AI for SMEs: from an idea to a production workflow

The shortest honest advice first: an AI initiative reaches production when you choose one process whose current costs you can measure, build the workflow with clear checkpoints, and take adoption as seriously as the technology. The rest of this guide explains that sentence.

Who this guide is for

For managing directors and department heads in businesses with around 20 to 500 employees who want to move beyond general discussions about AI and take a solid first step, without an in-house AI team or a strategy project lasting months.

What usually goes wrong

Four patterns come up repeatedly in our conversations:

1. Too many ideas, no order of priority. Every department has candidates, but nobody has a comparable assessment. The result: the project backed by the loudest voice gets started.
2. No baseline. Without measuring before the start, later claims of impact cannot be substantiated.
3. A demo without integration. A tool impresses in testing but fails because of data access, approvals, or simply because it runs alongside the systems people actually use.
4. Nobody makes adoption stick. The pilot works, but three months later the team is back to its old ways because introduction and ownership were missing.

The path in four stages

1. Prioritise the ideas you have

Assess candidates against four questions: how much recurring effort does the process require today? How accessible are the data and systems? How sensitive is the data? What happens if a result is wrong? A simple scoring framework is enough. What matters is assessing every candidate against the same criteria. Our article on prioritisation provides a detailed scoring framework.

2. Measure the starting point

Before anything is built, record minutes per case, cases per week, and the error or follow-up query rate. A week of tally marks is better than no baseline. These numbers are the only sound basis for demonstrating impact later.

3. Build with controls

Three things distinguish a workflow ready for production from a demo: integration with the systems of record, checkpoints where people decide before anything is sent externally or posted, and documented limits specifying the cases the workflow deliberately hands over to people. Our article on human-in-the-loop explains how to define those checkpoints in practice.

4. Make adoption stick

Introduction takes work: training on real cases, a named person responsible within the team, operating documentation, and a date to assess results against the baseline. Only then does a decision about scaling make sense.

A brief look at the legal framework

For most SME back-office workflows, the main considerations are the GDPR (data access and processing on behalf of a controller) and, depending on the intended use, transparency and risk obligations under the EU AI Act. The specific requirements depend on the use case and need legal review. A workflow with clear checkpoints and documented data flows provides the best starting point.

How to recognise a sound first step

• It concerns one process, rather than “the business” as a whole.
• There is a measured baseline.
• It is clear who decides on edge cases.
• The scope is limited, and the decision on what follows is explicitly left open.

The AI Opportunity Check is designed to help you assess which of your processes meets these criteria.