Many businesses start with AI by asking what it can do, and end up with a program nobody uses. A better question is: which work keeps getting left behind here? That is where the first gain is. This plan takes you from that question to a working first automation, without turning your whole business upside down.
- Choose one process: recurring work with a clear trigger and a clear outcome.
- Measure first: record how much time it takes and how often it goes wrong.
- Run a small trial: one channel, one team, a fixed period.
- Decide on human checks: what may happen automatically and what does a member of staff see first?
- Evaluate and decide: compare with the baseline and choose to expand, adjust or stop.
Step 1: choose the right process
Look back at the past month. Which enquiries were answered too late? Which quotes sat in draft for a week? Which calls were never returned? Work that comes back often, follows fixed rules and that nobody enjoys is the best candidate.
Good first candidates
- Recurring questions: a hair salon that gets the same questions about prices and free slots every day.
- Missed calls: an installer who is on a job when new customers ring.
- Reminders: appointments forgotten without confirmation, or quotes nobody follows up.
Better not yet
Processes that need tailoring every time, where a mistake costs money straight away or where the information is not recorded anywhere. Only take those on once you have experience with a simpler process. Which processes qualify is worked out per department in AI by business process.
Step 2: measure before you start
Without a baseline you only know afterwards how it feels, not what it delivers. Keep track for two to four weeks of:
- Volume: how many questions, enquiries or tasks come in each week.
- Time: how many minutes your team spends on each one on average.
- Response time: how long a customer waits for an answer or a confirmation.
- What goes wrong: missed calls, forgotten follow-ups, mistakes in appointments.
A simple tally sheet or an export from your diary is enough. What matters is that after the trial you measure the same things in the same way. Also note who was working in those weeks and whether it was a busy or quiet period, so you compare like with like.
Step 3: run a small trial
A trial is small enough to oversee and large enough to learn from. Choose one channel (only WhatsApp, for example), one location or team and a fixed period of a few weeks. Agree beforehand who follows the trial and when you will evaluate.
Make sure the information the automation draws on is correct: current prices, opening hours, services and the answers to common questions. An AI assistant working with outdated data gives outdated answers. If you use our industry software, the AI assistant takes that data from the same diary and price list as your team, so there is no second source to maintain.
Step 4: decide what a person checks
Before you start, decide which steps may happen automatically and which a member of staff sees first. In a trial you are strict: better too many checks than a mistake that reaches a customer.
- May happen automatically: answering a question about opening hours, sending a reminder.
- Goes to a member of staff first: a quote, a discount, a complaint, anything the assistant is unsure about.
How to set up those limits, approvals and fallback routes is worked out in human oversight in AI processes. How to check the answers themselves is covered in checking AI output.
Step 5: evaluate and decide
After the trial, compare the same points as in the baseline. Look not only at time saved but also at corrections: how often did a member of staff have to fix something? And ask your team and a few customers how it went.
- Expand: the figures are better and there are few corrections. Switch on the next channel or process.
- Adjust: it works, but the same things keep going wrong. Change the information or the checks and measure again.
- Stop: it delivers nothing or takes more checking than it saves. That is an outcome too.
If you would first like a broader view of where AI pays off in a small business and what drives the cost, read where AI actually pays off in a small business.