The Transformation Trap

I have sold transformation for twenty years. Here is what I know about how it ends.

It starts the same way every time. Web 2.0 transformation. Digital transformation. Cloud transformation. Finance transformation. Agile at scale. A new methodology, a new technology, or a new trend arrives that is going to change the way you do business. You do not want to be left behind, so you spend a few months planning and jump in the deep end.

The plans always look the same. Every big firm has a different flavor of the same meat and potatoes. Transformation touches every part of the company: operations, people, go to market, technology, org structure, and the change management program wrapped around all of it. The organizations that jump first are mostly the first to fail. Very few realize the benefits promised at the top of the cycle. There are always hidden external costs, third party dependencies that change their pricing, and requirements nobody saw that add cost, risk, and complexity. Then come three to five years of cleanup, resets, and restarts toward something the organization can actually achieve.

Transformation is a product. It has a sales cycle, a pricing model, and a refresh every five years. The refresh is what you are being sold right now.

The number that never moves

You do not have to take my word for it. The firms measure their own results and publish them.

In 2024, Bain reported that only 12 percent of business transformations achieve their original ambition. Eighty eight percent do not. The line I keep coming back to is Bain's own observation that 12 percent is the same number as five years earlier. The frameworks were refreshed twice in that window. The success rate did not move.

BCG got a similar answer in 2020. Across 825 executives and 70 client engagements, 30 percent of digital transformations met their targets, 44 percent created some value but missed, and 26 percent created little or nothing. The headline of that report was that six success factors could flip the odds from 30 percent to 80 percent. Four years later Bain measured 12.

McKinsey's 2018 global survey found that fewer than 30 percent of digital transformations succeed, and only 16 percent both improved performance and sustained the improvement.

Here is the part that should bother you. The famous "70 percent of transformations fail" statistic, the one that opens half the decks in the industry, was never measured. It traces to Hammer and Champy's 1993 book on reengineering, where the authors called their own 50 to 70 percent estimate unscientific. A peer reviewed trace in 2011 found no empirical basis for it. McKinsey repeated it without a source in 2015. HBR turned it into 900 billion dollars of waste in 2019 by multiplying the unsourced rate by a spending forecast. For thirty years the industry has sold both the disease statistic and the cure, and the disease statistic was made up. You do not need it. The real numbers from the firms' own surveys are worse, because they show no improvement.

Now look at the cures. Kotter's 1995 article in HBR gave us eight steps: urgency, a guiding coalition, a vision, communication, removing obstacles, short term wins, consolidating gains, anchoring in culture. McKinsey's 2018 research sorted 21 keys into five categories: leadership, capability building, empowering workers, upgrading tools, communication. McKinsey's 2023 book Rewired has six capabilities: roadmap, talent, operating model, technology, data, adoption. BCG's 2024 AI report has a 10, 20, 70 rule: 10 percent algorithms, 20 percent technology and data, 70 percent people and process. Accenture calls its version Total Enterprise Reinvention.

Every one of these is Kotter with a new noun in front of it. An independent comparison of the AI transformation frameworks scored the Big Four and MBB methodologies at 2.0 out of 5 on mid market applicability, because they were designed for Fortune 500 scale and treat change management as a separate workstream. If you run a company with 500 to 10,000 people, the framework in the deck was not built for you.

The consulting industry has run the same playbook for thirty years, measured its own results, found that nine out of ten programs miss, and responded by renaming the playbook.

Why the trap persists

From one standpoint, I understand the fascination. Three incentives hold the cycle in place, and none of them rewards realization.

The sponsor's clock. The average chief digital officer lasts about 31 months, the shortest tenure in the C suite. COOs average 3.3 years, CIOs 4.6, CMOs 4.1. A transformation with a three to five year benefits horizon is, for the executive who launched it, a launch and a departure. They need the big bang to hang their hat on for the next role. The deck is the deliverable. The realization belongs to a successor.

Nobody keeps the ledger. PMI's research on benefits realization found that 83 percent of organizations lack maturity in it. Promised benefits are a business case artifact. Realized benefits are almost never reconciled against them, and when they are, the sponsor has usually moved on. Executives get promoted on the strength of the launch. A missed realization has no owner, so it has no consequence.

The firm is paid on the plan. In fiscal 2025 Accenture booked 5.9 billion dollars of generative and agentic AI work, recognized 2.7 billion in revenue from it, and folded five business units into a group it named Reinvention Services. McKinsey says about 40 percent of its work is now AI related. This is not a conspiracy. It is a business model. A firm whose growth engine is the current wave will always tell you the current wave is different.

The sponsor leaves before the benefits are due, the ledger is never kept, and the firm is paid on the plan. Every incentive in the room rewards starting. None of them rewards finishing.

Cloud: a completed cycle

Cloud is the wave worth studying because it is over. Everyone can see how it ended.

It arrived as platform as a service. Windows Azure went live on February 1, 2010 as a pure platform: web and worker roles, blob storage, SQL Azure, a service bus. No virtual machines. The pitch was that you would modernize your applications onto the platform and the datacenter would go away. Azure did not ship infrastructure as a service until April 2013, after AWS had proven that renting servers was where the money was.

