Hope Frank, Global Chief Marketing Officer, Wand AI
Your Marketing Moat Is Memory, Not The Model
Once a CMO has designed a dual workforce and given each agent a badge, a boss and a budget, another question follows: What happens to everything that workforce learns?
Picture two marketing organizations licensing the same frontier model. On day one, their agents look equally capable. Twelve months later, one remembers that security language wins enterprise buyers in Germany, a creative format fatigues after four exposures and a discount that lifted conversion weakened retention. The other starts each planning cycle with a fresh brief. Accumulated experience creates that separation.
Stanford’s 2026 AI Index found models from four companies clustered within 25 Elo points on the Arena leaderboard. Its 2025 report showed the cost of querying a model at roughly GPT-3.5 performance falling from $20 to 7 cents per million tokens between November 2022 and October 2024, a reduction of more than 280 times.
The next model release may improve everyone’s baseline, but it can’t supply a company’s history of customers, decisions, failures and earned judgment. Experience also compounds, so every month of delay widens the gap.
Read More
