← Findings 🕐 10 min read
Findings

The Year-One AI Program Review: Five Questions for Your Annual Leadership Offsite

Brandon Sneider · March 2026

The 90-day check evaluates a pilot. The quarterly risk check-in maintains governance cadence.

Executive Summary

  • The median AI project that fails does so in 13.7 months — almost exactly at the one-year mark (Pertama Partners, 2,400+ initiatives, 2025-2026). The year-one review is not a formality. It is the moment when leadership decides whether to scale, pivot, or cut
  • Companies with pre-defined success metrics achieve 54% project success versus 12% without (Pertama Partners, 2,400+ initiatives, 2025-2026). If the metrics were not defined before the pilot launched, the year-one review requires defining them retroactively — then measuring honestly
  • Only 6% of companies report significant AI-driven EBIT impact; 88% deploy AI but only 39% see any EBIT impact at all, usually under 5% (McKinsey State of AI, n=1,993, June-July 2025). Most year-one reviews will show modest returns. The question is whether those returns are on a trajectory toward significance or have plateaued
  • 56% of AI programs lose C-suite sponsorship within six months — and those programs succeed at 11% versus 68% for those that maintain it (Pertama Partners, 2025-2026). The year-one review is the CEO’s opportunity to publicly recommit or redirect, not quietly let the program drift
  • 42% of companies abandoned most AI initiatives in 2025, up from 17% in 2024 (S&P Global VotE, n=1,006, 2025). The companies that avoided this had a structured review process that caught problems early enough to fix them

Why Year One Matters More Than Any Other Milestone

The 90-day check evaluates a pilot. The quarterly risk check-in maintains governance cadence. But year one is the strategic inflection point — the moment when AI shifts from “experiment” to “budget line item” or dies quietly through inattention.

The data is stark. Successful AI projects cost an average of $5.1M and deliver $14.7M in value — a 188% ROI. Failed projects cost $6.8M and deliver $1.9M, a negative 72% ROI (Pertama Partners, 2,400+ initiatives, 2025-2026). The difference is not technology selection. It is whether leadership evaluated honestly at the right moments.

For a 200-500 person company, the scale is smaller but the dynamic is identical. A $200K AI investment that delivered measurable value in two use cases is a success worth scaling. A $200K investment that produced enthusiastic demos but no workflow change is a warning sign that the year-one review should surface — not hide.

The Five Questions

Question 1: What Did AI Actually Cost Versus What You Budgeted?

Most companies underestimate AI costs by a wide margin. 85% of organizations misestimate AI costs by more than 10%, and nearly a quarter miss by 50% or more (CloudZero/AICosts.ai, 2025). Cost overruns average 380% at production scale versus pilot projections (MIT Sloan, 2025). At mid-market scale, the overrun is smaller but still real.

What to calculate:

Cost Category Budget Actual Variance
Software licenses $ $ $
Implementation and integration $ $ $
Training and change management $ $ $
Data preparation and cleanup $ $ $
Internal IT time (hours × loaded rate) $ $ $
Ongoing support and maintenance $ $ $
Total $ $ $

Pull data from the license audit (if conducted) and the TCO one-pager framework. The most commonly missed categories: internal IT time, data preparation, and the management hours spent governing the program.

What a healthy answer looks like: Total cost within 20% of budget. If the variance exceeds 30%, the budget process — not the technology — needs attention before year two.

Question 2: Which Use Cases Delivered Measurable Value and Which Did Not?

BCG’s research finds that 60% of companies generate no material value from AI despite investment. Only 5% — the “future-built” firms — create substantial value at scale, achieving 1.7x revenue growth and 2.7x ROI versus laggards (BCG, n=2,000+, September 2025).

The pattern among the 5%: they measure by use case, not in aggregate. “AI saved the company money” is not a finding. “AI reduced customer service response time from 4 hours to 22 minutes in the support team, handling 340 tickets per month that previously required manual triage” is.

For each AI use case deployed, answer:

Question Use Case 1 Use Case 2 Use Case 3
What baseline metric did you establish before launch?
What is that metric today?
How many employees use this tool daily?
What is the dollar value of time saved per month?
Did this use case change a workflow or just add a tool?
Would the team resist if you removed it tomorrow?

