AI doesn’t fail in CPG because the models are underpowered. It fails because the problem is poorly defined. The winning approach targets narrow, repeatable decisions — not whole departments.
The framework that separates winners from losers
Successful AI applications share three characteristics:
- Decisions that recur hundreds or thousands of times a year.
- Inputs that are unstructured — emails, PDFs, attachments, free text.
- Clear, measurable outcomes with tight feedback loops.
Strip any one of these out and the model has nothing to learn from.
Where this pattern has already played out
Finance and accounting. Vic.ai and AppZen got traction by automating specific decisions — invoice processing, three-way matching, expense auditing — rather than “transforming” the finance function.
Legal. Ironclad and Lexion won by focusing on contract workflows where patterns repeat across thousands of documents.
Operations. AI works in ticket triage, exception handling, and compliance — domains where the same decision shows up again and again.
Why marketing has resisted
Marketing involves subjective calls where even strong teams disagree on what the right answer looks like. Without a narrow, frequent, expensive problem, there’s nothing for AI to lock onto.
What makes CPG different
CPG generates endless repeatable decisions inside messy workflows. Specs change between document versions. Certificates of analysis arrive as inconsistently named attachments. Missing documentation halts production lines. These are the operating reality — not edge cases.
Workflow ownership is the unlock
The organizations that capture AI value observe the same decision repeatedly, capture raw inputs automatically, and learn from actual outcomes rather than stated intentions. “You don’t learn by seeing everything. You learn by seeing the same thing repeatedly.”
Compounding data effects
Once decisions flow through a single system, the loop tightens: more workflows generate more structured data, recommendations improve, outcomes get better, and more work pulls into the system across functional boundaries.
Where to look in CPG
Waystation focuses on supplier and compliance decisions, extracting structure from unstructured documents. DayZero does this for CPG accounting. Jampack AI does it for wholesale distribution. Moselle does it for inventory. Each one owns one repeatable decision.
The recommendation
Don’t implement “AI for marketing” or “AI for operations.” Identify one specific decision made repeatedly using information the team already has but can’t act on. Master that decision. That wedge unlocks the structured data and platform effects that move AI from interesting to inevitable.
