Get Started
Investing in an AI or data company?
Find out what’s proprietary before you pay for it.
Separate real AI advantage from a wrapper on someone else’s model
Diligence of data platforms, analytics products, and AI-native companies.
For data and AI companies, the valuation often rests on a claim of defensibility: proprietary data, better models, or workflows competitors can’t copy. We test that claim. We look at where the data comes from and who owns it, how models are built and monitored, and what it costs to serve each customer as usage grows.
What we find most often
-
Data architecture that grew reactively.
Pipelines built for product-market fit, with every customer’s data in shared stores and no layered separation between raw, cleaned, and served data. We map the fix on the AKF Scale Cube: split pipelines by stage (ingest, enrich, sync) as Y-axis services, and shard data stores by customer cohort (Z-axis) for isolation and performance.
-
Batch jobs that hit a wall.
A single vertically scaled process for large loads, where a big customer can stretch a job from hours to more than a day. We define the threshold at which it must be split, before customers notice.
-
Drift detected too late.
Model quality monitored through lagging signals such as support volume rather than statistical drift detection, so degradation shows up first in customer complaints.
-
Differentiation still being proven.
Patents and models that cover today’s product, while the newer agentic features in the pitch haven’t run at production scale; or AI applied to common use cases a competitor could copy.
-
Unit economics nobody tracks.
Compute and model costs not tied to usage, customers, or products, so margin at scale is unknown.
-
Availability without commitments.
Single-region or single-zone deployments and no customer SLAs.
-
Data rights and separation.
Customer-contributed data used in training without clear contractual rights, or without logical separation between customers.
-
AI knowledge in one head.
Model architecture understood by a single founder or CTO, with no documented succession.