Why people return to spreadsheets after an AI rollout

Essay

The spreadsheet reopens quietly.

The pilot has met its technical test. Training has been delivered. The new workflow is available. Yet a team leader keeps a parallel file because the system does not capture an exception their people face every week. Another manager asks for the old report before making a decision. A reviewer copies the model’s output into a private note and checks it by hand.

The simplest interpretation is that people have not adopted the tool. The more useful question is why the workaround still feels necessary.

A workaround is not only a behaviour to remove. It is evidence about what the official process has failed to make safe, clear or useful.

The senior-leader decision

Leaders often receive adoption as a dashboard: log-ins, completion rates, active users and time saved. Those measures can show whether the system is being touched. They cannot, by themselves, show whether people trust it with the work that matters.

Before applying more pressure, the senior team needs to distinguish three different gaps:

• Capability: people do not yet know how to use the system or lack the time, access or support to do so.
• Trust: people understand the system but do not believe its output, incentives or consequences are safe enough to rely on.
• Work design: the formal process does not reflect the exceptions, relationships or judgements through which the work actually gets done.

These gaps can coexist. They also require different leadership responses. Training may address capability. It will not settle who is accountable when the output is wrong, repair an incentive to keep a private record or make an unrealistic workflow fit reality.

Capacity is not willingness

Research on organisational readiness for AI adoption describes a socio-technical journey: technology, people, process and organisational conditions change together. That evidence does not provide a universal maturity score or predict whether a particular implementation will succeed. It does support a narrower point: technical readiness alone is an incomplete account of adoption.

A person may be fully capable of using a system and still choose not to rely on it. Their reasons may include weak data, unclear authority, fear that an error will be blamed on them, or a belief that the new process makes an important part of their work invisible.

Calling this resistance can close the conversation too early. Leaders should ask what the behaviour protects. The spreadsheet may preserve local knowledge. The manual check may compensate for missing evidence. The extra approval may be a response to responsibility that was never reassigned when the workflow changed.

Read the workaround as a signal

Not every workaround is wise. Some introduce risk, fragment data or preserve an outdated habit. But a leader cannot know which kind they are looking at without examining the work.

Choose one recurring workaround and trace it with the people who use it:

1. Name the moment. What has just happened when someone leaves the official workflow?
2. Name the need. What information, judgement or protection does the alternative provide?
3. Name the consequence. What could go wrong if the person followed the official process instead?
4. Name the owner. Who has the authority to change the system, the measure or the decision rule?

This turns “adoption” from a judgement about employees into a leadership question about the conditions they have been given.

What readiness looks like in the room

Readiness is not a single score. For a specific decision, leaders can look for several practical conditions.

People can explain what is changing

The team understands the purpose of the system, which part of the work it changes and which judgements remain theirs. If leaders offer different answers, employees are being asked to adopt ambiguity.

People can challenge the system

There is a usable route for questioning an output, recording an exception and escalating a consequence. The route has an owner and does not depend on personal influence.

Measures support the new behaviour

Targets and review practices do not punish the care the new workflow requires. If leaders ask for judgement but reward only speed, people will protect themselves through the process that best fits the measure.

Leaders use the same process they promote

Senior behaviour tells the organisation whether the change is real. When leaders bypass the system, request the old report or override a result without explanation, people notice the operating rule beneath the official message.

Move, pause or change the design

The decision is not always “roll out” or “stop”. A team may be ready for one workflow and not another. A limited release may be appropriate while an accountability question is resolved. The design may need to change before scale. In some cases the evidence may support continuing as planned.

What matters is that pace follows the live risk rather than a fixed prescription. A short pause is useful only if leaders know what must be learned or decided during it.

The senior team should be able to state:

• which behaviour is creating concern;
• whether the cause is capability, trust, work design or a combination;
• which leadership choice would change the condition; and
• what evidence would show that the response is working.

The spreadsheet is asking a question

Teams returning to old tools do not prove that the technology is wrong. They do show that the official workflow has not yet earned exclusive use.

The leadership task is to understand what the parallel process is carrying: context, discretion, safety, memory or control. Once that is visible, leaders can decide what belongs in the system, what should remain a human judgement and what the organisation must change around both.

If people know how to use the tool and still choose another path, the next intervention should begin with listening, not another training module.