The product produced accurate transcripts, summaries and topic tags. Managers opened it twice a month. Nothing in it told anyone what to do next.
STUDY 09 / 30 · ANONYMISED · NDA · SALES AUTOMATION · US · SERIES A
SALES AUTOMATION · US Call intelligence that produces a decision, not a transcript.
RoleLead designer
Timeline9 weeks
Team1 designer, 1 PM, 1 ML engineer, 1 frontend
VerticalRevenue intelligence platform · mid-market sales teams
Replaced the summary with a decision card: the single most consequential thing said, what it implies for the deal, and the one action it warrants.
Weekly active use among managers rose sharply. The product moved from a record-keeping tool to a deal-review instrument.
81%MANAGER WAUfrom 14%
2.4 / callACTIONS TAKEN FROM CARDfrom 0.1
19 daysDEALS FLAGGED AT RISK EARLIERmedian lead time
THE ARGUMENT
Why the obvious solution was wrong.
The study matters because the product problem was reframed before the interface was polished.
Conversation intelligence has an output problem, not an accuracy problem. The transcripts are correct, the summaries are faithful, the topic tags are precise - and none of that changes what a manager does on Monday. Producing a summary of a call is producing another thing to read. A manager with fourteen reps does not need more reading; they need to know which three calls this week contain something that changes a forecast.
The decision card names the single most consequential moment in the call, quotes it, states what it implies, and proposes one action with a one-click path to take it. Everything else - transcript, full summary, topics - remains available and stops being the default surface. The hardest design work was restraint: the model can extract twenty interesting things per call and showing twenty is the same as showing none. One is a decision. Twenty is homework.
THE INTERFACE CRAFT
The interaction, rendered as a working product surface.
The specimen below is code-native and uses the study's own design logic. The client interface remains protected.
SALES AUTOMATION
BUYING GROUP
DESIGN DECISIONS
Positions we would defend.
Each decision names the principle and the product consequence, not a stylistic preference.
Extract one thing
A model that can find twenty insights per call must be constrained to surface one. Ranking is the product; extraction is the commodity.
Quote, never paraphrase
A paraphrased insight requires verification against the transcript. A quote with a timestamp does not.
Default to the decision
A transcript-first interface is a tool for the archive. A decision-first interface is a tool for Monday.
PRODUCT LEADER READOUT
What transfers, and what should remain specific to this product.
A case study is useful when its operating principle travels without turning the original interface into a template.
Read the operating condition
For Revenue intelligence platform · mid-market sales teams, the transferable lesson is not a copied screen. It is the condition the interface had to make legible: The single most consequential utterance in the call, quoted exactly with timestamp and one line of implication. Not a summary of the call. Rebuild that visibility for your own roles, risk, terminology, and operating cadence.
Protect the design rule
A model that can find twenty insights per call must be constrained to surface one. Ranking is the product; extraction is the commodity. Keep that rule in the acceptance criteria, component states, and production QA record so later visual cleanup cannot erase why the interaction exists.
Measure behaviour after ship
The evidence record is 81% for manager wau, from 14%. Recreate the baseline and outcome window before rollout, segment the result by role and context, and state clearly what the measure cannot prove.
RESEARCH RECORD
The work behind the interface.
These artefacts connect the final interaction back to the evidence and product model that produced it.
Manager shadowing
Shadowed six sales managers through a full deal-review cycle to find where a call insight would actually have changed a decision.
Consequence ranking model
Built and tuned a ranking for utterance consequence with the ML team, validated against manager judgement on 400 calls.
One-vs-many test
Tested single-insight cards against five-insight cards for action rate; single won by more than 4x.
Risk lead-time analysis
Measured how many days earlier a card-flagged risk appeared versus the manual pipeline review that had caught it.
“It went from a thing we bought to a thing we run the pipeline meeting from.”
VP Revenue Operations, mid-market SaaS · under NDA
NEXT STUDY · 10 / 30 · MARTECH