Three of the biggest names in enterprise AI just bet billions that the hard part isn’t your data. On September 8, 2026, Accenture and Google Cloud announced a group built around 1,000 forward-deployed engineers, people who work inside a customer’s business and build the system there. Microsoft made the same bet in July 2026: $2.5 billion, 6,000 embedded experts. All three sell that work, so read the spend as a bet, not proof.
Why It Matters
McKinsey’s growth practice put a number on the same bet last month, in a piece on seven myths that hold growth back. The last is one you’ve probably heard: you need perfect data before you can introduce AI. Their answer, and the scope matters: in growth work, sales and pricing and marketing, 80 percent of companies don’t need a data lake or a rebuilt architecture to get value. That’s their stated experience, not a survey, and they benefit from you starting now instead of a year of plumbing. Weigh it accordingly. The useful part isn’t the number. It’s that the question changes from whether your data is ready to which process is ready first.
The Decision
So the question worth putting to your team is which job your next AI dollar buys. Rebuilding the data estate, the warehouse and the pipes, is a project with a vendor and a year attached. Getting one process clean enough to ship is smaller, and nobody champions it. That’s why the big project wins by default. Both are defensible. What isn’t is funding the first and calling it an AI project.
What To Do This Week
- Take the last AI proposal that stalled and find where the data problem enters. Read what it asks for.
- Pick three processes. For each, ask how many people a new joiner would have to go to before doing the work unaided. Write the names down.
- The shortest list wins. That list is your data gap, and it’s smaller to fix than an architecture.
What Not To Do
Don’t read this as permission to skip data work. In the same piece, Reckitt keeps its less data-rich regions on a simpler version of its tool while those teams build hygiene first. That’s what a failed test should produce: a smaller thing that ships, not a pause. And don’t ask your platform vendor whether you need the platform.
Signal Boost
Growth favors the bold: AI as force multiplier is where the 80 percent comes from. McKinsey, August 6, 2026. Seven myths, each with their answer under it. The other six are worth an hour with whoever writes your AI proposals.