“AI-enabled” has become a default descriptor across sectors. In many cases it is directionally true. The risk for investors is not that companies mention AI, but that the claim is priced as structural advantage before operational evidence exists.
Narrative inflation happens when AI language expands faster than measurable model impact. Teams present margin expansion, sales velocity, or retention uplift while the forecast still behaves like the pre-AI operating model.
Checklist: from language to evidence
Ask whether AI affects the product, the workflow, or the cost structure. Product-level claims should map to measurable usage and willingness-to-pay. Workflow claims should map to cycle-time or conversion changes. Cost claims should map to staffing or infrastructure deltas.
If the model shows none of these adjustments, the claim may still be strategic direction but not current underwriting evidence.
Common inflation patterns
Pattern one: “AI moat” with no data advantage economics in the model. Pattern two: “AI efficiency” with no reduction in service cost or support burden. Pattern three: “AI upsell” with flat ARPU and unchanged expansion assumptions.
These are not automatic red flags. They are prompts to reclassify confidence from base-case to hypothesis until evidence appears.
Practical IC framing
Separate AI claims into now/next/later buckets. Underwrite only “now.” Track “next” with explicit milestones. Treat “later” as option value. This protects upside while preventing narrative enthusiasm from distorting risk pricing.
Vendor spend and gross margin tell their own story
If AI is central to delivery, you should usually see movement in hosting, API spend or support efficiency over time—not always immediately, but on a timeline the company proposes. Flat gross margin with heroic AI language invites a simple question: where did the economic benefit go?
Sometimes the answer is reinvestment into R&D, which is fine if modelled. Sometimes the answer is that AI is still experimental, which is also fine if priced as such. The failure mode is silent conflation of roadmap slides with current COGS structure.
Customer-visible AI versus internal tooling
Internal copilots can improve employee productivity without moving customer-facing NPS or pricing. Customer-facing AI features may change conversion or ticket volume. Ask which class of claim is being made and which metric should move first if the claim is true.
Regulatory and procurement reality
In regulated contexts, AI claims may bump into procurement, legal review and customer security questionnaires. Roadmaps can be right while near-term revenue impact is nil. Underwrite the horizon where procurement friction clears, not only the horizon where demos look impressive.
If customers must approve model use contractually, pipeline stages should reflect that gate. Skipping it in the model while mentioning enterprise logos in the deck is another form of narrative inflation—enterprise interest is not enterprise deployment.