The AI J-Curve

Things always get worse before they get better. Surviving the dip in performance after Artificial Intelligence (AI) implementation in an organisation.

The Idea That Wouldn’t Stay Put

A few years back, I was in the audience for a fairly routine PowerPoint presentation at work. Slide after slide went by before until something caught my eye. A familiar shape. A curve. My curve!

Back in 2005 I had combined elements from the Changefirst method for People-Centred Implementation (PCI®) with the Kübler-Ross and Four Stages of Competence theories (developed in the 1960s) to explain a phenomenon I had witnessed time and again as an implementor of technology-driven business change: that stakeholders expect immediate performance improvement, but things always get worse before they get better. There is a dip in performance, before it rises to the new and higher state.

And there it was: the j-curve I had created years earlier, now sitting confidently in someone else’s deck. In small, almost apologetic text, it read: “adapted from David Viney”. Amusingly, the presenter knew me personally but had failed to notice my name in the image. Neither did anyone else in the audience. Resisting the tempation to interrupt, I instead quietly Googled that evening. Hundreds of references and a small but growing number of academic citations. An idea had escaped its confines and diffused into human knowledge; applied across digital workplace design (Meier et al., 2015), construction management (Aibinu & Papadonikolaki, 2020), nursing and healthcare (Young & Snowden, 2020), and - more recently - agile software development.

The AI J-Curve

Perhaps inevitably, the model is now providing useful to those working on the funding and adoption of artificial intelligence. For example, the Tony Blair Institute for Global Change used the J-Curve as a framework for their November 2025 paper 'Rebooting the UK's Tech Diffusion Ecosystem' (TBI, 2025) and Investment manager Pender Fund cited the framework in their recent paper, The Intelligence Economy (Pender, 2026).

To re-use this image, please attribute https://alchemy.consulting/ai-j-curve

The AI J-Curve summarises what happens during the transition from the current state (i.e. before AI was introduced) to the desired (end) state (i.e. when the benefits are fully realised). Whilst stakeholders expect an almost immediate improvement in performance (the red line in the chart), the reality is that performance falls initially (the blue line), as you have introduced instability and change into a system that, albeit sub-optimally, operated on comfortable and well-established habits.

In the Disruption Zone, immediately after the introduction of AI, the initial dip in performance is likely to be substantial; as McElheran et al. (2025) found in their evaluation of AI implementation in the US Manufacturing industry. This is particularly true when leaders are overconfident in the 'near-magical' properties of AI and fail to do enough to understand and manage the business change; as the RAND Corporation found in their 2024 study of why AI Projects were failing at a higher rate than for previous waves of technological innovation.

The leadership challenge is how to mitigate this dip in performance – sometimes called the 'valley of despair' – through awareness-building, governance & controls, reward mechanisms, and operating model redesign. This is reflected by the green line on the chart. However, it is also vital to manage stakeholder expectations; so they do not expect immediate, unrealistic performance improvement, bur rather hold their nerve –and support – until the desired state is reached and the benefits realised.

With good change management, the temporary impact of AI on performance is mitigated faster, and the Recovery Zone is shorter, before Breakthrough Zone productivity gains are attained.

Is AI Different to Previous Technology Waves?

This is a very interesting question. At WPP, I led the build-out of the Advertising Industry's first end-to-end Agentic AI Platform. 70,000 users across 110 countries in hundreds of fiercely independent , individually famous, and occasionally feral marketing agencies. Each had their own autonomy, budget, teams, and tools to defend. I am not going to say this wasn't a challenge. It was. But I would not say there was anything instrinsically different in terms of the J-Curve itself. It was still a new technology introduction and the temporary dip was expected and managed accordingly.

However, whilst the curve is not different, I would certainly say the scale of the change management challenge, the level at which it must operate, and the kinds of interventions required are at an order-of-magnitude higher than for more traditional, discrete, and self-contained technology projects. There are three particular differences I would point to, when compared to previous waves:

AI is Simultaneously Within & Across

Unlike previous technology waves, we have Agentic AI being introduced within processes at the same time Generative AI is impacting across all processes and, most importantly, in the manual workarounds and bridges that all too often exist along and between processes; a space I call the ‘informal layer’. Conventional change management tools like Prosci or ADKAR were not designed to manage this multi-dimensionality,

AI Functional Boundary Dissolution

Previous technology transformations disrupted roles within otherwise stable functional disciplines. A purchase ledger clerk was still a clerk after ERP introduction. They just had a better tool to use for the same job. By contrast, AI is flattening hierarchies and simultaneously dissolving the boundaries between disciplines themselves. A marketing director with access to Gen-AI can produce a credible technology roadmap for her marketing tech stack without involving their IT business partner. The CIO can return the favour, however, with a plausible go-to-market product strategy that requires no help from marketing. Everyone is now the instant expert… at everyone else’s job!

