implementai.today Free risk diagnostic

Free. Ungated. Nothing stored.

AI project risk diagnostic

Eight questions about your project, each one asked because a named study measured or identified that factor. You get a risk-factor load, the failure modes your answers point to, and the research behind each. It will not tell you whether your project will fail, because nothing can.

The questions below, and the study behind every one of them, are the useful part. They work whether or not you ever press the button.


AI project risk diagnostic

Read this first

This is not a validated statistical model and it cannot predict whether your project will fail. It is a structured way to check your project against the factors that published studies actually measured. Every question below exists because a named study measured that factor, and each one shows you which. Nothing is stored, nothing is sent anywhere, and there is no sign-up.

  1. Question 01 Where did this project come from?

    Why this is asked: RAND interviewed 65 data scientists and engineers. 84% of them named a leadership-driven cause, such as optimising for the wrong business problem, as the primary reason AI projects fail. That is what 65 experts cited, not an audit of real failures, and RAND lists "misunderstanding or miscommunicating the problem" first among its five root causes. RAND 2024

  2. Question 02 Is there one business metric, with a current baseline number, that this must move?

    Why this is asked: Gartner lists "unclear business value" among the reasons it forecasts generative AI projects will be abandoned after proof of concept. IBM's survey of 2,000 CEOs found only 25% of AI initiatives had delivered the ROI expected of them. Gartner 2024

  3. Question 03 Does the data this needs already exist, and can the team reach it today?

    Why this is asked: Gartner names "poor data quality" first among the causes in its forecast that at least 30% of generative AI projects would be abandoned after proof of concept. Gartner 2024

  4. Question 04 Who owns this once it is live?

    Why this is asked: Leadership-driven failures are the first of the five root causes RAND names, and the category 84% of its interviewees pointed to. An unowned system is an organisational condition, not a technical one. RAND 2024

  5. Question 05 Does it change work people already do daily, or ask them to adopt something new?

    Why this is asked: MIT's GenAI Divide report found 95% of organisations getting zero return, and names the cause plainly: "The core barrier to scaling is not infrastructure, regulation, or talent. It is learning." Tools that do not fit an existing workflow do not get the feedback they need to improve. MIT Project NANDA 2025

  6. Question 06 If the pilot works, what happens next?

    Why this is asked: Gartner's forecast is specifically about abandonment after proof of concept. McKinsey's 2025 survey found only 7% of organisations had fully scaled AI, and nearly two thirds had not begun scaling at all. McKinsey 2025

  7. Question 07 What are you actually building?

    Why this is asked: MIT reports that 60% of organisations evaluated custom or vendor-sold enterprise tools, 20% reached a pilot, and 5% reached production. General-purpose tools cross into real use far more often. Building bespoke puts you in the harder distribution. MIT Project NANDA 2025

  8. Question 08 Is there a hard spending cap, and does a named person get alerted when it is approached?

    Why this is asked: Gartner names escalating costs among the reasons it forecasts over 40% of agentic AI projects will be cancelled. One practitioner reported an unsupervised agent cron job spending $24.88 in two days with no cost guards and no human review. Gartner 2025

How the scoring works, in full

Each answer carries a weight from 0 to 3. The score is the sum of your answers as a percentage of the maximum 24. A failure mode is shown when any of the questions that trigger it was answered at weight 2 or above, and the three with the most triggers are displayed.

The weights are ours, not the studies'. No published research assigns a risk weight to "a committee owns it" versus "one named person owns it". The studies establish that these factors matter. The ordering of severity is a judgement, and you should feel free to disagree with it. We publish the rule so you can.

There is no hidden model, no calibration data, and no benchmark population. Calling this output a probability would be a lie, so we call it a risk-factor load: the proportion of the documented risk conditions your project currently carries.

Why it is not gated

The nearest comparable resources put a spreadsheet behind an email form. That is a lead-capture mechanism wearing a tool's clothing, and the audience for this material has learned to read it that way. We have nothing to sell you, so there is nothing to gate.

Questions, answered

Is this a validated statistical model?

No. It cannot predict whether your project will fail and it does not try. It is a structured way to check a project against factors that named studies measured or identified. The weights are a judgement about risk ordering, not a finding from any study.

What does the diagnostic measure?

Eight factors: where the project came from, whether a business metric with a baseline exists, whether the data exists and is reachable, who owns the system after launch, whether it changes existing daily work, whether there is a funded route to production, whether you are buying or building bespoke, and whether spending is capped and alerted.

Is it free, and do I have to sign up?

It is free and there is no sign-up. Nothing is stored and nothing is sent anywhere. The scoring runs entirely in your browser.

Why should I trust the questions?

Each question shows the study it comes from and links to it. Where a study is a forecast rather than a measurement, or blocks automated readers, we say so on the source itself.


We have run no surveys and delivered no AI projects. Our authority is that we read the primary research and show our working.

Next: what the failure rate actually is, and which of the numbers you have seen were never published by anyone.