AI implementation

A roadmap is a promise. Execution decides whether it is kept. The value of AI arises in the process in which a tool is used: a summarizing tool delivers little if nobody knows what to do with the summary, and a chatbot does not help if the knowledge base is out of date. That is why, during implementation, we look at the work itself, and stay involved until the new way of working is in place.

We implement process by process, in the order of the roadmap. Every process goes through the same steps: analysis, redesign, set-up, rollout, measurement and handover. That way your organization builds an approach it can repeat on its own after the first process.

Our approach

Every implementation starts with the question of what has to change in the work, and only then with the question of which tool fits. We work with a fixed core team from your organization: the owner of the process, two or three people who do the work every day, someone from IT and someone from information security or privacy. That team takes the decisions. We structure and accelerate.

Process analysis. Together with the people who do the work, we map the process: steps, handovers, systems, waiting times and the friction points. We measure the starting point: lead time, error rate, time per case. The best improvements rarely come out of a meeting room. They come from the front line.

Redesign. We redesign the process with AI as one of its components. We explicitly decide which steps are automated, where human judgment stays and how quality is safeguarded. For applications in a high risk class we set up human oversight, logging and documentation as the AI Act requires.

Tooling. We select and set up the tools that fit the process. Vendor-independent, with attention to data protection, integration with existing systems and manageability. Wherever possible we use what the organization already has. A new tool only comes into view when the existing one cannot do the job.

A trial in real work. Before a process changes organization-wide, the new design runs for a few weeks with a small team and real cases. Whatever goes wrong there, we fix in the design and not in a work instruction afterwards.

Rollout and change

New technology asks for new behavior. The rollout therefore gets the most attention of the whole engagement. A process that is right on paper and unused on the front line has delivered nothing.

Program management. We steer on the roadmap: planning, dependencies, budget and the decisions the management team has to take. In a rhythm of short cycles, so that every few weeks you see what has been delivered and where it chafes.

Change on the front line. We work with team leads and staff on work instructions, on-the-job training and room to learn. We take resistance seriously, because it usually points to something that is not right yet in the design.

Measuring. Up front we agree what success is: lead time, quality, satisfaction, cost. During the rollout we measure against the starting point and adjust. That way the management team sees afterwards what the process has actually delivered.

Handover. The goal is that your organization can do this itself. We train an internal process lead, document the approach and hand over the program as soon as that is responsible.

What you get

Every process delivers a fixed package. The implementation plan is its heart: one document that brings together the redesign, the choices on automation and oversight, the tooling and the measurement agreements.

In addition you get the work instructions for the new process, a configured and documented set of tools, the baseline and final measurements, and a playbook with which the organization can tackle the next process itself.

From pilot to standard practice

Many organizations have run a handful of pilots by now. Few of them have become part of daily work. That is rarely down to the technology. It is down to what is missing after the pilot: an owner, an adjusted process, a place in the budget and someone who maintains it.

That is why, during implementation, we treat a pilot as the first step of a rollout, with the same requirements for ownership, measurement and maintenance. Why isolated experiments rarely reach scale, and what a management team can do about it, is explained in the article Why isolated AI experiments rarely reach scale. How implementation fits into the integration phase is described in From exploration to integration: the three phases of an AI roadmap.

The result

A process that demonstrably works faster, better or cheaper, staff who understand and own the new process, and an approach that can be repeated for the next process. And an organization that can do it on its own afterwards.

Duration. A first process usually takes three to six months from analysis to handover, depending on the size of the process and the tooling required. Subsequent processes go faster, because the approach and the core team are then in place.

Involvement. The core team spends on average one day a week on the engagement. The management team takes a decision every four to six weeks, based on what has been delivered.

Who it is for. Organizations with a roadmap that requires execution, boards that want to stay in control of an AI program, department heads who want to improve a process and teams that need to adopt a new way of working.