In my experience, AI roadmaps do not fail during execution. They fail in three decisions made in boardrooms and leadership offsites, long before the first line of code is written and long before the teams who will be held accountable have any say.
When a company decides to pilot AI, there is an invisible pull toward the high-profile project. The one that makes the quarterly update sound visionary. The one that signals to the market that this is an AI-first company.
This is precisely the wrong place to start. Every new technology carries known unknowns - risks you can anticipate but not fully plan for and unknown unknowns that no one in the company has the expertise to solve. When you pilot on a high-stakes project, these surface in front of your most important customers and most unforgiving stakeholders.
What follows is predictable: early signs of strain appear, but admitting failure on a flagship initiative is politically unacceptable, so leadership doubles down. The hole gets deeper. When the project finally collapses, the search for a scapegoat begins. Teams who worked hard to deliver on a directionless mandate are blamed and let go. The CXOs who made the original call move on to the next initiative.
Pilot AI on a project that can afford to fail - but complex enough to generate real learning. In distribution, the instinct is to pilot on high cost-of-delay items. Resist that. Instead, pick an item that travels through multiple transit points and takes far longer than it should. Can AI reduce both the stops and the duration? This has enough complexity to produce genuine insight, while a stumble costs very little. The real output is not the logistics win - it is your organisation's growing capacity to tackle harder problems next.
Two questions guide good pilot selection:
- Does failure here materially hurt customers or revenue?
- Is there enough complexity that the team will genuinely learn?
If the first answer is no and the second is yes, you have found your starting point.
While pilot projects were being announced, another initiative was running in parallel: the AI reskilling programme. Employees were made to go through two-day workshops. Certificates were distributed. Employees are under the impression that they are part of the company's AI growth plans. They announce the Certificate with great enthusiasm on LinkedIn. Little do they realise that they had climbed only the first rung of a ladder. There are at least half a dozen more rungs before their profile makes the first cut for any meaningful AI role. Meanwhile, boardrooms are drawing lists of employees to be let go - many of them are those with the certificates whose AI ladder has been taken away.
Genuine reskilling starts with the right people. Identify five to ten high performers with a record of solving complex problems through unconventional thinking. Cognitive flexibility matters more here than existing technical credentials. Put them on the low-stakes, high-complexity pilot projects and track the following:
- What approaches were tried and what failed
- Skills and mindsets the work actually required
- How cross-team coordination differed from conventional projects
- The real learning curve — not the brochure version
These people are not just solving a logistics problem. They are generating the blueprint from which the rest of the organisation's reskilling can be designed - grounded in what your specific context demands, not a generic AI curriculum.
Be honest with the broader workforce about what the journey looks like. Invest specifically in those who show genuine drive. Acknowledge the hard truth plainly: despite best efforts on both sides, some will not make the transition. A company that gave people a real opportunity and a fair process can make that call with integrity. One that handed out certificates while writing termination lists cannot.
Before a single AI tool is selected, before a pilot project is named, there is a question that almost never gets asked in the boardroom: is our data actually ready for AI?
AI does not create order. It amplifies what already exists - including the mess. If your customer data lives across three spreadsheets, a WhatsApp group, and someone's memory, no algorithm is going to save you. Tribal knowledge - information passed through word of mouth, institutional habit, or one person's notebook - is not a foundation. It is a liability.
The painful irony is that companies rushing into AI roadmaps are often the same ones that have not done the unglamorous work of standardising their data, documenting their processes, or auditing where critical information actually lives. A simple spreadsheet done consistently beats a sophisticated AI platform sitting on top of chaos.
Before committing to any roadmap, answer three questions honestly:
- Where does our data live, and who controls it?
- Is it complete and reliable enough to draw conclusions from?
- Do we have a baseline process that AI can improve — or are we hoping AI will create the process for us?
If the answers are uncomfortable, the roadmap needs to wait. Fix the foundation first.
What a well-thought-out AI adoption looks like
The companies that emerge stronger from this period are not the fastest movers. They are the ones that asked the uncomfortable questions before anyone else did and had the discipline to fix what they found.
Check your foundation before you name a project. Audit where your data lives, whether your processes are documented, and whether your team has the baseline to build on. If the answers are uncomfortable, that is your starting point - not a roadmap slide.
Pick the pilot where failure is survivable and learning is guaranteed. Use it to build your organisation's capability, not to impress a board. Let your best problem-solvers run it, track everything they learn, and let that experience design the reskilling programme your workforce actually needs.
Be honest with your people about the stakes and the path. The ones who rise to meet it are your most valuable asset in this transition.
The roadmap is the easy part. The foundation, the right project, and the honest conversation with your workforce - that is what most companies skip.