YanflowBook a 15-min call
Work
Case study · Autonomous Agents

$2.4M recovered across fiscal year — with one pricing agent

A US-based Series C HR-tech SaaS, $40M ARR. Fourteen tier changes. Zero new headcount.

Yanflow8 min read
Revenue recovered$0.0M
Time to production0 weeks
New analyst hours0
Segments touched0 tiers

The company had done everything right on the product side. They'd grown from $4M to $40M in three years, built a category, and hired a real sales team. What they hadn't done was touch pricing.

Their list price hadn't moved in 18 months. Discounting was by feel — each AE negotiating against a private rubric of what a given size of company "usually" paid. The CFO could see, quarter after quarter, that deals were landing 8–12% below list. He didn't have the tooling to prove why or to fix it.

The CFO could see deals were landing 8–12% below list. He didn't have the tooling to prove why.

What we built

The decision loop
Signalsdata in
ScorerML model
Policybusiness rules
Human guardrailCFO review

We didn't replace the sales team. We gave them a second opinion. Every inbound deal gets scored against thirty signals — company size, vertical, contract length, competitor presence, seasonality, renewal vs new, and a half-dozen more that only mattered because we had their closed-deal data.

The agent writes a suggested price for each tier, with a confidence interval, and three sentences of reasoning. The AE sees it inside their existing CRM. They can accept, override, or reject — and every decision trains the next iteration.

The hard part wasn't the model. It was the integration: their billing system had twelve years of drift and five definitions of "tier." Two of our five weeks were spent cleaning the truth into one table.

What we'd do differently

We shipped the CFO's veto button on day one. In retrospect we should have shipped the AE's veto button on day one too. For the first three weeks, AEs assumed the agent was management watching them. Once we renamed the UI from "agent recommendation" to "second opinion" and made the override frictionless, acceptance jumped from 61% to 89%.

The receipts

  • 14 tiers moved in the first quarter live
  • $2.4M in recovered revenue across the fiscal year, measured against counterfactual list price on identical deal characteristics
  • Zero new hires on the revenue operations team
  • CFO now reviews the agent's weekly margin summary instead of running the model himself

We don't publish the customer name. We're happy to introduce you on the call.

Build brief

  1. 01DiscoveryWk 1

    Audit billing data, cohort segments, win/loss logs. Interview two CS reps.

  2. 02ArchitectureWk 2

    Design the decision loop: signals → scorer → policy → human guardrail.

  3. 03BuildWk 3–4

    Ship the pricing engine, shadow-running against last quarter's deals.

  4. 04Go liveWk 5

    Turn it on for two tiers. Two founders in the room. CFO veto.

  5. 05CompoundWk 6+

    Expand to all 14 tiers. Agent retrains weekly on closed-deal data.