Almost two-thirds of companies will abandon AI in 2026. Is yours one of them?
If we don’t invest in AI now, we’re going to be left behind — that’s a very popular sentiment among CEOs in Deloitte’s latest report on AI investment and ROI. Companies feel the urgency to incorporate AI into their workflows because of the “everything AI” boom. At the same time, the research demonstrates that many invest in AI without being able to specify its value.
As a result of the often haphazard investment, organisations are expected to abandon 60% of their AI projects already in 2026. As the Gartner forecast shows, the risk of abandonment is especially high when AI-ready data doesn’t support the project.
Interestingly enough, AI failures rarely come down to technology itself. In my practice, I almost never observe weak AI models that just can’t live up to the task. AI mostly fails because companies skip the necessities: data accountability, workflow ownership, risk boundaries, and a good operating model.
Here is a five-step readiness test for leaders to understand whether it’s the right time to incorporate AI features.
Step 1. What decisions will AI influence, and who owns the KPI?
AI only matters if it affects or changes decisions you already make. Name two specific decisions AI will influence, and assign one KPI owner to each. If you can’t tie decisions to an owned KPI, you’ll struggle to prove ROI.
Step 2. Are your datasets ready for AI?
No matter if you already work with an AI tool or are just evaluating your options, you still need to check whether your company has that one key dataset. It needs to be reliable enough to fuel an AI agent. Assign one person or team to own it and keep the data accurate at all times. With no owner and no data management, you’re working with a “data pile”, not a “data product”.
Step 3. Where will AI change the workflow?
Real AI adoption happens when it is built into the workflow and actually changes it: reorders a queue, automates steps, or gives recommendations when decisions are made. You also need to understand who will use it, when, and what they will do differently. Otherwise, it risks staying just another “dashboard with interesting insights”.
Step 4. Do you have an error budget and risk boundaries?
Every model makes mistakes, and you need to be ready for it. For starters, define the errors that are acceptable and those that are not. Make rules for how you’ll detect changes and drifts. Guardrails like these are of enormous help when you establish an error budget, escalation paths, and monitoring, so the system stays safe, whatever happens.
Step 5. Can you measure cost per outcome?
Being AI-ready means treating AI like a product you run every day, not just as a fun one-time pilot. Practically, this means you need to include some budgeting for ongoing improvement (think monitoring, retraining, evaluation, and integration). Besides, you need to measure unit economics tied to the outcomes. Track cost per resolved case, cost per prevented loss, or cost per retained customer, so you can see what works best and what doesn’t.
With due workflow ownership, data accountability, risk boundaries, and a good operating model, your AI investment won’t be lost and forgotten in a year.
