The Roadmap Bloat Problem
We are deep into Q1 planning season, and every AI roadmap we review has the same problem: it is too long. Eighteen months of accumulated AI experiments, each with a small team and a vague success metric, all carried forward because nobody wants to be the person who kills a project that might work eventually.
This is how AI programs die. Not from one big failure, but from the slow suffocation of spreading resources across too many half-alive initiatives.
The Kill List Exercise
Every AI roadmap review should start with a kill list. Before discussing what to add, discuss what to stop. The criteria are simple:
- Has it moved a business metric in the last 90 days? Not a technical metric. Not "model accuracy improved." A real business metric: revenue, cost reduction, time saved, customer satisfaction. If the answer is no, it is a kill candidate.
- Is there a clear path to production in the next 90 days? Projects that have been "almost ready" for two quarters are not almost ready. They are stuck. Either resource them fully or kill them.
- Would you fund this from scratch today? Sunk cost bias keeps projects alive long past their expiration date. Ask yourself: knowing what you know now, would you allocate new budget to this? If not, the budget currently allocated is being wasted.
Why Teams Resist Killing Projects
The resistance is always the same:
- "We have already invested $X in this." Irrelevant. That money is spent regardless of whether you continue.
- "It just needs a few more months." If a project needed more time 6 months ago and still needs more time now, the pattern is clear.
- "The team will be demoralized." The team is already demoralized. They know the project is not working. Freeing them to work on something impactful is a gift, not a punishment.
- "What if a competitor does it?" If you cannot make it work after 6+ months of effort, a competitor doing it is useful market research, not a threat.
What Killing Enables
Every project you kill frees three things: budget, people, and attention. In a constrained environment, which every AI program is, these are the most valuable resources you have.
We worked with a mid-market client in late 2025 that killed four of its seven AI initiatives. The three remaining projects each got doubled resources. Within 8 weeks, two of them were in production generating measurable returns. The four killed projects had collectively generated zero returns in 6 months.
Your AI roadmap for 2026 should be shorter than your 2025 roadmap. If it is longer, you have a prioritization problem, not an ambition advantage.