All use cases
AI Agent Automation2026-05-05

AI agents that don't just suggest — they do the work

A footwear group deployed Solya agents on markdown, transfers and reordering. Approvals stayed human; the busywork did not.

Outcome

+7% margin uplift

Network

5 banners · 50+ stores

Measured outcomes

+7%

Margin uplift · first end-of-season cycle

Days

Markdown decisions executed (was weeks)

100%

Audit trail on every action — who, what, when

What's wired up

Systems connected

MERCH

Merchandising system

Pricing & assortment

WMS

Warehouse (WMS)

Inter-store transfers

STORE

Store-facing tools

Tags & shelf-labels

Before · After

Before

Playbook on paper, scattered execution

Five banners, 50+ stores, a clear markdown playbook — and nobody actually following it consistently. Slow-moving stock was marked down too late, too uniformly, and the margin impact was significant.

After

Agents draft, humans approve

Each agent runs on its own cadence, builds a full proposal (rule followed, data used, expected impact) and routes it for approval at the team's existing thresholds. Approved actions execute directly — no re-keying into a workflow tool.

The challenge

The merchandising team had a clear markdown playbook on paper, but executing it across five banners and 50+ stores meant nobody actually followed it consistently. Slow-moving stock was being marked down too late, too uniformly, and the margin impact was significant.

What we changed

Solya agents were given the team's markdown rules, transfer guardrails and reorder logic, all formalised inside the intelligence layer — and the keys to the systems that execute them. Each agent runs on its own cadence, drafts the decisions, and routes them for approval based on the team's existing thresholds.

How decisions get made

An agent doesn't just produce a recommendation list. It builds a full proposal with the rule it followed, the data it used and the expected impact. Reviewers approve, edit or reject — and the agent learns which adjustments are systematic vs. one-off.

Where it lands

Approved actions hit the merchandising system, the warehouse, and store-facing tools directly. Nothing gets re-keyed into a separate workflow tool — the agent's output IS the workflow.

What changed

  • +7% margin uplift on the first end-of-season cycle
  • Markdown decisions executed in days, not weeks
  • Full audit trail on every action — who approved, what changed, what the agent saw

Related: pair this with continuous replenishment to close the loop upstream, or see how the same orchestration layer handles execution across systems without re-keying.

More use cases

From the blog