Dipp AIDipp AI

Case study · Retail

A retailer ran 240 concurrent pricing agents without a single runaway loop

Concurrency and recursion limits turned an experiment that had twice run away into a production system with a predictable bill.

Omnichannel retailer · 1,900 stores · published July 29, 2026

Download the PDFApply for early access

240

concurrent agents under a single enforced ceiling

0

runaway loops since deployment

two in the preceding six months

$0.11

cost per SKU-region pricing decision

from $0.74, unbounded

11 hrs → 0

detection lag on out-of-policy prices

now refused pre-commit

Summary

A retailer's markdown-optimisation agents had twice entered recursive loops that consumed a month's budget in a weekend. Orcher capped concurrency and recursion depth in the execution path, verified each price change against margin floors, and made the cost of a pricing cycle predictable enough to plan around.

Orcher components enforced

  • Cost Governance

    Routes by stakes and halts runaway loops.

  • Logic Scrubber

    Verifies the proposed action against systems of record before commit.

  • Observability

    Real-time cross-provider trace: which model, which role, what cost, what outcome.

Solution: Cost discipline & intelligent routing

The situation

What was happening before Orcher.

  • Markdown agents negotiated with demand-forecast agents, and the interaction had no depth limit.
  • Two incidents in six months consumed the quarterly budget over a weekend with nothing committed.
  • A pricing error committed at 02:00 on a Sunday took eleven hours to detect.

What Orcher enforced

The control surface, component by component.

  • Concurrency and recursion depth became enforced properties of the directive class, not conventions in application code.
  • Logic Scrubber verified each proposed price against margin floors, contractual price commitments and regional rules before commit.
  • Cost per pricing cycle was bounded up front, so the finance question was answered before the run rather than after it.
  • Halts and refusals were recorded as outcomes, giving the pricing team a feedback loop instead of a mystery.

The data

Measured over the engagement window.

ControlSettingTriggered in periodEffect
Concurrency ceiling240 agents17 timesQueued, none dropped
Recursion depth6 hops31 timesHalted and logged
Margin floor checkPer SKU-region1,204 refusalsPrices never committed
Contract price lockPer supplier agreement88 refusalsBreach avoided
Cycle cost ceilingPer pricing run3 timesRun completed at ceiling

Enforced control activity over one full markdown season.

Outcomes

What the organisation did next.

  • Markdown optimisation moved from a supervised experiment to a scheduled production process.
  • The pricing team plans agent cost as a line item with the same confidence as compute.
  • Refusals surfaced 88 supplier contract terms that the legacy pricing engine had been quietly violating.
  • The retailer applied the same ceilings to replenishment agents in the next season.
“Two runaways is a story you tell once. The third one ends the programme, so the limit had to live somewhere nobody could edit in a hurry.”
— Director of Pricing Science, omnichannel retailer (partner declined attribution by name)

Read further

The research behind this engagement.

Disclosure. Design-partner engagement, anonymised at the partner's request. Figures are measured by Orcher's own observability and ledger instrumentation over the stated period and have not been independently audited. Market figures carry their own source line.

More engagements

Other organisations that had to prove the same thing.

Run this pattern on your own workflow

Design partners deploy Orcher on one directive class, instrument it the way this study was instrumented, and keep the evidence whatever they decide afterwards.

Apply for early accessAll case studies