The Inverted Training Problem of AI

The Change Manager weapon-of-choice has always been user training. But AI inverts the problem. Instead of training the purchase ledger clerk to use a new tool, increasingly that same clerk is now training an AI to make their own job fully redundant. It's pretty difficult to motivate a user to engage in a training process which, by any measure, is a rational act of self-harm. As David Autor of MIT put it (in 2024), workers with 'foundational training' can leverage AI to do 'higher-value work'... but if a Nurse Practitioner can do much of the work previously reserved to a Doctor, then perhaps we will need less overall roles, people, and job posts in healthcare as a whole!

Board Meeting - image (c) Werner Pfennig

Change Management at a Higher Order of Magnitude

At this stage in my career, I increasingly work on an Interim or Fractional basis; helping board leaders navigate technology change. My main message is that AI will transform individual working lives, revolutionise businesses, and further disrupt economies and society. It also bears on the most fundamental question of all: what does it mean to be human in the age of AI? We need real vision and leadership to meet this moment, not sound bites for investors. Similarly, change interventions need to be board-led:

Informed, Board-level Governance

Arthur C. Clarke famously said ‘any sufficiently advanced technology is indistinguishable from magic’. Well, that may be a nice turn of phrase, but it was intended to be ironic, not a board instruction! I remain astonished at how many leaders expect AI to operate like Wikipedia, rather than as a probabilistic and generative tool; inherently subject to bias, hallucination, and miscalibrated confidence. Leaders need to get their hands dirty, and take time out to fully educate themselves in the workings, risks and opportunties. The Governance of Change needs to sit at board level. Why? Well, when AI diffuses change across an entire ship, the cockpit is the only place you can maintain coherence and grip.

Holistic Operating Model Design

In a similar vein, when AI is flattening hierarchies, eroding roles, and dissolving the boundaries between functions, it is vital for the Board itself to take a keen interest in how the organisation as a whole is being reshaped. Normally, the target operating model and organisation design might safely be delegated to each function. This is not such a time. Increasingly, Regulators expect to see a named C-Suite executive with overall AI accountability. And the EU AI Act, in particular, places a particular focus on HR / People platforms (used for hiring, performance management, and progression).

Values-based, Responsible AI 

In 1896, the UK suffered its first road traffic accident death. The coroner famously hoped "such a thing would never happen again", but now over one million people across the world die every year. The motor car has brought both great benefits to humanity and great harm, simultaneously from the same source. However, no one has ever seriously suggested we should go back to the horse and cart. Instead, we erected an entire safety industry; including everything from traffic lights to air bags, vehicle checks, and seatbelts. Like all truly transformative technologies, AI will be coterminously positive and negative for humanity; often in unforeseen ways. A good board needs to a strong position on why AI is positive for their business, and how the impacts will be mitigated, then lead the serious work required to properly balance the two.

Final Thoughts

It's been a genuine pleasure to see this model picked up by fellow practitioners in the field of AI; at a time reminiscent of the DotCom period when it was first conceived. It's also been lots of fun commercialising it and putting it to work more often in my own evolving practice, alongside other proprietory tools like the Value Flow Sankey (for AI Business Cases) and MASTER-AI™ Framework (for AI Governance).

Need some help?

I am now working as an independent, fractional consultant; advising clients on AI, Platforms, Cyber, and Change.

Learn More

Note: You can still read the Intranet Portal Guide (where the J-Curve was first popularised) on the Internet Archive, borrow it from the Open Library, or read excerpts on Google Books.

© David Viney 2003-2026. Licensed for re-use with attribution under CC BY 4.0

Preferred Attribution Code Snippet (HTML)

The <a href="https://alchemy.consulting/ai-j-curve”>AI J-Curve</a>, by <a href="https://www.david-viney.me/">David Viney</a>, licensed under <a href="http://creativecommons.org/licenses/by/4.0/">CC BY 4.0</a>

For Academic Citations (APA style)

Viney, D. (2026). The J-Curve of Change: A Practitioner Framework for Managing Organisational Performance During Technology-Driven Transformation. SSRN Working Paper. https://doi.org/10.2139/ssrn.6527978 

---

https://alchemy.consulting/ai-j-curveLike the Post?

https://alchemy.consulting/ai-j-curve