AI strategy
AI implementation in business, step by step
A practical guide to AI implementation, from scattered ChatGPT experiments to AI use that is organised, measured and useful to your team. Six steps, a scorecard for choosing where to start and a full worked example.
Why a method helps when you implement AI
An AI implementation that works starts from real business problems, tests one specific task with a few people and measures the result before rolling it out. This matters most in small companies where staff already use ChatGPT in their own way, each with their own habits, and nobody has decided which client details can go into it.
Gallup reported in May 2026 that 30% of US employees use AI at work several times a week or more, and 15% use it daily. Usage spreads on its own. What is often missing is a structure for deciding which tasks are worth changing, how to measure the result and which rules apply.
A use case is one specific task where AI helps, such as drafting the first version of a quote or sorting incoming emails. The six steps below help you find, test and extend them, and they work the same way in a 15-person accounting firm, an online shop or an estate agency.
6 steps for AI implementation in a business
Each step prepares the next. Skipping one usually costs time later.
- 01
Start from business problems
Before choosing a tool, write down where your team loses time or money. Quotes that take days, customer emails left waiting, monthly reports built by hand. Ask each manager for three tasks they would gladly hand over. That list is your starting point.
- 02
Map the processes
For each problem, describe the real process step by step, with who is involved, which tools they use, what data goes in and out and how long it takes. Do it with the person who does the task, because they know the exceptions. Without this map it is hard to see where AI fits.
- 03
Prioritise use cases
Run each candidate through the four-question scorecard further down. Start with the ones that collect the most yeses. Drafting replies to common enquiries is usually a better first project than an agent that sets prices.
- 04
Run a pilot and measure it
Test one or two use cases for a few weeks with a small group. Measure time per task, quality checked by a person and the number of corrections, before and after. A Harvard Business School and BCG study of 758 consultants found over 40% higher quality and 25% faster completion on suitable tasks, and worse results outside them. Your pilot shows which of the two groups your task falls into.
- 05
Train the team and set usage rules
A tool nobody knows how to use well ends up forgotten. Train the people involved with examples from their own work. Show them how to ask, how to check the answer and when to be sceptical. At the same time, publish clear rules on data, approved tools and human review.
- 06
Scale and maintain
If the pilot works, extend it to more people and document the process. Give each workflow an owner, review results regularly and adjust instructions when data or tools change. Then go back to your list from step 3 and pick the next use case.
How to choose your first AI use case: a four-question scorecard
Impact-versus-feasibility grids are useful, but they are often filled in by gut feel. These four questions are answered yes, partly or no, and force you to look at the actual task. The more yeses, the better the candidate.
Does it happen at least weekly?
A daily or weekly task pays back the set-up effort quickly and gives you data to measure within a short time. An annual report can wait.
Does it follow a rule someone can explain?
If the person who does it can describe how they decide, AI can help. If the answer is "it depends, I judge it as I go", that rule needs writing down first.
Is a mistake easy to fix before the customer sees it?
An internal draft that someone reviews can tolerate errors. A price sent to a customer cannot. To begin with, pick tasks where a person approves the output before it leaves the company.
Is the data it needs close at hand?
If the information sits in an email, a spreadsheet or the CRM, you can work with it. If it is spread across paper files or one person's memory, the project starts by organising it.
Example
A Brighton estate agency picks its first use case
Take, as an illustration, an estate agency in Brighton with six negotiators and one administrator. The director asks each of them for three tasks they would gladly hand over. Three come up again and again: writing property listings for the portals, replying to enquiries from those portals and the monthly report to landlords.
On the four-question scorecard, listings collect four yeses. They are written every week, follow a known structure, are checked by the negotiator before going live, and the data is in the property record. The landlord report scores worse, because it is monthly and pulls data from several sources. Portal enquiries sit in between.
The pilot runs for a few weeks with two negotiators. They use an AI assistant with shared instructions that set the agency's tone, forbid inventing features and leave out the owner's name and phone number. Before starting, each negotiator notes how long a listing takes to write. At the end they measure again and count how many corrections each text needed.
