Doubling Your AI Budget Won't Fix a Broken Sequence
94% of companies keep increasing AI spend even without measurable returns. The problem isn't budget — it's sequence.
We made a deliberate sequencing call at Rapid Data last year.
No customer-facing AI features until internal workflows were running on AI first. That meant automating support routing, internal documentation, release tracking — the operational backroom that customers never see. Nothing to demo. Nothing to announce.
For twelve months, while competitors were shipping AI features and running press releases about their AI roadmaps, we were quietly building the foundation. It was uncomfortable. The output wasn't visible. The investment was real.
Twelve months later: the internal foundation is now compounding into Rapid Next in ways that wouldn't have been possible otherwise. The team that built the foundation is the team building the product — and they know how to build with AI because they've been doing it for a year in a context where mistakes had low stakes.
I'm not telling this story because the outcome was obvious. At the time, the sequence felt risky. The pressure to ship customer-facing AI was real. We held the line on sequence anyway.
What the data says about everyone else
BCG's AI Radar shows what happens when companies don't hold that line: 94% of companies are still increasing AI investment — even without measurable EBITDA impact. The budgets keep growing. The P&L stays flat.
This isn't a data quality problem. These are real investments producing real outputs — features that work, pilots that complete, models that run. The disconnect is that "working" and "compounding" are not the same thing.
The companies spending most are often the ones furthest from returns, because they skipped the foundational stages and went straight to the expensive ones.
The sequence that actually compounds
I use a simple framework to diagnose where a company is in its AI journey:
PLANT — Automate one internal process completely. Not partially, not experimentally — completely. Measure what changes. Then move to the next process. This stage produces real efficiency gains and, more importantly, it builds the team's instinct for where AI works and where it doesn't.
BUILD — Build actual AI capability inside your team. Not access — capability. There's a meaningful difference between a team that has Copilot licenses and a team that knows how to design workflows around AI, evaluate outputs, catch failures, and iterate fast. Capability is built by doing, not by training decks.
HARVEST — Now you can build customer-facing AI that compounds. Your team knows how to build it. Your internal processes aren't fighting you. Your product decisions are informed by what you've already learned from a year of internal deployment.
The failure mode I see most often: companies skip PLANT and BUILD entirely. They license foundation models, hire an AI lead, announce a product initiative, and wonder why the features don't move the P&L.
They're trying to HARVEST before they've PLANTED anything.
Why the wrong sequence is so common
The pressure runs in the wrong direction.
Customer-facing AI is visible. It shows up in demos, in press releases, in competitor comparisons. Internal process automation is invisible — it doesn't make a conference talk and it doesn't impress a board. So the incentive is to skip to the visible thing.
The other factor is accountability. If you spend six months automating internal workflows and there's nothing to show customers, you need the confidence that the sequence will pay off. That's a harder internal sell than shipping a feature.
The companies that get this right are usually the ones where the person making the AI investment decision is also the person who has to explain the P&L in twelve months. They don't have the luxury of separating the technology decision from the business outcome.
The diagnostic question
Before the next AI investment decision, one question is worth asking: which stage are you actually in?
Not which stage you've budgeted for — which stage your team is genuinely capable of executing. A company that has never completed a PLANT project isn't ready to HARVEST. Adding budget doesn't change that. It just makes the gap more expensive.
The AI budget conversation in most boardrooms goes: "We need to invest more." The conversation that should happen first: "Why isn't what we already have producing returns?"
In almost every case I've seen, the answer is sequence. The investment came before the foundation.
Doubling the budget is a reasonable answer to the wrong problem. The right answer is slower, less visible, and significantly more durable.
What stage are you actually in right now — PLANT, BUILD, or HARVEST? And does the investment you're making match the stage you're in?