Case study · free PDF
Decision models vs foundation models
Which model should make each decision in your marketing stack? I built a game to measure it. A purpose-built decision model and nine foundation models answered the same well-defined decisions; this is what it cost and how long it took.
- Why small, repeated decisions add up in marketing automation
- A fair, recorded benchmark: latency, cost and agreement per decision
- When to use a decision model, when to use a foundation model, and how to combine them
From one recorded run: 20 game decisions per model, called from Australia on 5 October 2026. A tenth foundation model, Grok 4.7, was left out after 2 of its 20 calls timed out; it was also the slowest, so leaving it out narrows the gap. The PDF covers the method and its limits.