How a Director Should Evaluate an AI Proposal: 7 Questions Before Approval
By Pedro Sonda, dooop
To evaluate an AI proposal, a director should ask seven questions: (1) what business problem does it solve; (2) how much is solving this problem worth; (3) what evidence exists that the solution works in our context; (4) what is the true cost, including people, data, integration, and operation; (5) what risks does it bring to customers, data, and reputation; (6) who will operate and be accountable for the solution after implementation; and (7) when and by what criteria will we decide whether to continue. A convincing demonstration does not replace these answers. Without them, the proposal is not ready for approval.
Updated October 2026.
Why Evaluate More Rigorously
| Data | Source |
|---|---|
| At least 30% of generative AI projects would be abandoned after proof of concept due to poor data, inadequate risk controls, rising costs, or unclear value; costs are less predictable than other technologies | Gartner, 2024 |
| 64% of CEOs acknowledge that fear of falling behind leads to investing in technology before understanding its value | IBM, 2025 |
| Gartner estimates that among thousands of vendors claiming agentic AI, only about 130 are genuine; the rest practice "agent washing," rebranding existing products | Gartner, 2025 |
| 32% of organizations have refrained from purchasing at least one software or feature because they built it internally using agentic programming tools | McKinsey, 2026 |
Interpretation: haste, vendor marketing, and unpredictable costs increase the risk of approving fragile proposals. Alternatives have changed: some solutions previously purchased can now be built internally.
dooop Framework: The 7 Questions
| # | Question | Good Answer | Warning Sign |
|---|---|---|---|
| 1 | What business problem does it solve? | Specific problem, with an owner and current metric | "It will increase productivity" |
| 2 | How much is solving it worth? | Estimate with explicit assumptions and range (pessimistic, likely, optimistic) | Single number without assumptions |
| 3 | What evidence exists in our context? | Test with our data and processes, or a verifiable comparable case | Only demonstration with vendor data |
| 4 | What is the true cost? | Total cost: licenses, consumption, integration, data, people’s time, human review, maintenance | Only license price |
| 5 | What risks does it bring? | Proportional risk assessment; data involved; error plan | "The tool is safe" |
| 6 | Who operates and is accountable? | Business owner, support, and recurring budget defined | "The vendor takes care of it" |
| 7 | When do we decide whether to continue? | Date, indicator, and stop criteria defined before starting | Project without defined end |
About Question 4: The Hidden Costs
The proposal usually shows the technology cost. The cost that weighs most is often people: managers’ time to redesign processes, training, human review, and monitoring. BCG observed that in companies generating the most AI value, about 70% of effort goes to people and processes (BCG, 2024).
About Question 7: Approve in Stages
Instead of approving the entire project, approve the next stage with criteria to release the following one. See How to Decide if an AI PoC Should Continue and How to Define AI Project Success Before Starting.
How to Decide Based on the Answers
| Result | Decision |
|---|---|
| All 7 answers good | Approve, with review on agreed date |
| Questions 1, 2, and 7 good; others incomplete | Approve only a limited test to generate evidence |
| Question 1 or 2 unanswered | Return; proposal lacks business case |
| Question 5 or 6 unanswered, with sensitive customers or data | Do not approve until risk assessment and owner exist |
Buy, Build, or Wait
For each proposal, also ask: is there a simpler alternative (an already licensed tool, a process change)? Could it be built internally? Does waiting six months change anything? To estimate value more carefully, see How to Estimate the Value of an AI Initiative.
What to Do Now
- Turn the 7 questions into a mandatory form for every AI proposal above a defined value.
- Apply it to the proposal under review closest to approval.
- Define who participates in the evaluation: business, finance, technology, and risk.
To generate AI proposals that already have a business case, learn about dooop’s AI Opportunity Discovery.
Frequently Asked Questions
What if the vendor refuses to test with our data? It is a sign to consider. Limited tests with controlled data and confidentiality agreements are common practice and reduce decision risk.
Should internal proposals undergo the same scrutiny? Yes. The evaluation size can be proportional to value and risk, but the seven questions apply to any proposal.
Who should have the final say? The director of the area that will receive the results, with participation from finance and risk owners. Technology assesses feasibility but does not decide alone.
Also Read
- How to Assess the Feasibility of an AI Use Case
- How to Decide How Much to Invest in Artificial Intelligence
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.