What really happened is that modernizing applications for PaaS turned out to be a rewrite, and most large organizations could not do it at the pace the plan assumed. So they moved what they could as IaaS, which for a company that already owned its datacenters meant moving the same servers to someone else's building at consumption pricing. For many, the opex bill came in higher than the capex it replaced. The change required across technology, operations, and the revenue generating parts of the business was massive, and those changes do not happen easily in established companies. They take years.

The receipts arrived a decade later. Andreessen Horowitz estimated in 2021 that cloud spend was suppressing margins enough to cost the top 50 public software companies 100 billion dollars in market value, and pointed to Dropbox saving 75 million dollars over two years by leaving. A Barclays survey of CIOs in late 2024 found 86 percent planned to move workloads back to private or on premises infrastructure, the highest reading in the survey's history. Flexera puts the share of cloud spend that is simply wasted at about 27 percent.

Fifteen years in, cloud is not a transformation. It is a workload placement decision that gets revisited every budget cycle. Some things belong there, some do not, and the sensible organizations are optimizing the mix. That is what it always was.

Cloud did not transform the enterprise. The enterprise spent a decade and a great deal of money discovering which workloads it should have moved in the first place. That is optimization with a very bad first three years.

AI transformation: the same wave, with a meter running

Last week I wrote about judgment versus intelligence, and the argument was that companies are aiming agents at the work with the most judgment in it, where the objective cannot be written down and the value is theorized. That is the pilot. The transformation is what gets wrapped around the pilot, and it is following the cloud script line for line.

Start with the firms' own diagnosis. McKinsey's June 2025 paper on agentic AI opens with what it calls the gen AI paradox: nearly 80 percent of companies have deployed generative AI, and roughly 80 percent report no material impact on earnings. About 90 percent of function specific use cases are stuck in pilot. The prescription is a new architecture, a new operating model, and a CEO playbook for agents.

Read that again. The firm diagnosed that the last wave produced no earnings impact for four out of five clients, and prescribed the next wave. The cure for the last transformation is always the next transformation.

The independent numbers say the same thing. MIT's NANDA project found that 95 percent of enterprise generative AI pilots deliver no measurable P&L impact against 30 to 40 billion dollars invested. S&P Global found that 42 percent of companies abandoned most of their AI initiatives in 2025, up from 17 percent the year before, with the average organization scrapping 46 percent of its proofs of concept before production. The top reason cited was cost.

Cost is where this wave is different, and worse. An agent does not make one model call. It plans, retrieves context, calls a tool, checks its own output, corrects itself, and calls again. A task that was a single completion in a chatbot is ten to twenty model calls in an agent, and the models doing the reasoning are the expensive ones. Per token prices have fallen sharply and the bills have gone up anyway, because the architecture multiplies calls faster than the price falls. Call it the recursive intelligence trap.

Gartner has been tracking the bill. In 2024 it warned that CFOs learning AI costs as they go were producing estimates off by 500 to 1,000 percent. In January it predicted that by 2030 the cost per resolution for generative AI in customer service will exceed the cost of an offshore human agent, and said plainly that full automation will be prohibitively expensive for most organizations. In June it predicted that AI coding costs will overtake the average developer's salary by 2028, with monthly bills per developer moving from 20 to 100 dollars toward 2,000 to 5,000, and noted that there is no direct relationship between rising token consumption and rising productivity.

Even the people selling the tokens have noticed. Sam Altman told CNBC in June that enterprise customers had already burned through their 2026 AI budgets, and that the complaint he now hears is "I am spending a ton of money on AI, and I know some great stuff is happening, but I know there's a ton of waste." He called it the most fair criticism of AI right now.

So here is the cycle, again. The pilot runs against work that requires judgment. The cost of running it at scale is a multiple of what the business case assumed. The organizational change needed to support the benefit is two years behind the technology. The pilot fails, the transformation plan goes in the trash, and three to five years of cleanup begin toward a plan the organization can actually execute.

The AI transformation deck is the cloud transformation deck with the nouns changed, and the meter runs faster.

When does transformation become running the business?

I am not arguing against the technology. Agents are going to take an enormous amount of low judgment work off the books, and that is real. I am arguing against the category. "Transformation" is what you call optimization when you want a bigger budget and a shorter clock.

Look at what the survivors of every wave actually did. They placed small bets with kill criteria, sequenced by how much judgment was in the work, and started at the bottom. They kept a ledger: every promised benefit had an owner, a baseline, and a date, and realized was reconciled against promised continuously, whether or not the sponsor was still in the building. They modeled cost at scale, not at pilot, because a unit economic that does not hold at ten times the volume with a twenty times token multiplier is not a business case. And they funded the organizational change ahead of the technology instead of wrapping it around the end.

None of that is transformation. It is management. It is the next cycle of optimization and growth, done by people who intend to be there when the benefit is due.

That is why value realization is a first class object in Analyzt AI rather than a slide at the end of the program. Promised benefits, owners, baselines, and realized against promised, with the gap quantified and visible to whoever holds the seat now. The organizations that get out of the trap are the ones that can answer, at any point in the cycle, what was promised and what has actually shown up.

The next time the deck lands on your desk, ask three questions. Who owns the benefit in year four? What does this cost at scale, not at pilot? And what would we do differently if we called it optimization instead? If the answers are nobody, we don't know, and the same thing with a smaller budget, you have found the trap.

---

If you have been through a wave or two yourself, I would like to hear how it ended. The beta is open and free while it lasts. And if you run a consulting firm that keeps the ledger and gets paid on what shows up, we would like to work with you.