“Would you tell me, please, which way I ought to go from here?”
“That depends a good deal on where you want to get to,” said the Cat.
“I don’t much care where—” said Alice.
“Then it doesn’t matter which way you go,” said the Cat.
“—So long as I get SOMEWHERE,” Alice added as an explanation.
“Oh, you’re sure to do that,” said the Cat, “if you only walk long enough.”
Lewis Carroll, Alice in Wonderland
A common problem
AI initiatives often stall before they create real value. This is often referred to as the ‘POC graveyard’ where exciting initiatives fail to be operationalised or add real value, and instead languish in a prototype state, where the investment in developing them is unable to be realised.
This isn’t because the leaders who are driving this change are careless or because the teams lack technical capability. The usual cause for the initiatives to stall is because organisations attempt to lock in certainty around the solution too early.
In many AI initiatives, particularly where there is pressure for leaders to execute a hastily drawn up strategy, and there is executive fear of being left behind, leaders request a roadmap before they have aligned on the problem, the outcome, or the conditions required to succeed.
In such cases, a roadmap will appear, but it will be lacking the depth of thought and connection that delivering an Ai solution requires. The result: business value is not realised and the roadmap quickly becomes roadkill.
The Anxiety of the Future
Jeff Bezos often referred to the idea of open mental loops and closed mental loops. A closed loop is something that a leader does not need to think about any more – it is either delegated, or someone has committed to execute it on the executives behalf – this could be a delivery group, a project that the executive has governance over, or an external vendor.
An open loop is something that the executive needs to keep thinking about – it takes up a certain part of their mental load. It is still ‘one their plate’. Most executives are extremely time and bandwidth constrained – they have a huge number of demands on their time and mental load. This can be a huge drain on energy and source of stress. The desire to offload this uncertainty is immense.
A roadmap, or better yet, a committed long term plan helps the leader manage this anxiety. They do not have to worry about it (for now). Milestones are laid out and the illusion of certainty is given by the roadmap. The loop can be closed. The roadmap promises certainty, the work is underway and the future appears structured.
In this context the rush to a committed roadmap can be less about reflecting our current understanding and more about managing anxiety and managing mental load.
Roadmaps can be hugely valuable in a volatile technical environment such as AI, as long as we acknowledge the inherent uncertainty in the roadmap, and we do not treat it as a committed plan.
Plans pretend certainty, a good roadmap acknowledges uncertainty. It can be very tempting when someone comes to us and says they have certainty over the future and can manage it. But if this is not the case, we are simply kicking the risk down the road.
A plan assumes we know the sequence, the milestones, and the outcome. A roadmap is supposed to be something different. It is a statement of intent based around our current understanding of the situation. It is a communication and directional guide that evolves as the organization learns and as the delivery evolves.
But when a roadmap is requested too early, it becomes a plan in disguise. It pretends the future is clearer than it actually is.
Assumptions can get locked in
When an AI roadmap is requested prematurely, several false assumptions quietly become embedded.
We have a clear idea of the goal we want to achieve
The scope is well understood and clearly defined
The sequence of work is knowable
The constraints of delivery are stable
It is very rare for any delivery, but particularly for AI Initiatives, that any of these are truly settled. The roadmap therefore reflects guesses rather than alignment. And once written, those guesses start behaving like commitments and become locked in. Any variance from the initial assumption is treated like a failure to deliver according to plan, rather than a learning experience.
When assumptions are locked in too early:
- teams optimize the wrong sequence
- dependencies appear late
- priorities shift
- constraints change
The roadmap then becomes misleading. Not intentionally misleading. But structurally misleading. Leaders start asking, “Why did the roadmap change?” Instead of asking, “What did we learn?” because their primary driver in accepting the roadmap was that it gave them the illusion of certainty.
The Questions That Must Come First
If we want a rigorous roadmap that gives us a good understanding of AI development we must answer a number of questions.
What problem are we actually solving?
What outcome defines success?
What trade-offs are acceptable?
What decisions will we make when new information appears?
Until these questions are addressed, any roadmap will be fragile. Because the destination itself is unclear, and we do not have guiding principles that allow us to understand how we evolve the roadmap over time.
In order for a roadmap to be useful it must address the above questions at a bare minimum, that gives leadership clear principles for how they address the inevitable evolution of the roadmap.
Use the Three Pillars to make your AI Roadmap Great
Any well-defined AI Roadmap has three key pillars ;
- Clear business outcomes
- Access to expertise about how the work will be done
- A strategy or operating model for AI inside the organization
These are the three critical pillars of a well defined roadmap and they need to be in place to enable a rugged roadmap that has a solid foundation but can evolve over time.
A roadmap needs clear business outcomes as an aligning principle as the understanding of the work, the technology, and the scope evolves. It is our underlying ‘why’?
We need expertise familiar with AI technologies to connect the business outcomes with the work. While scope is certain to evolve, we need a well-informed understanding about the capabilities of AI and even if AI is the right solution to solve a particular problem.
And lastly, an overarching AI strategy to manage issues of compliance, risk management, and how AI systems will be operationalised and managed within the organisation is needed. Without a strategy that outlines how we deliver, manage risk around, and operate AI systems, we will hit roadblocks at every turn, and we cannot put together a realistic roadmap of our time to deliver Ai outcomes.
In summary, the temptation to put together a roadmap for AI delivery can be strong – there is pressure in many organisations to ‘just get started’ or to signal progress.
We must do the foundational work to ensure the roadmap is built on a clear idea of business outcomes, a solid technical foundation based on expertise, and a cohesive AI strategy outlining acceptable use, risk management and how delivery and support will occur for AI.
Without this we will find that the roadmap as a tool for mitigating risk is a mirage that leads us into trouble down the road.
SPCT & AI-Native Trainer