When you run a business, another list of AI tools can leave you with more questions than answers. You still need to decide where to begin, what information to use and whether the finished work will be good enough.
Start with a task you already understand. Familiar work gives you something concrete to judge, and a reason to keep going if the experiment proves useful.
That uncertainty about where to begin is common. The UK government's 2026 AI adoption survey identifies limited skills, expertise and understanding of use cases as major barriers, alongside uncertainty about the business case. It also reports that many improvements are incremental rather than transformational. Read the government's survey summary.
Look for a task with clear edges
Write down three pieces of work that regularly take more attention than you would like. Examples might include preparing a weekly update, turning approved product information into a first draft, or answering a familiar type of enquiry.
Then ask five questions about each:
Does it happen often enough for an improvement to matter?
Can you describe a good result in plain language?
Is accurate source information available?
Can someone knowledgeable check the output?
Could you test it without exposing confidential information or making commitments to customers?
A task with a clear input and a reviewable output is easier to test than a broad goal such as “automate our marketing”. Narrow that goal to something like “prepare three draft emails from an approved campaign brief”.
Choose the part AI will help with
Break the task into its actual steps. A routine customer reply might involve understanding the question, finding the current policy, drafting an answer and sending it.
For a first experiment, you could keep the person responsible for understanding the enquiry and choosing the correct policy. AI can help prepare a draft. The person then checks and sends it.
This gives the experiment a manageable scope. It also makes clear where a mistake can be caught before it reaches a customer.
Record how the work happens now
Before changing anything, complete a few representative examples using your current process. Note the time taken, the checks involved and the common difficulties.
Choose measures that fit the task. For a weekly report, you might track preparation time, missing information and the number of corrections. For a draft service page, you might check whether all required information is present and every claim is supported.
Include the effort needed to gather information and review the result. A quick draft that takes longer to correct may not be an improvement.
Try a small, controlled version
Use public, fictional or appropriately approved information to begin. Give the tool a clear task, the relevant source material, the intended audience and the format you need.
Ask it to flag missing information rather than fill the gaps. Then inspect the output yourself. Test a range of examples, including an awkward or incomplete one. A successful demonstration does not tell you how the process handles exceptions.
A fictional example
Imagine an independent visitor attraction that regularly answers questions about parking, opening hours and group visits. This is an illustrative example, not an Amazency client case study.
The owner could prepare an approved information sheet and use AI to draft answers to a set of fictional enquiries. A staff member would check every detail, especially seasonal opening hours and whether a booking is actually available.
The trial would compare total preparation and review time with the existing approach. It would also record mistakes, missing information and any replies that need to be written from scratch. No enquiry would be sent automatically.
The aim is to discover whether this particular task becomes easier while keeping the answers accurate. There is no assumed time saving.
Decide what happens next
After the trial, review the evidence. Continue if the output meets your quality standard and the overall effort is worthwhile. Refine the instructions if the same fixable problem keeps appearing. Stop or choose another task if checking the work is too difficult or the consequences of an error are too high.
Capture what you learned in a short project brief:
The task and why it matters.
The information the tool may use.
The output you need.
The checks a person must complete.
The measures you will use to judge the trial.
The conditions that would make you pause it.
That brief gives you a useful starting point for your next conversation about AI. You have a business question to answer and a practical way to explore it.
Ready to find a starting point? Tell Amazency which task you would like to improve and how you handle it today.