AI Investment Durability Scorecard
Evaluate AI companies by distribution, economics, defensibility, and platform exposure
- Difficulty
- Moderate
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 96%
The scorecard evaluates whether an AI company can preserve value as models become cheaper, open source, and easier to access. Start with distribution: determine whether the company already owns customer relationships or a platform through which it can bundle AI capabilities. Examine economics next, including supplier dependence, retention, churn, and whether rising market demand strengthens margins. Then assess differentiation by looking for proprietary technology, data, infrastructure, network effects, or difficult manufacturing capability. Finally, measure platform exposure: thin applications built on interchangeable APIs face replication and pricing pressure, while infrastructure providers can benefit regardless of which end-user application wins. The framework favors picks-and-shovels suppliers, strong incumbents, and products with defensible distribution over undifferentiated AI wrappers.
Origin
Extracted from Marketing Against The Grain during Matt Wolfe, Kipp Bodnar, and Kieran Flanagan’s discussion of where durable AI investment winners may emerge.
Core principles
- 01Prefer businesses that benefit across multiple AI winners
- 02Treat distribution as a durable strategic advantage
- 03Demand economics that improve as adoption expands
- 04Distinguish proprietary capability from thin API wrappers
- 05Discount products exposed to rapid replication and churn
How to run it
- 1
Map distribution
Determine how the company reaches customers and whether it controls an established platform, channel, or audience. Strong distribution can let an incumbent bundle AI features before a point solution builds equivalent reach.
Pro tip Separate genuine recurring access to customers from temporary launch attention.
Watch out Product quality alone does not guarantee economical customer acquisition.
- 2
Inspect the economics
Evaluate gross margins, infrastructure costs, retention, and dependence on upstream model pricing. Ask whether broader AI adoption improves or weakens the company’s economics.
Pro tip Model what happens if API prices change or a free substitute appears.
Watch out Rapid revenue growth can conceal severe churn and expensive inference costs.
- 3
Identify hard differentiation
Look for proprietary technology, scarce hardware capability, exclusive data, network effects, or workflow depth. Rate how difficult each advantage would be for a capable competitor to reproduce.
Pro tip Require evidence of a moat beyond prompts and user-interface polish.
Watch out Code layered over a widely available API is usually easy to replicate.
- 4
Measure platform dependency
List the models, clouds, APIs, and marketplaces the product relies upon. Determine whether those suppliers can raise prices, copy the feature, or bundle it into their own platform.
Pro tip Favor companies that benefit from multiple competing model ecosystems.
Watch out A single upstream dependency can erase pricing power or product access.
- 5
Assess substitution and churn
Estimate how easily customers can switch to a cheaper, free, or incumbent-provided alternative. Products embedded deeply in valuable workflows should score better than novelty tools with shallow usage.
Pro tip Look for recurring workflows and accumulated customer context that increase switching costs.
Watch out Excitement and trial volume are not the same as durable retention.
- 6
Compare picks and shovels
Benchmark the candidate against infrastructure providers that earn revenue across many downstream winners. Invest only when the application company’s upside and defensibility justify its additional concentration risk.
Pro tip Include compute, chips, cloud capacity, and enabling platforms in the comparison set.
Watch out Infrastructure businesses can still be overpriced or technologically displaced, so durability does not replace valuation analysis.
In the wild
The discussion notes that Microsoft partnered with both closed-source OpenAI and the open-source LLaMA 2 ecosystem. Although the projects compete, workloads from both can run on Azure, allowing Microsoft to benefit from activity across different model strategies.
→ Microsoft gains diversified exposure to AI adoption rather than depending entirely on one model winning.
AI applications and cloud-compute systems require GPUs, and many use Nvidia hardware. Instead of selecting one chatbot, image generator, or voice application, an investor can examine the supplier serving demand across those categories.
→ The investment thesis captures broad AI compute demand while reducing dependence on a single application winner.
Common mistakes
Mistaking an API wrapper for a moat
A polished product can appear differentiated even when competitors can reproduce its core capability with the same underlying model.
Ignoring customer churn
Fast innovation makes novelty-driven AI subscriptions vulnerable to cancellation when cheaper or better alternatives appear.
Backing technology without distribution
A technically impressive point solution may still lose when an incumbent bundles a similar feature into an existing platform.
Is it for you?
Best for
It is best for evaluating AI startups, public technology companies, vendors, or potential employers during a fast-moving market cycle.
Not ideal for
It is not ideal for short-term trading decisions driven primarily by valuation, timing, or technical market signals.
From the transcript
“I tend to think that the best place to be is the, you know, the shovel sellers, right?”
“I think there are two things to really consider to understand where the big opportunity long-term is gonna be, and which is, Matt touched on…”
“If you are just writing code on top of somebody else's APIs, that's not very differentiated”
From the episode
These 9 A.I. Tools Are Better Than ChatGPT… (#145)