How to Define Criteria for Investing in an AI Project: 6 Eliminatory Criteria and Thresholds for Each Stage
By Pedro Sonda, dooop
To define investment criteria for AI projects, use two layers. First, six eliminatory criteria, without which no project receives funding: a named business owner, a problem with an indicator, a measured baseline, an estimated total cost, classified risk, and a defined stopping criterion. Then, classificatory criteria (value, feasibility, risk, and alignment) to rank those that pass. Each project stage (test, pilot, operation, scale) also has an evidence threshold that releases the next investment. Written criteria before evaluation reduce decisions driven by enthusiasm.
Updated October 2026.
Why Explicit Criteria
| Data | Source |
|---|---|
| Unclear value, rising costs, inadequate risk controls, and poor data are among the causes of generative AI project abandonment | Gartner, 2024 |
| 65% of CEOs say they prioritize AI use cases based on return calculation; still, only 25% of initiatives delivered the expected return | IBM, 2025 |
| 63% of organizations with high AI maturity perform financial return analysis | Gartner, 2025 |
Interpretation: Calculating return is not enough if basic conditions (owner, baseline, full cost, risk) are not verified. Eliminatory criteria exist for this purpose.
dooop Framework: The 6 Eliminatory Criteria
| # | Criterion | Why It Eliminates |
|---|---|---|
| 1 | Named Business Owner | Without an owner, no one operates or is accountable for the result |
| 2 | Problem with Indicator | Without an indicator, there is no way to know if it worked |
| 3 | Measured Baseline | Without a "before," there is no comparison with the "after" |
| 4 | Estimated Total Cost | License is the smallest part; people, data, integration, and review usually weigh more |
| 5 | Classified Risk | Without classification, it is unknown which controls to require |
| 6 | Stopping Criterion | Without it, the project can drag on indefinitely |
All six are verifiable on a one-page form; see How to Define the Success of an AI Project Before Starting.
The Classificatory Layer
Among projects that pass the six criteria, investment order comes from scoring value, feasibility, risk, and strategic alignment. The scoring method is detailed in How to Prioritize Artificial Intelligence Use Cases.
dooop Framework: Thresholds by Stage
| Stage | Investment Released | Threshold to Pass to Next Stage |
|---|---|---|
| Test | Small, weeks | The solution works with real company cases, with an acceptable accuracy rate |
| Pilot | Moderate, months | The indicator improved against the baseline; users adopted it; risks are under control |
| Operation | Full for one area | Cost per unit of result within forecast; owner and controls functioning |
| Scale | For other areas | Result sustained over a full cycle; data and support handle the volume |
Thresholds must be written before each stage begins. Changing the threshold after seeing the result is the most common way to keep alive a project that should stop; see When Should a Company Stop an Artificial Intelligence Project?.
Different Criteria by Nature of Investment
Efficiency projects may have faster return thresholds; growth or learning projects (options) may have intermediate evidence thresholds. The important point is that the threshold is consistent with the nature; see AI, Capital, and Strategy: What Leadership Needs to Connect.
What to Do Now
- Adopt the 6 eliminatory criteria as a condition for any funding request.
- Define the thresholds for the four stages for ongoing projects.
- Review current projects that do not meet any eliminatory criterion.
To structure AI criteria and portfolio, explore the dooop AI Opportunity Discovery.
Frequently Asked Questions
Do eliminatory criteria block small projects? For small projects, the six criteria fit in a few lines. What they prevent is starting without an owner, without measurement, and without knowing when to stop.
Who verifies the criteria? The person responsible for the AI agenda or the committee, before taking the project to an investment decision.
What about exploratory projects without a clear indicator? They can be treated as learning options, with a small budget, short term, and a success criterion in terms of learning. Even so, they need an owner and a stopping criterion.
Also Read
About dooop
dooop is a consultancy based in Caxias do Sul (RS) that prepares leaders and organizations to transform artificial intelligence into business capability through strategy, governance, and training.