Engageably

05 — Journal

← Back to Journal

AI engagements that start with a tool decision miss the P&L

May 25, 2026 · 7 min read · Method · Benjamin Rodriguez

Executive TL;DR

  • Most AI engagements still start with a vendor or model pick before anyone names a P&L consequence. That is a capital allocation error wearing a technology costume.
  • What separates a real investment from another pilot is one thing. A named revenue or cost line the work has to move, written down before the build starts.
  • CFOs keep defunding AI programs mid-cycle because no one tied them to a line the CFO has to defend. The first budget revision finds them. The second one ends them.
  • One question reframes the engagement before a dollar moves. Which line on the P&L does this work move, by how much, and what triggers the next funding decision.

The Shift

AI procurement in 2026 looks a lot like enterprise software procurement in 2010. Tool-led. Vendor-sequenced. Cut off from any operating outcome until someone asks for a renewal budget.

The spend is too large to keep handling this way. Worldwide AI spending is forecast to reach $2.59 trillion in 2026, a 47% year-over-year jump. Gartner has said directly that CIOs struggle to prove value and show tangible business outcomes (Gartner, 2026). The return signal is worse than the spend signal. More than 80% of enterprises report no measurable impact on enterprise-level EBIT from their generative AI use (McKinsey State of AI, 2025). A review of more than 300 enterprise initiatives found that roughly 95% of generative AI pilots returned zero measurable financial value (MIT Project NANDA, 2025).

This is not a vendor problem or a talent problem. It is a framing problem. And it starts in the first conversation between the operator and whoever is pitching the engagement.

Why It Matters Now

The window for exploratory justification is closing. CFOs are asking for P&L accountability on transformation lines earlier than in any past digital cycle. Only 14% of CFOs at U.S. companies between $500M and $10B+ in revenue report clear, measurable impact from their AI investments (RGP CFO survey via CFO.com, 2025). At the same time, 42% of companies walked away from most of their AI initiatives in 2025, up from 17% the year before (S&P Global Market Intelligence, 2025). Abandonment is no longer the tail outcome. It is becoming the median one.

A tool decision made before anyone names a P&L consequence is not a strategy. It is a procurement event.

What Most Companies Are Still Doing

The default pattern is familiar. An internal champion in product or IT surfaces a capability. A vendor demo follows. A pilot gets approved against a vague efficiency thesis, usually a phrase like productivity uplift or faster cycle time. No line on the P&L is attached. The CFO inherits a cost center and a request to wait for proof. The proof rarely arrives in the form requested. No one specified the form.

Then one of two things happens. The pilot runs forever with no funding threshold to clear, quietly turning into a permanent operating expense. Or finance defunds it at the first budget revision, because it has no line on the P&L to defend it. RAND found that more than 80% of AI projects fail, roughly twice the failure rate of conventional IT projects. It traces most of those failures to organizational and framing issues, not technology limits (RAND Corporation, 2024).

What the Best Operators Are Doing Instead

The operators getting durable value from AI work have flipped the sequence. They name the P&L consequence first. They pull the capability requirement from it. They resolve the tool or model decision last, not first. The evidence is now quantitative. When accountability for AI value sits with the CFO, 76% of companies report a great deal of value, versus 53% under CIO or CTO ownership and 32% under functional executives (Harvard Business Review survey via CFO Brew, 2026). The seat that owns the outcome changes the outcome.

1. Name the revenue or cost line first. Identify the specific metric the engagement is accountable to before any system or vendor conversation begins. Not a theme. A line. Gross margin in a named segment. Cost to serve in a named channel. Win rate in a named deal band. If you cannot name the line, the work is not ready for funding.

2. Derive the capability requirement from the outcome. Let the financial target specify the functional need, not the reverse. The question is not what this model can do. The question is what would have to be true in the actual work for that line to move by the targeted amount. Then ask which capability makes that true.

3. Set a return trigger, not a timeline. Define the condition under which funding continues or stops, independent of the project calendar. A return trigger reads like a sentence with a number in it. Calendar gates measure activity. Return triggers measure consequence. Consequence is what the CFO funds.

4. Make the tool the last decision, not the first. The right tool is whichever one meets the capability requirement at acceptable risk and cost. Evaluate it last, against criteria already written down. This is the step where most engagements start today. It belongs at the end.

Implications for the Next 12 Months

AI Budgets Will Face a P&L Audit Whether Operators Prepare or Not

CFOs at mid-market and enterprise companies used to approve transformation lines on strategic intent. Now they require a defined return mechanism. Gartner’s most recent finance survey shows 39% of CFOs naming AI acceleration as a top-five 2026 action item. Only 36% feel confident they can deliver real enterprise impact from AI (Gartner 2026 CFO Top Priorities, 2025). That gap between mandate and confidence is where the audit pressure comes from.

Engagements without a defined return mechanism will surface as impairments or forced write-downs, not as learning investments. The reclassification is already happening at companies that funded multiple tool-first pilots in 2024 and 2025 and cannot point to a moved line.

The Sequencing Error Compounds Across the Portfolio

Companies running several tool-first pilots build up technical and contractual debt. That debt erodes their negotiating leverage when vendors reprice or consolidate. Every commitment you make before defining the outcome raises the cost of resequencing later. The portfolio version is worse than any single engagement. Contracts overlap. Data flows tangle. The renegotiation window narrows with every additional signature.

The Engagement Model Itself Becomes a Differentiator

Some operators can show that they built their AI work around a P&L consequence, not a capability catalog. They will have an easier conversation with their CFO, with any investor or acquirer, and with any external party reviewing program quality. The differentiator is no longer model access or talent. Those have flattened. The differentiator is now the discipline of the framing. That framing shows up on the first page of any program memo.

Executive Next Step

Before approving or extending any AI engagement, require one written answer to one question. Which line on the P&L does this work move, by how much, and what condition triggers the next funding decision. If that answer does not exist, the engagement is still in the procurement phase, no matter how far the build has progressed.

Sources

← Back to Journal