PLAYBOOK

How to Build a Message Performance Agent From Sales Calls and Deal Outcomes

A step-by-step playbook for finding out which of your messages land with buyers, using the sales calls and deal outcomes you already have.

Try Rippit free
or request demo
By Lauren Alexander, VP of Marketing at Rippit · Published November 19, 2026

A message performance agent reads your sales calls, finds each claim your team made, records how the buyer reacted and links it to what happened to the deal. Each week it tells product marketing and enablement which messages buyers lean into, which they question and which calls show it. It can’t prove that a message caused a win; for that, turn the strongest patterns into tests.

Your team believes a claim is working because it shows up in every deck. The calls tell you something more useful: whether buyers repeat it back, ask “how, exactly?” or move straight to price. That reaction is evidence you already have, sitting in recorded calls next to the opportunity they belong to.

Key takeaways

  • Track claims, not calls. One call can contain five claims and five different reactions.
  • Record the buyer’s next words, but don’t treat silence as agreement.
  • Compare like with like. Segment, stage and seller shape outcomes more than any one message.
  • Treat patterns as test candidates, not proof that a message works.
  • Keep it running weekly so launches and new objections show up while they still matter.

How do you build a message performance agent in 6 steps?

List the claims, record each time one is said, capture the reaction, compare fairly, review with people, then test and keep it running.

Step 1: List the claims you want to track

Write down the claims, proof points and value messages your team is meant to use, in plain language, with a short example of each. Version the list: when a launch adds or retires a claim, the analysis should know which version applied to which call.

Step 2: Record one row per message moment

Use one record per message exposure, not one row per call:

What one message moment records
FieldWhat it holds
Call and opportunityThe call, the account and the Salesforce opportunity it belongs to
ClaimThe normalized claim from your list, and the list version
Exact wordsWhat the seller actually said
Speaker and roleWho said it, seller or buyer
Buyer reactionThe buyer’s next words, if any
Stage at the timeThe deal stage when the claim was made
Later outcomeWon, lost, no decision or still open
Coverage and confidenceWhether the call was fully transcribed and how clear the match is

Separate the exact language from the normalized message label. Store the next buyer reaction when available, but do not assume silence means agreement.

In Rippit: Each field is an AI column on your Gong calls, with a choice list for the claim and the reaction. A lookup adds the stage and outcome from Salesforce.

Step 3: Compare like with like

Deal outcomes are affected by segment, seller, stage, product, price, competition, timing and many other factors. Compare messages within reasonably similar populations and report coverage and missingness.

Useful descriptive findings include:

  • a claim is common but often followed by clarification requests;
  • a proof point appears more often in a defined won set than a comparable lost set;
  • a message performs differently by persona or stage;
  • objection language changes after a launch.

These are test candidates, not causal conclusions. In this walkthrough, Dev Kapoor, Director of Product Marketing, checks four claims from a recent launch against six weeks of sales calls.

Six weeks of calls, read claim by claim, compared by persona and posted to #product-marketing every Monday.
Weekly message summary · #product-marketing4 launch claims · last 6 weeks · 112 calls on 46 opportunities
  • “Live in a day, without IT”41 callsTest next
  • “Every report updates itself”33 callsTest next
  • “Replaces your spreadsheet tracker”29 callsBuyers question it
  • “Enterprise-grade permissions”9 callsNot enough calls

Why “Live in a day, without IT” is worth testing

Message
The seller says: “You can be live in a day, without waiting on IT.”
Buyer response
On 26 of the 41 calls, the buyer’s next words repeat the point or ask what setup involves: “So our ops lead could set this up herself?” On 6, the buyer asks about a security review instead.
Deal context
Said mostly in mid-market discovery calls, by 7 different sellers. It came up in 11 of 16 won mid-market deals and 5 of 14 lost ones of similar size and stage.
What it can’t tell you
Whether the claim helped win, or whether simpler deals made it easier to say.
Next test
For four weeks, half the mid-market team opens discovery with the claim and a two-minute setup walkthrough. Compare how many opportunities reach evaluation within 30 days against the other half.
From four launch claims to the two worth testing next, with the calls behind each one.

The same finding sends two teams different ways. Sales enablement keeps the line in the mid-market talk track while the test runs. Product marketing takes the six security-review questions to whoever owns the security FAQ.

Step 4: Review findings with people

Product marketing should review taxonomy drift, ambiguous exposures and high-impact findings. Sales enablement should validate whether the message was delivered as intended. RevOps should approve opportunity joins and outcome definitions.

Protect customer confidentiality and exclude calls or uses that fall outside approved consent and governance.

Step 5: Turn observations into experiments

Use the agent to select a message and audience, then evaluate it through an approved test: randomized where feasible, or a carefully designed phased or matched comparison. Predefine the outcome and window.

Feed results back into the reusable message model so the agent distinguishes observed patterns from tested effects.

Step 6: Turn it into a weekly agent

Run the analysis every week on new calls, and after every launch, so changes in buyer reactions show up while they still matter.

Ask Rippit

Every Monday, for sales calls from last week, tag each of our launch claims the seller made, the buyer’s reaction and the deal stage. Compare reactions by persona with call counts. Post a summary to #product-marketing with the claims worth testing next, the claims buyers question most and links to the calls.

In Rippit: Rippit builds it as a pipeline: AI columns on each call, a rollup by claim and persona, and a Slack action. Test it on a sample and check the cost before you put it live.

Where does Rippit fit, and where doesn’t it?

Rippit reads every call next to its deal and keeps the analysis running. It doesn’t decide which message is true or replace a real test.

Rippit brings in sales calls from Gong and other call recorders and Salesforce opportunities as tables, adds the message columns your CRM doesn’t have and links every value back to the call behind it.

Where it doesn’t fit: it won’t prove that a message caused a win, it doesn’t replace message testing, and it reads only the calls you connect and are permitted to analyze. Treat the output as a set of well-sourced hypotheses.

Which agents should you run next?

Connect message performance to win/loss reasons, positioning gaps and competitor claims.

Frequently asked questions

Can a message performance agent prove a message increases win rate?

No. Calls show which claims were made and how buyers reacted, but many other factors shape a deal. Use the patterns to pick messages to test, then measure the test.

How many calls do I need?

There is no fixed number. You need enough calls per claim and segment to compare, with the call counts and coverage shown next to every finding.

Which calls should be excluded?

Exclude calls outside your consent and governance rules, calls without a usable transcript and internal calls. Report how many were excluded so readers know what the counts cover.

Can I use calls from Gong?

Yes. Rippit’s Gong data integration brings calls in as a table, read-only, and Salesforce opportunities join them on the account or opportunity ID.

What is Rippit?

Rippit is a Growth Data Platform that helps you and your team build your edge. Anyone can build AI agents on all of their customer and business data, with no engineer needed. Read every sales call next to its Salesforce opportunity, and see which claims buyers lean into, question or ignore. Keep the analysis running for your team, in Rippit or in the AI assistant you already use through our MCP server.

For this agent

Reads
Gong sales calls matched to Salesforce opportunities
Compares
How buyers respond to each launch claim, grouped by persona and deal stage
Delivers
A weekly Slack brief for Product Marketing with the claims worth testing and links to the calls behind them

Build your first agent on your own customer and business data.

Start free with $10 in credits. No credit card or engineer needed.

Try Rippit free
or request demo

Keep reading

More from the Rippit Library.

View the library →