Why AI tools end up in the drawer

In many companies we meet, the pattern repeats. Management buys licences for an AI assistant, a presentation session takes place, a few curious employees test the tool for three weeks, and then usage fades. Six months later nobody knows who is still using it, and the spending continues quietly.

The usual diagnosis is resistance to change. We think it is mostly wrong. People do not refuse a tool that saves them an hour a day. They ignore a tool that fits into none of their real tasks. The tool sleeps in the drawer because nobody changed the work around it.

The drawer is an organisational problem, not a technology problem

An AI tool is a means. Until someone says what it is for, who uses it, in which process and with what expected result, it remains one option among a hundred in the day of an already stretched employee. In an SME, the day is full. Nobody has time to search for where a novelty might apply.

Our position is simple: adoption is decided before the purchase, not after. It belongs to senior and middle management, not to the IT department or a training provider.

Three causes we see most often

1. A tool with no defined task

"Use AI to be more efficient" is not an instruction. It tells nothing to the accountant, the sales manager or the workshop lead. By contrast, "draft the first version of replies to recurring quotation requests" or "summarise past exchanges with a client before each meeting" is a task that can be tried, measured, improved or dropped.

A use that cannot be stated in one concrete sentence will not survive the first busy week.

2. An individual gain nobody captures

Suppose an employee saves twenty minutes a day. What does the company do with it? If nothing changes in objectives, scheduling or volume handled, the gain dissolves. The employee has no reason to change habits, and management has nothing with which to defend the investment. A gain that is neither visible nor reallocated ends up interesting nobody.

This is a steering question as much as a usage question, and it is close to the concerns of the finance function: knowing what you are trying to obtain and how you will see it. We return to this in our work on finance and performance.

3. Unclear rules on what is allowed

Many employees hold back out of fear of getting it wrong. Can a client contract be pasted in? A list of employees? A quotation with confidential terms? Since the revised Swiss Federal Act on Data Protection (FADP) came into force on 1 September 2023, the question is no longer theoretical for Swiss companies, and the European AI Act, whose text is now finalised, will add its own requirements in phases for those operating in the Union.

Without a simple rule, two behaviours emerge. The cautious abstain. Others use personal tools outside any control. In both cases the official tool stays in the drawer. One page of clear rules, with three categories of data (free to use, to be anonymised, prohibited), does more for adoption than many awareness sessions.

What a forgotten tool really costs

Take a deliberately simple example, with explicit assumptions. A company of 120 people buys 40 licences at roughly CHF 30 per month. That is CHF 14,400 a year. Add two hours of training per person, at a fully loaded cost of about CHF 70 per hour: CHF 5,600. The total approaches CHF 20,000 in the first year.

That amount is bearable. The cost that matters lies elsewhere: management time, lost credibility with teams ("another tool nobody mentions any more") and above all opportunity cost. A company that has lived through one failure approaches the next with suspicion. If only ten of the forty users open the tool each week, the true price per active user is four times the listed price, and that calculation is rarely made.

What we recommend to leaders

Start with one use, one team, one owner

Choose a team and a repetitive task whose result can be observed. Name an owner on the business side, not in IT, who has the authority to adapt how work is done. Our experience on getting an AI assistant adopted in customer service shows that success depends less on the choice of tool than on the place it is given in the advisers' daily routine.

Attach the tool to a real moment in the work

The tools that survive are those that appear at the right moment: when a file is opened, before a meeting, at the start of a new job. The case of accelerating seasonal staff onboarding illustrates it well: when the assistant answers the questions newcomers actually ask, at the moment they ask them, it becomes a reflex rather than an option.

Define one indicator before launch

One indicator is enough: handling time, response delay, rework rate, number of requests absorbed without escalation. Set it before deployment, measure it after four to six weeks, and decide: continue, adjust or stop. Stopping is a legitimate decision. A tool explicitly abandoned costs less than a tool tolerated indefinitely.

Write the rules before the use cases

One page is enough. Which data are allowed, which are not, who approves a new use, whom to ask. This document protects the company under the FADP and reassures employees. It is reviewed every six months as regulation and tools evolve.

Train on the task, not the tool

A generic AI training session produces enthusiasm and little change. A forty-five-minute session centred on the team's own task, using its own (anonymised) documents and examples of good and poor outputs, produces habits. Afterwards, an internal contact available every week for a month is worth more than a single day of training.

A deliberate stance: fewer tools, more organisational work

The reflex is to multiply tools so that everyone finds their own. We recommend the opposite. One well-integrated use in one team, measured and explained, creates an internal example that other teams will come looking for on their own. Ten tools available to everyone mostly create ten opportunities not to use them.

AI is not always the right answer. In some cases a better-written procedure, a document template or a clearer decision rule solves the problem with no technology at all. Saying so is part of the job, and it is what we defend in our approach to organisation and adoption: start from the real work, then choose the tool.

Where to begin

Take an honest inventory. Which AI tools have you paid for? Who uses them every week? For which task? With what observed result? If the answers are vague, you already have your starting point. If you would like an outside, structured and non-binding view of the real state of your usage and of the two or three initiatives worth the effort, our diagnostic is designed for that.