AI Adoption Déjà Vu: Why We’re Repeating the Same Mistakes Again (and What to Do Instead)

Every few years, a new technology arrives that promises to transform everything.

• ERPs were going to standardise the enterprise.

• CRMs were going to “own the customer”.

• RPA was going to automate the back office.

• Big data and machine learning were going to make every decision “data-driven”.

• Now it’s AI’s turn.

What’s striking, looking across these waves, is not the technology itself – it’s how predictably organisations mishandle it. The tools change. The pattern doesn’t.

Right now, with AI, a lot of organisations are once again buying solutions before they’ve

understood the problems. They’re automating processes they don’t fully understand.

They’re optimising things that should probably be redesigned from scratch. And they’re surprised when value doesn’t show up.

We’ve been here before. The question is: does leadership want to learn this time, or pay to repeat the lesson?

1. We’ve Been Here Before

If you want a reminder that smart people with big budgets can still get technology adoption badly wrong, you don’t have to look far.

IBM Watson for Oncology

Once held up as the future of AI in healthcare, Watson aimed to read medical literature and assist doctors with treatment decisions. The ambition was huge; the reality was messy. Watson struggled with real-world complexity, produced questionable recommendations, and proved difficult to integrate into clinical workflows. Billions were invested before IBM sold off much of Watson Health.

Zillow Offers

Zillow attempted to turn its Zestimate algorithm into a home-buying business. On paper it worked. In a shifting market, the algorithm overvalued homes while the company scaled aggressively. The result: overpaying for thousands of homes and shutting the programme with losses exceeding $500 million.

RPA Hype and Disappointment

Numerous surveys have shown high failure or stall rates in RPA programmes. Many organisations tried to automate poorly understood processes, underestimated edge cases, or treated automation as a shortcut rather than part of broader operating model change.

These failure stories are not about “bad tech”. They reflect human and organisational patterns:

- Leading with marketing instead of evidence

- Choosing the wrong problems to solve

- Ignoring process complexity

- Underestimating change management

- Scaling before learning

AI has simply given us a more powerful tool with which to repeat the same mistakes.

2. The Core Pattern: Solution First, Problem Later

At the heart of most failed tech adoptions is a simple pattern:

Leaders perceive a problem, fall in love with a solution, and only then ask whether it fits the reality on the ground.

The sequence is familiar:

- A leader feels pressure and concludes, “We need to do something with X technology.”

- Vendors arrive with polished demos and industry case studies.

- A tool is bought before the business fully defines the problem.

- Only after contracts are signed does real discovery begin: mapping processes, talking to users, analysing data.

By then, the organisation is constrained by a decision it was not ready to make.

AI adoption research consistently highlights the same barriers that affected past digital initiatives: unclear objectives, poor data readiness, and weak change management.

3. Old Mistakes, New Tools: How AI Is Replaying the Same Movie

Recent studies across digital, automation, and AI adoption point to the same recurring organisational behaviours. Surveys highlight that most AI and automation challenges are people- and process-related—not technical—mirroring patterns seen in RPA and earlier digital initiatives. Research also shows high failure rates when organisations automate without understanding real processes, reinforce outdated workflows (“paving the cow path”), or skip robust as-is analysis. These themes are reflected in the four mistakes below.

Mistake 1: Decision-Maker Intuition Replaces User Reality

Leaders often feel close to the work without actually being close. Multiple studies highlight that user-related issues and weak stakeholder engagement are among the top reasons digital and AI initiatives fail, reinforcing this disconnect. They perceive problems, but the lived reality for users or customers is more nuanced. The result is low adoption or workarounds.

Mistake 2: Automating Processes That Aren’t Understood

Many organisations still map only the “happy path”. Research on automation and RPA consistently shows that failing to account for exceptions and real-world variability is a leading cause of automation breakdowns. Exceptions, judgement calls, system quirks, and workarounds are overlooked. AI designed for the idealised version of a process breaks when meeting real-world complexity.

Mistake 3: Incremental Efficiency Instead of Process Redesign

Sometimes a process shouldn’t be automated — it should be replaced. Modernisation research frequently warns against “paving the cow path,” where organisations embed outdated processes into new technology without rethinking them. Yet businesses often reach for incremental efficiency because redesign feels riskier.

