We implement AI in your company, starting with one process.

Who this implementation is for

For a small or medium-sized company where questions recur, information gets copied between places or staff repeatedly prepare similar material. A first step makes sense when the task has an owner, examples and a way to judge the outcome. If the team cannot yet describe the work or its rules conflict, we first clarify the process.

Possible starting tasks include drafting responses to enquiries, finding answers in documents or preparing content from approved materials. These are illustrative uses without client results.

First decision: task, data, ownership

We pick a recurring task that takes time and has examples to evaluate. We map the current steps and unnecessary handoffs. We compare a workflow change, ordinary automation and model-assisted work; fixed rules often benefit from a predictable flow.

When deeper analysis is needed, we propose a separate paid discovery sprint. The process map, priorities and acceptance criteria remain with your company even if another team delivers the implementation. Details: process diagnosis

How the work runs and how long it takes

The introductory call takes 30 minutes and establishes fit. We then agree the diagnosis scope, simplify the workflow, design a pilot using real and appropriately prepared examples, launch the solution and check it with its users. Each stage ends with an asset or a decision to approve. Delivery time and the owner's time commitment depend on scope, staff availability and data; the proposal specifies both before work begins.

A team member provides examples and checks whether the result helps. The owner approves the objective, data boundaries and any expansion. If the pilot does not demonstrate value, we record what we learned and keep the solution as it is.

What your company receives

Deliverables depend on the selected task. Agreed assets have an owner in your company, and acceptance uses examples and measurable criteria. We identify who updates sources and who responds to errors after launch.

• Map of the task, data and decision owners
• Pilot criteria and examples for evaluating results
• Working flow or assistant with defined access
• Team instructions, exception handling and handover
• Comparison of time, quality, rework and cost against a baseline

Example first-stage deliverables

This is an illustrative discovery scope for one task. We agree the implementation scope separately for each project. Inputs are a description of current work, approved documents, examples of correct and incorrect output, and a conversation with the person doing the task. Data preparation follows agreed access rules.

We hand over the current and proposed workflow map, approved sources and permissions, a baseline and a pilot brief: evaluation cases, acceptance criteria, decision owner and stopping conditions. Acceptance means reviewing these assets together and recording a decision: pilot, simpler automation or no further implementation.

This stage does not automatically include production integration, migration of the entire knowledge base, company-wide training or ongoing support. The proposal separates delivery work from subscriptions, API or model usage, hosting and support. It names account owners, a cost limit and the person maintaining sources; further work needs separate approval.

Data, people and the model's limits

For each tool, we define permitted data and access. Confidential documents and personal data do not go into a staff member's unapproved personal account. A model can help summarise a conversation or locate facts in a document, while people check its output. Employment, legal and financial decisions remain with accountable people.

In the EU, Article 4 of the AI Act addresses measures to support the ability of staff to use AI systems. The scope depends on use and risk. The Commission explains it here: AI literacy questions and answers

You can start writing down approved tools and data rules with our template: AI use policy for a small business

How we measure the outcome and agree a price

Before the pilot we record a baseline: task time, rework, quality and operating cost. At acceptance we compare similar cases, including failures and cases that need a person. Agreed criteria show whether to expand, simplify the workflow or stop the pilot.

Price depends on the task, data, integrations, access and ongoing support. Diagnosis is a separate paid stage; implementation, tool usage and further support have separately stated scopes and quotes. One price cannot describe every company.

Where to go after the first task

Want to plan your first step yourself? Start with the guide: how to start using AI in a small business

After acceptance, your team can operate the solution. If you need continued support, we agree on ownership of sources, new versions, errors and access before another stage begins.