Why the Most Successful AI Strategies of 2026 Start with a ‘Stop-doing’ List

How many new AI solutions, pilots, or features have you seen in the past few weeks? Dozens, maybe even hundreds? In today’s landscape, technology leaders might feel overwhelmed by the sheer number of AI options and uncertain about how to implement them effectively.

But when the market is so crowded, there is another approach to achieve results. Instead of listing all the AI initiatives you should try—there is already too much of that—start with a ‘stop-doing’ list. It’s a good way to cut through the noise and refocus efforts on actions that can generate real business value.

In this article, we’ll look at a list of seven ‘stop-doing’ things for technology leaders who want to make AI transformation actually work.

1. Stop launching AI initiatives without a clear way to measure value

Too many AI pilots begin with enthusiasm and quietly fade out without delivering results. Enthusiasm is quite natural—AI offers real potential to improve productivity—but without a clear definition of impact, even the most promising pilot struggles to move forward.

Before starting any AI initiative, you need to define how success will be measured. If you don’t know how to evaluate performance, estimate long-term cost, or assess operational impact, there’s no reliable way to prove business value. A practical way to approach this is to measure impact across three levels.

• Technical level. Here, you need to see how the system performs in real conditions: how fast it responds, how often it succeeds, and how performance evolves over time. Key metrics here include latency SLO, success/containment rate, model precision per task, drift/degradation patterns, and rollback triggers.
• Economic level. At this level, the focus shifts to cost and efficiency. This includes cost-to-serve (per request, per model, per use case), cost avoided, reduced errors or faster decisions, and time-per-task or cycle time reduction.
• Business level. Finally, answer a straightforward question: Does the solution improve your business outcomes? Checking out resolution rates, customer outcomes, process throughput, revenue or margin impact, and cycle time compression in engineering or ITSM all help answer that question.

One practical rule to remember: if you can’t track these metrics regularly, they won’t support real decision-making. And with AI initiatives, you need to make weekly check-ins to stay in control.

2. Stop measuring ROAI the same way you measure ROI

Traditional ROI models don’t fully apply to AI, and this is where many leaders get misled.

First, AI introduces variable operating costs. Every interaction with a model has a cost, so value depends on actual usage rather than projections.

Second, AI systems evolve over time. Models degrade, data shifts, and prompts change. Measuring ROAI requires accounting for reevaluation, retraining, and governance overhead—factors that are often underestimated.

Third, AI creates value only when it becomes part of everyday workflows. A highly accurate model that no one uses delivers no return. Low adoption directly impacts ROAI, which is why measuring actual usage is just as important as measuring performance.

3. Stop managing AI as a scattered pipeline of experiments

When every team runs its own pilots, it can feel like progress is happening. In reality, this fragmented approach rarely leads to meaningful outcomes.

A more effective way to go would be to treat AI as a focused portfolio. Instead of dozens of disconnected experiments, group initiatives into a small number of value areas, such as:

• personal productivity,
• process automation,
• IT service management,
• engineering efficiency,
• AI-driven innovation.

This structure makes it easier to see where value is expected and which of your pilots is worth scaling.

See also: How AI is Forcing an IT Infrastructure Rethink

4. Stop ignoring your data foundation

There’s a common saying in technology: you can’t build future systems on a legacy core. That doesn’t mean you necessarily need to rebuild everything from scratch, but it does mean taking a close look at your data before introducing AI.

Even strong AI initiatives often fail because they rely on fragile or poorly governed data. A data readiness audit helps uncover these risks early. Before launching any solution, define:

• data sources,
• quality standards,
• update frequency,
• access and controls.

Getting this right upfront helps prevent 90% of downstream issues that would otherwise surface much later in the process.

5. Stop treating governance as something to add later

Governance is often postponed to keep pilots moving quickly. But this is one of the main reasons those pilots never reach production.

If a system cannot be audited, evaluated, or safely rolled back, it is not ready for real use, regardless of how promising the initial results look. Production-grade AI requires governance from day one. This includes clear model and prompt records, reliable evaluation datasets, continuous monitoring, and the ability to revert changes instantly when needed. Without these elements, scaling becomes difficult and risky.

6. Stop assuming bigger models will solve the problem

For technology leaders, it might be tempting to use the most powerful models available. It might seem that those models will definitely help to achieve better results. The truth is, larger models often introduce higher costs and slower response times without delivering proportional value.

A more effective approach is to match the model to the problem. Larger models should be reserved for tasks that require complex reasoning. For everything else, smaller and more efficient models—combined with strong retrieval and caching—are often a better fit.

7. Stop relying on traditional delivery structures

Large, siloed teams were not designed for the speed and iteration AI requires. They tend to slow down decision-making and introduce unnecessary friction.

AI initiatives benefit from a different model: small, cross-functional teams—often called AI Pods—where 3–5 specialists work alongside AI tools across the entire software development lifecycle.

This setup shortens feedback loops, improves quality, and helps move beyond isolated pilots into systems that are actually used in day-to-day work.

What skills CIOs need to lead AI effectively

Alongside a clear ‘stop-doing’ list, successful AI transformation requires a shift in leadership skills.

• Product thinking. Treat each AI initiative as a product, with clear ownership, defined SLAs, KPIs, measurable outcomes, and cost-to-serve boundaries.
• Engineering literacy. You might not need to write code, but you need to understand how systems behave—how different models compare, how data flows, how retrieval patterns work, and how AI integrates into the software lifecycle.
• Governance fluency. Knowing what to ask for—evaluation baselines, auditability, rollback mechanisms—helps ensure systems are both effective and safe.
• Clear vision. Perhaps most importantly, you need to be able to cut through the noise and choose solutions that will deliver real value. Every AI initiative should answer a few core questions: what does it improve, which KPIs it helps to achieve, what it costs to operate, how fast it performs, and how risks are managed.

Successful AI strategies in 2026 will be defined by leaders’ ability to stay focused and consistent as the market becomes increasingly saturated. In many cases, that means doing less—not adding more to an already overloaded to-do list. Prioritize several simple principles: know how to measure impact, validate your data foundation, and embed governance early. With this focus in place, AI can be integrated into real workflows and become a practical tool that improves how your business operates.