Engageably

05 — Journal

← Back to Journal

Your AI initiative is a pile of pilots, not a strategy

May 27, 2026 · 8 min read · Capability · Benjamin Rodriguez

Executive TL;DR

  • Most mid-market and enterprise AI programs are not strategies. They are pilot portfolios funded by curiosity and defended by activity counts.
  • Companies greenlight pilots because they are politically safe and cheap to start. Not because they tie to a competitive claim the CEO will make in public.
  • The cost shows up on the books now: stranded spend, vendor lock-in, talent that will not stay for a portfolio with no graduation path, and a widening gap between AI spend and P&L.
  • This one belongs to the CEO. Turn the pilot inventory into an operating architecture this year, while you still have the leverage to set the terms inside the company.

The Shift

The cost to launch an AI pilot has fallen to near zero. The cost to scale one has not budged.

Adoption is now nearly universal. Embedded value is not. McKinsey’s most recent reading finds that close to nine in ten organizations use AI regularly. Yet most have not embedded it deeply enough to move enterprise financials (McKinsey & Company, The State of AI in 2025, 2025). Only about 6% qualify as high performers, attributing more than 5% of EBIT to AI. Only around 39% report any measurable EBIT effect in the past year (McKinsey & Company, The State of AI in 2025, 2025). The story is not that adoption is rising. The story is that adoption and impact have come apart.

A pile of pilots is now the baseline, not the win. Treating it as progress is the first strategic error.

Why It Matters Now

The companies that already turned pilots into an operating architecture are compounding. They are rewiring the work, not decorating it. McKinsey ranks workflow redesign at the top of 25 attributes that drive EBIT impact from generative AI. Yet only about 21% of generative AI users have redesigned any workflows (McKinsey & Company, The State of AI (Rewired report), 2025). The pilot-heavy company is still counting use cases launched. The leader is counting workflows changed. That gap compounds every quarter.

Counting pilots is a vanity metric. The only number that matters is how many of them have changed how the company competes.

What Most Companies Are Still Doing

The failure pattern is consistent and operational. Pilots sit inside one function with no map of who else they touch. Companies judge success at the demo, not at the workflow. They fund AI as a technology line item, not as a capability that forces the operating model to change shape.

MIT’s NANDA research found that roughly 95% of enterprise generative AI pilots deliver no measurable P&L impact, with rapid revenue gains concentrated in a narrow 5% (MIT NANDA, The GenAI Divide: State of AI in Business 2025 (via Fortune), 2025). S&P Global’s 2025 enterprise survey shows the discipline gap hardening. 42% of companies abandoned most of their AI initiatives in 2025, up from 17% in 2024. The average organization scrapped 46% of proofs of concept before production (S&P Global Market Intelligence (451 Research VotE: AI & Machine Learning 2025), 2025). Nearly half of organizations that invested in generative AI report no single enterprise objective saw a strong positive impact (S&P Global Market Intelligence, 2025).

What sustains the pattern is not ignorance. It is politics. A pilot carries less accountability than a program. The AI team’s mandate and the P&L owner’s authority sit in different reporting lines, so neither side has to defend an outcome. Everyone claims activity. No one claims a number.

What the Best Operators Are Doing Instead

The move that separates them is not better pilots. It is an architecture decision made at the CEO level that says, in plain language, which capabilities must compound across the company and which are genuinely contained experiments. Four pieces separate that posture from a pile of pilots dressed up as a strategy.

1. A defined capability thesis. The company ties AI spend to a specific competitive claim it is willing to make. We will underwrite faster than peers. We will service at a lower cost-to-serve. We will price with sharper margin discipline. Without that claim, every pilot looks equally interesting, which is the same as saying none of them are.

2. Sequencing that puts workflow integration ahead of tool deployment. Buying a tool takes weeks. Rewiring a workflow takes quarters. The operators winning right now sequence the harder thing first because it is the only thing that produces durable margin. McKinsey’s finding that workflow redesign is the single biggest EBIT driver is not a hint. It is the brief.

3. A shared success metric between the P&L owner and the AI function. Not parallel scorecards. One number. If the CRO owns conversion and the AI team owns model performance, the handoff will fail every time. When both sides answer for the same outcome, the political safety of the pilot evaporates and a real program emerges.

4. A defined graduate-or-kill discipline. Every pilot has a defined window. It graduates to a program, or you kill it. MIT’s data shows tools bought from specialized vendors succeed about 67% of the time. Internal builds succeed roughly a third as often (MIT NANDA, The GenAI Divide: State of AI in Business 2025 (via Fortune), 2025). That is not a build-versus-buy debate. It is evidence that governance and vendor strategy decide graduation rates. And graduation rates decide whether the portfolio is an investment or a sunk cost.

Implications for the Next 12 Months

The Compounding Integration Gap

Companies that have not made an architecture decision by year-end will face a much harder integration problem in 2027. Vendor lock-in, data fragmentation, and internal skill atrophy stack on top of each other and feed each other. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, driven by cost, unclear business value, and weak risk controls (Gartner press release, June 25, 2025, 2025). The cancellations will not be evenly distributed. They will concentrate in the companies that never decided what the portfolio was for.

The Talent Signal

Applied AI talent reads an operating model the way a director reads a script on the first pass. One meeting in, they know whether the work will ship or sit. Companies with a coherent architecture keep that talent. Pilot portfolios with no graduation path lose it, and the gap is widening faster than compensation can close it. You will see it in hiring leverage now and in retention by mid-year.

The Financial Exposure

Pilot spend you cannot trace to a workflow outcome or a competitive capability gets harder to defend every quarter. In a tightening operating environment, that pressure resolves one of two ways. A write-down on tooling that never reached production. Or a forced consolidation onto whatever vendor has the most of your data already, which is rarely the vendor the CFO would have picked on the merits. Neither is a strategic migration. Both are concessions.

Executive Next Step

Map the current pilot portfolio against one question: which of these, if scaled, changes how we compete. Every pilot that cannot answer it is overhead, not investment. Set the review for the next 30 days, name a single owner, and publish the kill list inside 60 days.

Sources

← Back to Journal