Somewhere in every organisation there is a team that built its workflows around a tool that no longer exists. They migrated data, retrained processes, learned the interface. Then the product was discontinued, or the vendor pivoted, or the budget disappeared. The manual processes they had abandoned had to be rebuilt from memory.
That team remembers the effort. They rarely remember the external reason it failed. What remains is simpler: a feeling that this kind of thing does not work for us.
The Numbers Behind the Feeling
Forty-two percent of companies scrapped at least one AI initiative in 2025 — up from seventeen percent the year before, according to S&P Global Market Intelligence. Large enterprises abandoned an average of 2.3 initiatives each, at an average sunk cost of $7.2 million per project.
Behind each number is a team that did the work. They attended the workshops. They cleaned the data. They adjusted their processes. And then it stopped — not because they failed, but because something outside their control changed.
When someone arrives with the next AI proposal, they are not entering a blank conversation. They are entering a room shaped by what happened last time.
Three Shapes of the Scar
The disappearing-vendor scar. A tool was adopted, workflows rebuilt around it, and the provider changed direction. The investment was real. The outcome was nothing. Every subsequent tool evaluation carries an unspoken question: how long before this one goes away too?
The overengineering scar. The team evaluated a platform that could solve their problem — but required a migration, a six-month implementation, and a budget they could not justify. They went back to spreadsheets. Not because spreadsheets work well, but because spreadsheets do not promise anything they cannot deliver.
The staleness scar. A reporting initiative that required manual data aggregation produced outputs so slow they were stale on arrival. The team abandoned it. This is not scepticism about AI. It is a belief, earned through repetition, that current information is structurally impossible to maintain.
What the Scar Does
The visible effect is resistance. The deeper effect is that scarred teams build coping mechanisms that look like rational decisions — and are.
Teams develop hard validation rules: if they do not have the capacity to check the output themselves, they will not use the tool. This is not a philosophical position about human oversight. It is a constraint driven by liability and experience. If the team has no time to review what an agent produces, the agent stays unused.
These are not irrational responses. They are reasonable adaptations to an environment where technology has repeatedly failed to deliver on its promises.
What This Means
The first conversation about AI is never really about AI. It is about what the last thing promised to do — and did not.
The instinct is to lead with capability: what the agent can automate, how much time it saves. But a team carrying scar tissue hears capability claims the way someone who has been burned hears "this time will be different." The information may be accurate. The response is not shaped by information.
What works is harder than a capability pitch. It means asking what specifically went wrong last time, and showing how this proposal is structurally different from what failed. The scar is not the obstacle to adoption. The scar is the information.
Wyrework helps organisations rewire how work gets done — one workflow at a time. See where your governance gaps live at wyrework.ai/tools.
Source: S&P Global Market Intelligence, Voice of the Enterprise: AI & Machine Learning (2025).