19/05/2026
Most operators are spending on AI that won't compound.
We audited 12 deployments across our practice over 22 months. Every deliverable got one of three labels: COMPOUNDING, LINEAR, or DEPRECIATING.
The pattern was clean.
Compounded every time:
· Eval harnesses (5/5)
· Multi-tenant platforms (3/3)
· Customer knowledge layers (2/2)
· One-page strategic artifacts (4/4)
Depreciated within 9 months:
· Single-purpose AI features without harnesses
· Persona tools (4/4 abandoned within 6 months)
· Vendor-locked AI subscriptions
The three traits compounding deliverables share: INFRASTRUCTURE, SHARED, PROVENANCE.
Three questions to run on any AI investment before you commit:
1. Will it be infrastructure or a feature?
2. Shared or single-purpose?
3. Will it have provenance or be a black box?
Two or more yeses tends to compound. Any zero tends to depreciate.
We've built a 1-page checklist with 9 specific questions you can run on any AI investment under $50k.
Comment COMPOUND below and we'll send it. No newsletter. No sales sequence. Just the checklist.