The last question is the most revealing. If the answer is “no one would notice,” the use case delivered adoption without value. If the answer is “the team would revolt,” the use case delivered genuine workflow integration — and that is where year-two investment should concentrate.

Question 3: Did the Organization Get Better at Using AI, or Just More Familiar With It?

Usage and proficiency are different things. OpenAI’s enterprise data shows “frontier workers” send 6x more messages than the median — but more messages do not equal more value (OpenAI State of Enterprise AI, n=9,000 workers, 2025). Deloitte’s 2026 survey finds workforce AI access expanded 50% in one year, from under 40% to 60% of workers, but fewer than 60% of those with access use it daily (Deloitte, n=3,235, August-September 2025).

Organizations with formal AI training programs report 2.7x higher proficiency scores and 4.1x higher user satisfaction (Larridin, 2025). The distinction matters: a company where 200 employees have AI licenses but 30 use them daily has an adoption problem, not a technology problem.

What to measure:

  • Adoption rate: Employees with access who use AI tools at least weekly (target: 60%+)
  • Proficiency distribution: What percentage of users are past the “ask it a question” stage and into structured workflows, custom prompts, or integrated processes?
  • Training completion: What percentage of employees completed formal AI training versus self-guided exploration?
  • Champion network: Have internal AI power users emerged? Are they formally recognized? Are they contributing to governance? At mid-market companies, individual contributors — especially those under 35 — are often more fluent with AI than the IT team tasked with governing it. This is a feature to channel, not a problem to solve.

Question 4: What Risks Materialized and What Controls Worked?

Only 28% of S&P 100 companies disclose both board-level AI oversight and a formal AI policy (ISS Corporate, Harvard Law School Forum, March 2026). Mid-market companies have even less formal governance. The year-one review should document what happened — including the incidents that did not escalate.

Risk inventory for the year-one review:

  • Did any employee use AI on data the acceptable use policy prohibits? (If the answer is “I don’t know,” the monitoring controls need attention)
  • Did any AI output reach a customer or partner without human review?
  • Were there accuracy failures that required correction? How were they caught — by process or by luck?
  • Did any vendor change terms, pricing, or data handling practices during the year?
  • What did the cyber insurer ask about AI at renewal? Could the organization answer confidently?

Deloitte’s 2026 data reveals a troubling readiness gap: technical infrastructure readiness sits at 43%, data management at 40%, and talent readiness at just 20% — all lower than the prior year (Deloitte, n=3,235, 2025). The year-one review should assess whether the organization’s readiness improved or declined against its own baseline.

Question 5: What Should Next Year’s AI Program Look Like?

This is the question the year-one review exists to answer. The data should inform one of three decisions:

Scale: The pilot delivered measurable value, adoption exceeded 50%, risks were manageable, and the cost was within budget tolerance. Year two doubles down: expand to additional teams (see the pilot-scaling framework), increase the budget by 25-50%, and formalize the governance structure that worked informally in year one.

Pivot: The technology worked but the use case did not — or the use case worked but the technology was wrong. Year two changes the target, not the ambition. 46% of AI projects get scrapped between proof of concept and production (S&P Global, n=1,006, 2025). The companies that succeed after a pivot are the ones that documented what they learned, not just what they spent.

Reduce: The program consumed resources without producing measurable value, adoption stayed below 30% despite training investment, and the cost-to-value ratio was negative. Year two reduces scope to a single high-value use case with strict metrics and a 120-day deadline. This is not failure — it is discipline. The alternative is the slow bleed of another $200K into tools no one uses.

The budget decision:

Year-One Outcome Year-Two Budget Recommendation Rationale
Measurable value, high adoption Increase 25-50% Scale what works to additional teams
Mixed results, moderate adoption Hold steady Fix adoption gaps before adding scope
No measurable value, low adoption Reduce 50%, single use case Prove value in one place before expanding