With the results in hand, the agency runs hands-on training for the whole team using their own listings, publishes a one-page usage policy and makes one negotiator responsible for maintaining the instructions. The next item on the list, portal enquiries, is prepared with its process already mapped.
Common mistakes when implementing AI in a business
They show up in companies of every size. Most are avoided by following the steps above.
Starting with the tool
Buying licences before knowing what they are for. The team finds no use for them and the spend gathers dust.
An over-ambitious first project
An agent that does everything takes months and wears everyone down. A narrow, measurable use case delivers results in weeks and builds confidence.
Measuring nothing
Without a baseline you cannot tell whether AI saves time. Time the task before the pilot, even roughly.
Banning everything
If the company offers no approved tools, people use their own in secret, with personal accounts. Clear rules and an approved tool bring that use into the open.
Leaving training until last
Many people try once, get a mediocre answer and give up. A hands-on session with their own work changes that first impression.
Forgetting human review
AI can make up facts or be confidently wrong. Any text that reaches a customer needs a person to check it.
Data and an AI usage policy for your business
An AI usage policy does not need twenty pages. A short document your team can read in ten minutes is enough, provided it answers three concrete questions.
Which tools are approved, and with which type of account. Free and business versions do not always handle the data you enter in the same way, so check each provider's terms before approving a tool.
Which data must never go into an external tool. The usual examples are customers' personal data, financial information, contracts and health data. An estate agency can ask AI to write a listing without including the owner's name or phone number.
Who reviews what. Define which tasks need human review before they go out, who is responsible for each automated workflow and who to ask when in doubt. Review the document every few months, because the tools change quickly.
Setting up your first pilot, and which steps Obinova covers
Ask each manager for their three tasks and score the list against the four questions. Pick the top scorer, time it and find two people willing to try it. That is everything a first pilot needs.
Obinova is an AI agency that helps companies put AI to work for their people. Our AI audit and consulting covers steps 1 to 3. We map your processes with the people who run them and prioritise use cases with you, starting with a launch session and continuing with follow-ups several times a year.
Our n8n automation service handles the pilot and scale-up, with workflows and AI agents that connect your email, CRM, spreadsheets and internal tools. Our training prepares your staff on ChatGPT, Claude, Gemini and Microsoft Copilot with everyday use cases, on a dedicated video platform or in person at your offices.
Frequently asked questions
Where should a small business start with AI adoption?
Start with a list of repetitive tasks that currently eat into your team's time. Pick one that happens often, is easy to measure and carries little risk if AI gets it wrong, such as drafting replies to customers. Try it with two or three people for a few weeks and compare the time before and after. That result tells you what to do next.
How much does AI implementation cost for a business?
It depends on how many processes you want to review, the tools you already use, the type of automation and how many people need training. A focused pilot takes far less effort than a rollout across several departments. At Obinova we give you a quote after a first meeting, once we understand your situation.
How long does an AI pilot last?
A sensible pilot runs for a few weeks. The team needs time to get used to the tool, and you need time to gather reliable data on speed and quality. The exact length depends on how often the task happens, since a daily process gives results sooner than a monthly one. The key is to decide at the start what you will measure and when you will decide.
Do we need a technical team to implement AI?
Using assistants such as ChatGPT, Claude, Gemini or Copilot does not require technical staff, although practical training helps a great deal. Automations that link several tools do need someone to build and document them. With good documentation, your team can maintain them without relying entirely on an outside supplier.
Is implementing AI worth it for a small company?
Starting is usually worth it, provided you choose the first task well. In a small company, every hour the team stops spending on copying data or writing repetitive text is noticeable. A measured pilot gives you the answer in your own figures before you invest more, and lets you stop if the result does not justify the effort.
What data can we share with an AI tool?
It depends on the tool, the type of account and the provider's terms. As a cautious rule, do not enter customers' personal data, financial information or confidential documents into tools your company has not approved. Write these rules into a short usage policy and make sure the whole team knows it.