Start With the Workflow, Not the AI Model
Evaluating supply chain AI solutions?
Focusing on model selection and token burn is a mistake.
Efficiency is pointless if you're driving in the wrong direction.
Start with the workflow, not the solution.
Collaborate with the teams living the pain to understand what each workflow is meant to accomplish, and where it breaks down.
A few questions to get you started:
Why does the work exist?
Killing unnecessary tasks is a high-leverage move, so delete before optimizing. Understand what it’s meant to accomplish, and for whom. Ask the teams that depend on the output what they'd miss if it disappeared.
Is ownership clear?
Visibility and accountability gaps cluster at handoffs. When information is scattered and coordination is manual, the next handoff is often short on details. This risk compounds without ownership. Every open item should have a clear owner who’s accountable for closing the loop. Without one, the item drifts until it’s on fire.
Where does work wait?
In supply chain execution, information rapidly loses value because it’s tied to a narrow decision window, yet most of the timeline is spent waiting, not working. Perfectly synchronized data across partners and systems isn't achievable, and isn't the goal. Eliminate the overhead of closing gaps (a delayed PO confirmation, a missing ETA). These are expensive bottlenecks, and the best place to start with automation or AI.
Can the inputs be trusted?
Supply chains have a data problem, and AI won’t fix it for you. To reconcile information across systems and trading partners, you need shared vocabulary, clear relationships, and rules that hold up. Without this, you're moving the problem around. AI will inherit and amplify the ambiguity in your data.
When the workflow breaks, what fails?
The blast radius of supply chain issues varies wildly, so triage is critical. Find the issues creating the most risk and the key decisions that drive results. Eliminate the low-value work around these decisions, and keep them in human hands with the context, time, and authority to act.
The real opportunity lives in the answers to these questions. You'll know what to pilot, what to automate (AI or otherwise), and what to leave in human hands.
