Transformation & Change

Change Management for AI Adoption: From Prototype to Daily Use

Marin Niehues15 September 202511 min read

Marin Niehues standing with arms crossed and a lanyard in front of a blue wall with a logo.

Change management for AI adoption is the deliberate redesign of workflows and roles so that an AI application is used reliably in day-to-day operations. In many programs the technology is ready well before the business units accept it. That gap decides whether a prototype becomes a tool that shortens lead times and improves decisions. This article sets out the order in which executive and business-unit leadership should answer the key questions and how they can measure progress. This article provides an overview and does not constitute legal advice.

Why AI adoption follows its own logic of change

An AI application changes the workflow itself. Whoever used to write proposals or sort inquiries takes on a different task: they formulate instructions and check results. Responsibility and quality standards shift accordingly. Conventional software projects know this shift too. With AI it is more pronounced, because results vary and trust has to be built first.

Research shows why this care pays off. A field study with more than 5,000 customer service agents found that AI support raised the number of resolved inquiries per hour by 14 percent, and by 34 percent for new and less experienced agents. A field experiment with 758 consultants showed the other side: on tasks outside the model's capabilities, participants using AI were 19 percentage points less likely to produce a correct solution than those without it. The benefit therefore rests on two conditions that change management creates: employees actually use the application, and they know where not to trust it.

This yields a test for every initiative: a pilot counts as a success only when the business units use the application in their daily work. Why many pilots miss this threshold is described in why AI pilots do not reach operations. Change management plans for this threshold from the start and treats it as a deliverable in its own right. Where change management sits within the overall course of an introduction is shown in the article From pilot to scale.

The framework: five questions based on the ADKAR model

The ADKAR model describes five stages an individual passes through in a change: Awareness, Desire, Knowledge, Ability and Reinforcement. For AI adoption, this yields five questions that executive and business-unit leadership should answer for each application:

  • Awareness: Why is the workflow changing, and what does that mean for the individual role?
  • Desire: What advantage does the business unit gain in its own daily work?
  • Knowledge: Which steps and rules does each role need?
  • Ability: Where does the team practice on real cases until the application works reliably?
  • Reinforcement: Which responsibilities and metrics secure use over the long term?

The stages build on each other: a stage succeeds only once the previous one has been reached. A common pattern is to invest early in training and skip the rationale. The result is attendance without use. Starting with awareness creates the foundation on which knowledge can be absorbed at all.

For a single application in one business unit, ADKAR is sufficient as a framework. Whoever introduces a portfolio of applications across several units adds Kotter's program view with a sponsor and a guiding coalition.

Roles and decision paths

Four roles decide how an introduction unfolds. Each needs a name, a mandate and a fixed place in the rhythm of the initiative.

  • Executive sponsor: sets the goal and removes obstacles between units.
  • Product owner or delivery manager: steers the scope and sequence of the rollout and reports progress.
  • Business-unit owners: define how the workflow changes in concrete terms and take responsibility for results after go-live.
  • Multipliers in the teams: work with the application early and feed back to the rollout team.

Decision paths define who approves exceptions and who decides on changes to the application. In Germany, where a works council exists, it must be informed at the planning stage. Applications that are objectively capable of recording the behavior or performance of employees are subject to co-determination, and that applies to almost every AI tool with user logs. Early involvement prevents delays shortly before launch.

Enablement within the workflow

Enablement works where the work happens. An effective format consists of short sessions at the workplace in which teams process their own documents and cases with the application. It is complemented by a contact person for each team and by binding rules: which inputs are permitted, who checks results, and when is an issue escalated?

The EU AI Act also requires companies that deploy AI systems to ensure a sufficient level of AI literacy among their staff. Enablement is therefore also an obligation.

Feedback matters just as much. Users' observations flow back to the rollout team at a fixed rhythm and lead to adjustments to the application or the rules. Employees thereby experience that their input has an effect. This creates acceptance in the workforce.

Why do employees not use AI tools even after training?

