Whitepaper

From AI pilot to daily operations

Artificial intelligence · Summary

The problem

Many companies test AI in pilots. Many of these pilots do not become part of daily operations. This whitepaper tells scaleups and enterprises how to close this gap.

In August 2025, Gartner forecast that up to 40% of enterprise applications will include task-specific AI agents by 2026. The figure in 2025 was less than 5%. In a separate forecast, Gartner said that companies will cancel more than 40% of agentic AI projects by the end of 2027.

Thus, adoption can be fast and failure can be frequent. Consulting decisions and delivery work must therefore stay together.

Select the correct work

Start with the result, not with the model. Select work where you can measure the effect on cost, time, or risk.

In our predictive maintenance case study, an industrial operator put equipment failures in order of cost. The project then focused on the 20% of assets that caused 80% of the downtime cost. Unplanned downtime on this equipment decreased by approximately 30% in the first year.

Make the data ready

A model cannot correct bad source data. Examine the data before you build the model.

In our document classification case study, a six-week assessment found a problem. Approximately 40% of new documents needed pre-processing or normalisation. We built this data layer first. The underwriting team then got 4× throughput on document review for each analyst.

Regulated companies must also answer these questions about their data:

Where did the data come from?
How did the system process the data?
Who had access to the data?

Use the existing tools

People usually do not accept an AI feature that makes them change tools. Put the AI in the workflow that they already use.

In our customer service case study, the AI was inside the existing ticket system. The project team released it in stages and examined the quality at each stage. Agents now handle approximately 60% more tickets in each shift.

Set limits for agents

An assistant suggests an action, and a person makes the decision. An agent can do the action. This change increases the responsibility of the company.

Answer these questions before an agent starts work in production:

Which actions can the agent do without approval?
Who approves the other actions?
Who stops the agent if a problem occurs?
Where does the system record each action?
Who is accountable for the result?