For AI tied to strategy, McKinsey and BCG are the obvious calls; for rollouts across many business units, Accenture, Deloitte and IBM bring partnerships with the major model makers and their own agent platforms; Slalom, Thoughtworks and EPAM build and modernize the systems; Booz Allen serves federal agencies; and Quantiphi is the Google Cloud and AWS expert. The answers further down explain why so many AI pilots stall, how to spot exaggerated AI claims and how to avoid being locked in to a consultant’s AI system.
Why do so many AI pilots fail to pay off?
Mostly because they never become part of how the business runs. MIT’s Project NANDA, which reviewed more than 300 public AI initiatives and interviewed people at 52 organizations for its July 2025 report, found that despite $30 billion to $40 billion of enterprise investment in generative AI, 95% of organizations were getting zero return, and only 5% of custom enterprise AI tools reached production. Its interviews suggested that working with an outside provider beats building alone, with partnerships reaching deployment about two-thirds of the time against about one-third for internal builds, though the authors note that the figures are self-reported and may not prove cause and effect. Data is the other common failure point: in a 2024 Gartner survey of 1,203 data management leaders, 63% lacked, or were unsure whether they had, the right data management practices for AI, and Gartner predicts that through 2026 organizations will abandon 60% of AI projects that lack AI-ready data. Ask any consultant how it will measure business results after the pilot and what has to be fixed in your data first.
How can we spot exaggerated AI claims?
Regulators call it AI washing, and they have started punishing it. In March 2024 the SEC fined two investment advisers, Delphia and Global Predictions, a combined $400,000 for claiming to use AI they did not, and in April 2025 it charged the founder of the shopping app Nate with raising more than $42 million on claims that AI completed purchases when contract workers in a call center in the Philippines did much of the work. The FTC’s Operation AI Comply, launched in September 2024, included a $193,000 penalty against DoNotPay, which had billed itself as the world’s first robot lawyer, and Gartner estimated in June 2025 that only about 130 of the thousands of vendors selling agentic AI offered real agents, with many rebranding chatbots, robotic process automation or assistants. Ask a consultant for production systems you can confirm with the client, which models and data it used, and how much of the work is automated rather than done by people.
How do we avoid getting locked in to a consultant’s AI system?
Agree on ownership and exit terms before any work starts. Gartner predicted in September 2026 that by 2028, 70% of enterprises will abandon agentic AI built for them by vendors’ forward-deployed engineers, trapped by soaring costs and unable to evolve the systems on their own, and its analyst Mukul Saha said the best-scoped engagements set out governance, business value delivery, intellectual-property ownership, project co-ownership, knowledge transfer and an exit strategy from day one. In practice that means a contract giving you the code, prompts, evaluation data and model settings, documentation your own staff can maintain, a schedule for training them and the freedom to switch model providers. EPAM, for one, is expanding its network of forward-deployed engineers, so ask any firm that embeds engineers in your teams how and when they will hand the work over.