Employees do not use AI tools despite training when they see no advantage in their own workflow, when the tool is too cumbersome in daily use or when it remains unclear who is accountable for results. Technology acceptance research has named two central drivers of use for decades: perceived usefulness and perceived ease of use. Later models add the influence of the environment, meaning the behavior of managers and colleagues, as well as the supporting conditions available.

Training addresses only one part of this, namely knowledge. Five causes appear particularly often in practice:

Each of these causes has its own remedy, and none of them is another training session. Assign the cause to the ADKAR stage at which use stalls, and act there.

How can you tell that an AI application has been adopted?

An AI application has been adopted when it is used regularly, changes the workflow measurably and delivers the benefit set out in the business case. The three levels build on each other:

  • Usage: active users per role and return rate after the first weeks.
  • Workflow: lead time and rework rate per case.
  • Outcome: the benefit defined in the business case, such as processing time saved or better decision quality.

The levels are read in this order. If usage falls, the cause usually lies in awareness or desire. If usage stays high while the workflow is unchanged, the application sits beside the process and is not built into it.

The evaluation belongs in the units' regular management reporting, with a fixed date and a named owner. A business case that derives benefits and costs robustly provides the target values for the third level.

How long does it take to introduce an AI application in a business unit?

For one application in one business unit, plan for eight to twelve weeks from the first user group to stable use. Each further wave is faster because rules and multipliers are already in place. A staged approach has proven effective: a first user group, an evaluation on a fixed date, then expansion to further teams.

Niehues Consulting as a partner for adoption

Niehues Consulting supports AI adoption from a delivery perspective. The approach follows the Niehues Delivery Model with the phases Understand, Build and Introduce. For the adoption of an application, three of its screening questions matter most: Who owns operations from day one? Are users enabled and processes adapted? Which metric shows after 90 days that the solution is landing?

First, goal and benefit as well as the affected roles are clarified. Then prototypes are built that prove the benefit in the real workflow, and maturity is developed until the business units accept the application. Finally comes the transfer into operations with internal or external teams and the introduction with adapted workflows and responsibilities. Project experience ranges from VC-funded scale-ups to DAX 40 groups, with assignments in 20+ countries on six continents.

The comparison shows how this approach differs from two common patterns.

The first conversation is a free initial consultation of 30 minutes. It clarifies which stage your initiative has reached and which role needs to be filled next. Book an appointment

Frequently asked questions

What is the difference between change management and project management for AI?

Project management steers the scope and schedule of the application, while change management steers the change in work that comes with it. In AI adoption, both run in a single plan led by a delivery manager or product owner.

What is the ADKAR model?

ADKAR is a model for change at the level of the individual, with the stages Awareness, Desire, Knowledge, Ability and Reinforcement. The stages build on each other, and action starts at the first stage not yet reached. For AI adoption, it translates into five screening questions per role.

How long does it take to introduce an AI application in a business unit?

For one application in one business unit, plan for eight to twelve weeks from the first user group to stable use. Further waves are faster because rules and multipliers are already in place.

Which metrics show whether an AI application is being adopted?

Three levels are enough: usage per role, change in the workflow and the benefit from the business case. Usage is the leading indicator, while the benefit proves the effect.

Why do employees not use AI tools even after training?

Because training conveys knowledge but establishes neither the benefit in one's own workflow nor integration into the process. Common causes are missing rules for input and review, managers who do not lead by example and an application that sits beside the actual system.

Is AI training mandatory for employees?

Yes, in the form of a competence obligation. The EU AI Act requires companies that deploy AI systems to ensure their staff have a sufficient level of AI literacy. The company decides on the format, and traceable documentation is advisable.

How should companies deal with employees' fear of losing their jobs to AI?

With an honest statement on which tasks will change and what freed-up time will be used for. Employees who help shape the change early as multipliers experience it as something they can influence. Vague reassurance without a plan increases concern.

Does the works council have to be involved in an AI introduction?

Yes, in Germany, as soon as the application is objectively capable of recording the behavior or performance of employees, and that applies to almost every AI tool with user logs. The works council must be informed at the planning stage and may call in an expert for AI matters.

Let’s talk about your initiative.

In 30 minutes we clarify where your organization stands and which step will pay off most now.

Request a consultation

30 minutes · free · no obligation