The hidden cost of an AI tool nobody uses

Somewhere in your company there is a tool that was bought with real enthusiasm, demonstrated to the management team, used properly for about three weeks, and has been renewing quietly ever since. Everyone knows which one. Nobody wants to be the person who raises it.

The standard framing of this problem is the licence fee, which is the least interesting part of it. Here is what it actually costs.

The four costs, in ascending order of size

The licence. Real, annoying, and usually the smallest number in this list. If it were the only cost, shelfware would be a rounding error rather than a pattern worth writing about.

The attention already spent. Someone selected it, someone negotiated it, someone configured it, someone ran the training session, and a manager gave it space in a team meeting. In a company of eighty people that attention is the genuinely scarce resource, and it has already been spent whether or not the tool is used.

The credibility. This is the cost nobody puts in a business case and the one that lasts longest. After the second unused tool, the next AI proposal in that company meets a room that has learned something. They will not say "we do not believe you"; they will ask more questions, take longer to decide, and want a smaller pilot. That is a rational response to evidence, and it makes the next genuinely good idea harder to get through.

The opportunity you did not take. The largest by a distance and completely invisible. A company can only run so many changes at once. The one that was occupied by the unused tool was not available for something that would have worked.

Crates stacked and unused. The licence fee is the smallest part of what shelfware costs you.

Why tools go unused, and it is almost never the tool

We have seen the same four causes repeatedly, and only one of them is about the software.

Nobody owned it after purchase. A tool with no named owner degrades immediately, because every tool needs someone to answer questions in week two when the initial enthusiasm has worn off and the first awkward case appears.

It required a process change nobody agreed to. The tool works, but using it properly means the sales team has to enter something they never entered before. That is not a software problem and no amount of configuration will fix it.

The people who had to use it were not in the room when it was chosen. This is the most common cause by a wide margin. A tool selected by management for a team, without that team present, is a tool that will be used exactly as long as someone is watching.

There was no definition of what success looked like. Without one, nobody can tell whether it is working, so nobody escalates when it is not, so it drifts into the renewal cycle.

Only the second of those four is even partly about the product.

The four costs, and who notices each

The pattern is the point: the only cost that appears in the budget is the smallest one, which is why shelfware gets discussed as a procurement issue rather than as the capability problem it actually is.

The number to check this quarter

A concrete exercise that takes an afternoon. List every AI or automation tool the company pays for. Per tool, write down: who owns it by name, how many people used it last month, and what would break if you cancelled it tomorrow.

The third column is the useful one. If the honest answer is "nothing," you have found either a saving or a decision to actually implement the thing properly. Both are better than the current state, which is paying for optionality you are not exercising.

The principle underneath

This is one of the ten principles Radical works from, and it is stated more bluntly internally than most companies would put it on a website: a tool without a team that can run it is shelfware. We build the skill into the people who stay, not into a report that gets filed.

That is also the honest reason we are sceptical of the tool-first sales motion generally. Not because the tools are bad, many are genuinely good, but because a licence transfers a capability to your company only if somebody there can run it. Absent that, you have bought a subscription to a possibility. The same argument, applied to consultancy rather than software, is in Waarom een adviesrapport geen capability is.

What to do with the one you already have

Three options, and "leave it" is not among them.

Cancel it, and write down why, so the next proposal does not repeat the mistake. Or give it a named owner, a six-week window and a defined success measure, and treat it as a fresh pilot rather than an existing asset. Or keep it deliberately because a small number of people genuinely rely on it, and stop counting it as an AI initiative.

All three are defensible. Renewing silently for a fourth year is not.

About this page

The four-cost breakdown and the four causes are Radical's own framing, drawn from what we see in assessments. The shelfware principle quoted here is principle ten of the ten in our internal charter. No vendor is named because the argument is about the buying pattern rather than any particular product. Written by Radical's own team; no client data was used.