Key Data Points

  • 13.7 months: Median time from AI project approval to failure (Pertama Partners, 2,400+ initiatives, 2025-2026)
  • 54% vs. 12%: Project success rate with pre-defined metrics versus without (Pertama Partners, 2,400+ initiatives, 2025-2026)
  • 380%: Average cost overrun at production scale versus pilot projections (MIT Sloan, 2025)
  • 6%: Companies reporting significant AI-driven EBIT impact (McKinsey, n=1,993, June-July 2025)
  • 68% vs. 11%: Success rate with sustained C-suite sponsorship versus lost sponsorship (Pertama Partners, 2025-2026)
  • 42%: Companies that abandoned most AI initiatives in 2025, up from 17% in 2024 (S&P Global VotE, n=1,006, 2025)
  • 188% ROI: Average return on successful AI projects; −72% on failed ones (Pertama Partners, 2,400+ initiatives, 2025-2026)
  • 60%: Workforce AI access rate in 2025, up from under 40% in 2024 — but fewer than 60% of those with access use it daily (Deloitte, n=3,235, 2025)
  • 2.7x: Proficiency improvement with formal training programs versus self-guided learning (Larridin, 2025)

What This Means for Your Organization

The year-one review is the single most consequential governance moment in an AI program. It determines whether the CEO presents a structured narrative of learning and progress — or defends a cost line that nobody can justify with data. The companies that get this right share three characteristics: they defined success metrics before they started, they measured honestly against those metrics, and they made the scale-pivot-reduce decision based on evidence rather than sunk-cost psychology.

Most mid-market AI programs will show modest year-one results. That is normal. BCG’s data confirms that only 5% of companies generate substantial AI value at scale — but those companies started exactly where the audience is now. They just measured more rigorously, learned more deliberately, and made harder decisions at the twelve-month mark.

The year-one review template above takes a leadership team approximately two hours to complete — one hour to gather the data across the five questions, one hour to discuss and decide. It produces a one-page summary suitable for a board presentation or leadership offsite slide. If the review raises questions specific to how the findings apply at your scale and in your industry, I would welcome the conversation — brandon@brandonsneider.com.

Sources

  1. Pertama Partners — AI Project Failure Statistics 2026. Synthesis of 2,400+ enterprise AI initiatives tracked through 2025-2026, aggregating data from RAND Corporation, MIT Sloan, McKinsey, Deloitte, and Gartner. Source credibility: moderate-high (practitioner synthesis of multiple independent sources, large aggregate sample). https://www.pertamapartners.com/insights/ai-project-failure-statistics-2026

  2. McKinsey & Company — The State of AI in 2025. Survey of 1,993 respondents across 105 countries, June-July 2025. Source credibility: high (independent annual survey, large sample, consistent methodology). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai

  3. S&P Global Market Intelligence — Voice of the Enterprise: AI & Machine Learning, Use Cases 2025. Survey of 1,006 IT and business professionals across North America and Europe. Source credibility: very high (premier market intelligence firm, rigorous methodology, large sample). https://www.spglobal.com/market-intelligence/en/news-insights/research/ai-experiences-rapid-adoption-but-with-mixed-outcomes-highlights-from-vote-ai-machine-learning

  4. BCG — The Widening AI Value Gap, September 2025. Survey of 2,000+ executives globally. Source credibility: high (premier consulting firm, large sample, annual study). https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap

  5. Deloitte — The State of AI in the Enterprise, 2026. Survey of 3,235 business and IT leaders across 24 countries, August-September 2025. Source credibility: high (largest sample in category, multi-industry, multi-geography). https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html

  6. OpenAI — The State of Enterprise AI, 2025. Survey of 9,000 workers across nearly 100 enterprises. Source credibility: moderate (vendor-published, but large sample and detailed methodology). https://openai.com/index/the-state-of-enterprise-ai-2025-report/

  7. ISS Corporate / Harvard Law School Forum on Corporate Governance — US AI Oversight Through Three Lenses, March 2026. Analysis of S&P 100 AI governance disclosures. Source credibility: very high (independent governance research, premier academic institution). https://corpgov.law.harvard.edu/2026/03/11/us-ai-oversight-through-three-lenses-investor-expectations-the-sp-100-and-company-specific-analysis/

  8. Larridin — The AI ROI Measurement Framework, 2025. Practitioner framework with training program data. Source credibility: moderate (advisory firm, methodology described but sample sizes not specified). https://larridin.com/blog/ai-roi-measurement

  9. CloudZero / AICosts.ai — The State of AI Costs in 2025 and Complete Guide to AI Pricing 2025. Cost benchmarking data across enterprise AI deployments. Source credibility: moderate (vendor-adjacent, but cost data is independently verifiable). https://www.cloudzero.com/state-of-ai-costs/

Brandon Sneider | brandon@brandonsneider.com March 2026