Mistake 4: Tech Implementation Without Understanding the As-Is

Transformations move from as-is to to-be. Process mapping research reinforces that without a clear understanding of the current state—including pain points, constraints, and exceptions—technology adoption becomes guesswork. Without a clear understanding of the starting point, the path to the desired state is guesswork. With AI, small misunderstandings can have outsized operational consequences.

4. Why This Keeps Happening (and Why Leaders Are Complicit)

It’s tempting to blame vendors or IT. But leadership behaviour drives many of these mistakes.

Truth 1: Hype Is Attractive

Boards and executives feel pressure to “do something with AI.” Announcing tools is visible and immediate. Discovery is slower and harder to showcase.

Truth 2: Discovery Is Undervalued

Process analysis, problem definition, and operating model design are perceived as overhead rather than core work.

Truth 3: Uncertainty Is Uncomfortable

It’s easier to commit to a tool than say, “We don’t yet know where AI can add value — we’re going to explore first.”

Truth 4: Accountability Is Fuzzy

No single person typically owns the bridge between strategy, technology, and operations. AI becomes everyone’s and no one’s problem.

AI’s long-term potential is very real — similar to early RPA, analytics, and machine learning, which took years to mature. But the current hype can distract leaders from thoughtful adoption. Meanwhile, risks around privacy, hallucinations, compliance, and governance are material and must be addressed.

5. A Different Playbook: Problem-Led, Process-Aware AI Adoption

1. Start With the Problem

Before talking tools: - Define the problem clearly. - Identify root causes. - Document a concise problem statement and objectives.

2. Close the Gap Between Leaders and Users

Engage: - Frontline staff - Middle management - Customers (where appropriate)

User insight reduces the risk of solving imaginary problems.

3. Map and Challenge the Process

Before implementing AI: - Map the end-to-end process, including exceptions - Understand why the process exists - Decide whether it should be eliminated, redesigned, simplified, or automated

4. Treat AI as One Option, Not the Strategy

Consider alternatives such as: - Policy changes - Training - Process redesign - Integration - Traditional automation

AI may still be the right choice — but for the right reasons.

5. Invest in Change Management

AI initiatives fail for many of the same reasons broader transformations do — and research consistently shows that underestimating change management is one of the most persistent and costly mistakes. Even with the right technology and well‑defined processes, adoption collapses if people aren’t brought on the journey.

In practice, underestimating change shows up as:

- Announcing new tools without explaining the problem they solve

- Assuming training alone will drive adoption

- Overlooking frontline concerns or workarounds

- Ignoring the behavioural and cultural shifts needed to embed new ways of working

- Failing to provide sustained leadership visibility and reinforcement

Stronger change management means:

- A clear narrative about why change is needed and how it helps users

- Early involvement of SMEs and frontline teams in shaping requirements

- Practical training focused on new behaviours, not just system features

- Ongoing feedback loops and iteration after go‑live

- Leaders consistently reinforcing expectations, removing blockers, and celebrating progress

Good change practice isn’t an add‑on — it is often the difference between AI becoming a strategic asset or an abandoned tool.

6. Establish Guardrails and Governance

Decide:

- Who approves AI-generated content?

- How data privacy is protected

- Rules for using external models

- How AI outputs are monitored and corrected

6. Five Questions to Ask Before Buying Your Next AI Tool

1.      What specific problem are we trying to solve, and how do users describe it?

2.      Have we mapped the real process, including exceptions?

3.      What does success look like, and how will we measure it?

4.      Have we considered non-AI options, and why is AI best for this case?

5.      Who owns the change — process, data, risk, governance?

If you can’t answer these, you are not ready for AI. You are ready for discovery.

7. Closing Thoughts

AI is not a passing fad. Like digital, automation, and data before it, AI will reshape how value is created. But the organisations that win will be those that:

·       Start with real problems

·       Redesign processes, not just automate them

·       Pair transformation leaders with business SMEs

·       Invest in change management and governance

We’ve already paid for a masterclass in how not to adopt technology. The question now is simple:

Do you want AI to be your next expensive lesson — or your first real application of what you’ve already learned?

If you’re exploring AI or concerned about repeating old patterns, feel free to connect with me on LinkedIn or reach out via my website. I help organisations design and deliver transformations that work in the real world.