Pull up your organisation's AI roadmap. If quarter one is "platform selection," quarter two is "data foundation," and business value first appears somewhere in the hazy back half of year two, you are holding the standard template — and the standard template is backwards.
Why the standard template fails
Technology-first roadmaps assume you can specify the platform before you know the workload. In practice, the workloads discovered along the way never quite fit the platform bought up front, and the roadmap's back half becomes a quiet negotiation between what was purchased and what is actually needed. Worse, by deferring value to year two, the program spends its political capital before producing anything a CFO can see. Most transformations don't die of technical failure; they die of impatience — and the impatience is justified.
The inverted roadmap
Start from decisions, not systems. List the ten decisions your organisation makes most often where better information would most move the P&L — pricing, replenishment, credit, routing, staffing, churn intervention. Rank them by value and by data readiness. The intersection of high value and available data is your quarter one, and it defines — not follows — your technology choices.
Then build the thinnest possible path from data to that one decision, in production, with a named business owner. The platform emerges from generalising what the first three such systems have in common, which means you buy or build it knowing exactly what it must do. Funding stays honest too: each quarter's expansion is financed by the measurable returns of the last, rather than by a promissory note against year three.
Value cases first, models second, platform last. It feels upside-down to anyone trained on ERP-era programs. It is also, consistently, how the transformations that actually ship are sequenced.