AI won't fix broken processes. It makes them faster.
AI can't rescue bad workflows. If your returns policy is clunky, replenishment logic is off, or your warehouse layout fights demand patterns, AI will speed up poor decisions and amplify bad data.
Value comes when insight connects to execution. That's where outcomes change - not in dashboards, but in how merch, fulfillment, transportation, and inventory actually move.
Start with outcomes, not an "AI initiative"
The question to ask: what business result should change? Less waste. Fewer stockouts. Faster put-back on returns. Better on-shelf availability without bloating inventory.
Andrea Morgan-Vandome, SVP and chief innovation officer at Blue Yonder, puts it plainly: "retailers have to change their operational procedures. Otherwise you're using AI to get more bad data."
Returns: fix the flow, then apply intelligence
Returns aren't just a processing speed problem; they're a workflow problem. The goal is to move goods back to sellable inventory faster, cut dwell time in transportation, and capture reason codes at the source.
"As you do those returns and now that you know why it is being returned that data can be used to improve processes," Morgan-Vandome said. That insight only matters if it feeds back into planning, merchandising, fulfillment, or transportation - automatically.
From insights to execution (in real time)
"AI is also an enabler of helping people bring together end-to-end processes. And you can drive big value out of it," said Morgan-Vandome. That means linking demand sensing to replenishment and store execution so decisions don't stall in analysis.
Example: a grocer reduced waste and protected on-shelf availability, not by "better forecasts" alone, but by coordinating forecast signals with replenishment rules and the execution windows that actually exist in stores and DCs.
Warehouses: stop waiting for the next redesign
Layout is no longer a once-a-year project. With AI agents, slotting and flow can be evaluated continuously against real demand and labor constraints.
"In a warehouse, the issue is how to lay out the warehouse. Previously that was done at specific time periods. With an AI agent, you can start to look at how to lay out the warehouse in the most efficient way and the most efficient way may be store-ready pallets." Insight is useful only when tied to actions WMS can execute.
Speed beats perfection
Long transformations have their place. But the wins are coming from teams that act fast on high-impact decisions and let AI rank what matters first.
Retailers are proving you don't need perfect data. Short proofs of concept show which inputs truly constrain performance. In one three-month test, a retailer lifted sales and cut out-of-stocks using the same inventory - by changing how decisions were made. "That's where AI helps prioritize what to fix," Morgan-Vandome said.
Embedded beats bolt-on
The most effective AI is often invisible. "Our solutions are built from the bottom up with AI," Morgan-Vandome said. It's embedded in the workflow, not a separate tool that analyzes data off to the side.
Pulling data out, analyzing it elsewhere, then pushing recommendations back creates lag and adoption friction. "Where we make it part of the solution is how we apply it to change operational procedures and give it to planning."
The operator's playbook: connect insight to action
- Define the outcome. Pick one: reduce waste, cut stockouts, compress put-back time, improve on-shelf availability. Set a measurable target and a deadline.
- Map the decision chain. For that outcome, list the 5-7 decisions that move the needle (e.g., safety stock rules, promotion timing, vendor MOQ/lead time, return disposition, slotting rules).
- Fix returns at the source. Capture reason codes at scan-in, set disposition rules (reshelve, rework, liquidate), and track put-back SLA. Route insights to planning and merchandising automatically.
- Close the loop. Turn insights into triggers your systems can execute: OMS for order holds, WMS for slotting and waves, TMS for consolidation, planogram changes for stores.
- Pilot fast, scope small. Pick one category or DC-region pair. Run a 60-90 day test. Freeze other variables. Measure OSA, waste, dwell time, and cycle time.
- Be data-pragmatic. Identify the 3-5 fields that change outcomes (e.g., promo calendar accuracy, vendor lead time, return reason). Clean those first; leave the rest for later.
- Use agents for execution. Let AI propose and stage actions (re-slot picks, generate store-ready pallets, adjust order quantities). Keep human-in-the-loop thresholds for risk.
- Instrument everything. Track decision latency, exception rates, and time-to-value. Review weekly. If a recommendation isn't executed, fix the handoff - not the model.
- Governance with teeth. Assign owners for data, models, and decisions. Define rollback plans. Document who can override what, and when.
From innovation to adoption
Morgan-Vandome's remit spans innovation and supply chain advisory, with a clear tilt toward adoption. Teams stay engaged through implementation so operators know not just how to use the tech, but why it exists, and what decision it should change.
Everything tracks back to a business outcome. "Because if what we are doing doesn't actually change the outcome, why would you do it?"
Bottom line
AI amplifies whatever you feed it. With clear targets and process change, it compounds value. Without that, it compounds noise.
Need to upskill your ops team on practical AI?
Browse role-based learning paths at Complete AI Training to accelerate adoption where it counts - in execution.
Your membership also unlocks:
