What Is the Durable AI Index?

The Durable AI Index is 10dem's scoring framework for deciding which AI use cases are worth building. It rates every candidate on three axes: Impact (how much business value it creates), Feasibility (how hard it is to build and integrate), and Stickiness (whether the team and the process will actually adopt it and keep using it). Most prioritization tools stop at the first two. The Index adds the third because Stickiness is what separates a pilot that demos well from one that survives to month six. A high score on all three predicts durable value. A gap on any one predicts a stall.

Why does AI need its own prioritization framework?

Standard project scoring was built for software with a predictable adoption curve: you ship a feature, people use it because it is in their workflow, done. AI breaks that assumption. An AI tool can work technically and still be ignored, worked around, or quietly abandoned, because it changes how people make decisions and who they trust. MIT found that 95% of AI pilots deliver no measurable business impact. That is not a technology failure. Most of those pilots ran, produced output, and demoed fine. They failed at the part scoring frameworks never measure: whether the work actually changed.

The Durable AI Index exists to make that invisible risk visible before you spend the budget. It forces a conversation about adoption at the selection stage, not after the tool is already built and sitting unused.

What does each axis of the Durable AI Index measure?

The three axes are deliberately simple to score, because a framework people will not run is worse than no framework.

Impact asks: if this works, how much does it move a number the business cares about? Revenue, margin, cycle time, error rate, capacity freed. You are scoring the size of the prize, not the coolness of the technology. A chatbot that shaves a little time off a task nobody does often scores low here even if it is technically impressive.

Feasibility asks: can we actually build and integrate this with the data, systems, and skills we have? This covers data quality and access, model reliability for the task, integration into the systems where work happens, and security or compliance constraints. A use case that needs clean data you do not have is low Feasibility, no matter how valuable it would be.

Stickiness asks: will the people whose work this touches adopt it, trust it, and keep using it after the novelty wears off? This is the axis everyone ignores, so it gets its own section below.

You score each axis, and the pattern of scores tells you what to do. High on all three: build it. Strong Impact but weak Feasibility: a sequencing problem, fix the data first. Strong Impact and Feasibility but weak Stickiness: the most dangerous case, because it will demo well and then die.

Why is Stickiness the axis that actually matters?

Stickiness is the differentiator because it predicts the failure mode that the other two axes cannot see. Impact and Feasibility both describe the tool. Stickiness describes the people and the process around the tool, and that is where AI value actually leaks out.

Consider what low Stickiness looks like in practice. The output lands in a system nobody checks. The tool asks people to change a habit they have held for a decade. It produces answers the team does not trust enough to act on without re-checking, which means it adds work instead of removing it. Ownership is fuzzy, so when it breaks in month three no one is accountable for fixing it. None of these show up in a demo, because a demo is a controlled performance by the one person who wants it to succeed. They show up in month six, when that person is busy and everyone else has drifted back to the old way.

McKinsey found that end-to-end workflow redesign is the number one driver of AI value. That is Stickiness restated: value comes from redesigning the process so the AI is load-bearing, not from bolting a tool onto a workflow that stays exactly the same. Scoring Stickiness up front forces you to ask who changes their behavior, what the new process looks like, and who owns it, before you commit to building anything.

How does the Index prevent pilots that demo well and die?

RAND put the AI project failure rate at roughly 80%, far higher than typical IT projects. The Durable AI Index attacks that number at the point of decision. Here is how it changes what you fund.

First, it kills seductive low-Stickiness projects early. The use case that scores high on Impact and Feasibility but low on Stickiness is exactly the one that gets funded under a two-axis model and then dies. The Index flags it before the money is spent.

Second, it turns a weak Stickiness score into a design brief instead of a rejection. A low score is not always a no. It tells you what has to be true for the project to work: this needs a workflow change, a trust-building step, a named owner, a training plan. You either commit to that work or you do not build the tool. Either answer is better than building and hoping.

Third, it makes prioritization a portfolio decision. Instead of chasing whichever demo impressed a stakeholder last week, you can rank a whole backlog and fund the use cases that score across all three axes. That is how you spend a finite AI budget on the projects that will actually stick.

How do you run a Durable AI Index scoring session?

You gather the people who understand the business value and the people who do the work today, in the same room. For each candidate use case you score Impact, Feasibility, and Stickiness, and, this is the important part, you make someone defend the Stickiness score out loud. Who changes their behavior? Do they trust it? What is the new process? Who owns it after launch?

The disagreements are the point. When the person who wants the tool scores Stickiness high and the person who would use it scores it low, you have just found the risk that would have killed the project in production, and you found it in an afternoon instead of in six months. Score the backlog, rank it, and fund from the top.

Frequently asked

Is the Durable AI Index a numeric score or a discussion tool? Both, and the discussion matters more. You assign a score on each axis to rank the backlog, but the value comes from the conversation the scoring forces, especially the argument over Stickiness. A number you cannot defend out loud is not a real score.

How is this different from a standard impact-versus-effort matrix? An impact-versus-effort matrix is essentially Impact and Feasibility with no third axis. It works for conventional software because adoption is largely assumed. AI breaks that assumption, so the Index adds Stickiness to measure the adoption risk those matrices leave out.

What if a use case scores low on Stickiness? Treat it as a design brief, not a rejection. A low Stickiness score tells you the process redesign, trust-building, and ownership work required to make it durable. If you are willing to do that work, build it. If not, do not, because the tool will not stick.

Who should be in the room when you score? Both the people who own the business outcome and the people who do the work the AI would touch. If only the sponsors score it, Stickiness gets rated by the people least likely to have to change their own habits, and you lose the whole point of the axis.

Where to take this next

The Durable AI Index is the spine of how we prioritize, and it is the first thing we run in an engagement. If you want to see it applied to your backlog, start with our services, browse use cases by function on our AI-for hub, or get in touch to score a live pilot. To go deeper, read how to prioritize AI use cases and why AI pilots fail.