Every vendor will tell you what AI can do. Nobody wants to talk about what it can’t.
So let’s talk about it.
The Honest Assessment
AI is exceptional at:
- Pattern recognition across massive data sets
- Consistent application of defined rules
- Content generation at scale
- Real-time response to defined triggers
- 24/7 monitoring without fatigue
AI is poor at:
- Making judgment calls with incomplete information
- Understanding context that isn’t in the data
- Managing relationships and stakeholder trust
- Recognizing when the rules have changed
- Handling genuinely novel situations
Commerce sits at the intersection of both. That’s why the best teams aren’t replacing humans with AI — they’re designing systems where each does what it’s best at.
What Humans Must Own
1. Strategic Framing
AI can tell you what’s happening. It cannot tell you what matters.
“What should our category strategy be for the next 18 months?” is a human question. The data informs it. The judgment is human.
2. Relationship Context
A key account is struggling. The data shows they’re ordering less. The AI can’t know that the buyer’s mother is in the hospital, or that they had a bad experience with your support team last month, or that they’re quietly evaluating your competitor because of a LinkedIn post they saw.
The human managing that relationship has context that the algorithm doesn’t and probably can’t.
3. Exception Handling
When the rules change, humans decide what to do. When a global event disrupts supply chains, when a regulatory change invalidates your pricing strategy, when a PR crisis emerges — these aren’t edge cases for AI to handle. They’re the moments where human judgment is most critical.
AI handles the steady state. Humans handle the exceptions.
4. Ethical Choices
AI will optimize for the metric you’ve given it. If your metric is conversion rate, AI will increase conversion rate — even if the strategy involves dark patterns, misleading copy, or predatory targeting.
These choices require human accountability. The AI doesn’t care. You have to.
The Right Division of Labor
The best commerce operations look like this:
| Task | Who Does It |
|---|---|
| Monitoring competitor prices | AI — always |
| Deciding repricing strategy for commodity goods | AI — with human-defined rules and floors |
| Analyzing why conversion rate changed | AI + Human — together |
| Deciding to enter a new category | Human — with AI-generated analysis |
| Writing product descriptions | AI — human edits and approves |
| Managing key account relationships | Human — AI provides data support |
| Responding to a PR crisis | Human — AI monitors and informs |
| Setting pricing strategy for premium positioning | Human — with AI competitive analysis |
The Risk of Getting This Wrong
The failure mode isn’t “AI does everything and fails.” The failure mode is more subtle:
Human over-reliance — teams stop developing judgment because AI handles the steady state, then they can’t handle exceptions when they arise.
AI over-reliance — teams abdicate judgment to algorithms because it’s easier than thinking, then find themselves unable to make decisions when the algorithm doesn’t have an answer.
Both are bad. Both are common.
Building for the Partnership
The practical framework:
- Define clear roles — which decisions are AI-first, which are human-first, which are collaborative
- Maintain human expertise — don’t let judgment atrophy; require humans to practice it
- Instrument the partnership — track where AI is succeeding and where human intervention is required
- Review the division — quarterly, check if the role assignment still makes sense as capabilities evolve
The commerce operations winning over the next decade aren’t the ones replacing humans with AI. They’re the ones building human-AI systems that leverage the strengths of each.
That’s a harder design problem. It’s also the only one that actually works.
If you’re thinking through the human-AI division of labor for your commerce operation, that’s worth a focused conversation. Every team structure is different.