and summaries, you know which input to fix first.
Spotlight.ai can run the same trace on last quarter's closed and slipped deals using your Gong recordings, so you can see what the evidence said about each one before the quarter ended. It has done this work across $9B+ of pipeline autonomously qualified and executed.
FAQs About Forecast Accuracy With Clari and Gong
Why is my sales forecast wrong even with Clari or Gong?
Clari rolls up what reps and managers believe, and Gong records what was said. The judgment that turns conversations into a qualified deal still comes from reps reading summaries, so the forecast inherits that opinion.
Does Gong improve forecast accuracy?
Gong gives you an accurate record of every call, which helps when someone checks it. Accuracy improves only when the qualification behind each deal is tied back to that record and re-checked as the deal moves.
Are AI call summaries reliable for forecasting?
Summaries describe one call at a time and drop hedges and history. Forecasting needs to know whether an answer held up across several calls, emails and meetings, which a single summary can't show.
What is the difference between a populated CRM field and a qualified deal?
A populated field means something was written into it. A qualified deal means the buyer confirmed the criterion and it still holds today, which requires evidence and regular re-checking.
Does Spotlight.ai replace Gong or Clari?
Spotlight.ai runs on top of Gong, using its recordings as a source, and adds evidence-based qualification and a bottom-up forecast. Teams keep the recorder they already pay for.
How do I check if my forecast is built on evidence?
Take your five largest commit deals and find the buyer's words behind each qualification field. Fields that trace only to rep notes or summaries mark where the forecast is resting on opinion.
Comments