Where should a business start with AI?

1. List what repeats. Spend a week noting every task your team does more than a few times a week: replying to the same questions, copying data between systems, chasing people, building reports.
2. Put a cost on each. How often it runs, how long it takes, who does it. A rough hours figure is enough.
3. Pick one. The task that runs most often, follows the clearest rules and touches data in a system you can connect. Not the most impressive one.
4. Define success in one sentence. "Every enquiry gets a reply within five minutes." "The Monday report arrives without anyone building it." "Supplier invoices are logged without re-typing."
5. Pilot it, measure it, then decide. Run it on real work, compare to the baseline, and only then choose the next process.

Use three filters, in this order.

• Volume. A task done 50 times a week beats one done twice a month, even if the second looks more interesting.
• Rule clarity. If two staff would handle the same case differently, agree the rule first.
• Data access. The information has to live somewhere a system can read: a CRM, a shared inbox, a spreadsheet, an accounting tool.

Good first candidates in most small and mid-sized businesses:

• Enquiry intake and first reply.
• Chasing unpaid invoices or unanswered quotes.
• Logging documents that arrive by email.
• Compiling a weekly report from two or three tools.
• Booking and reminding appointments.

Poor first candidates: anything you do a few times a year, anything still changing month to month, and anything where one wrong output causes legal or financial damage.

A pilot, on one process, with the baseline recorded before you start. A good pilot runs on real work, not test data, and often runs alongside the manual process for a short period so the team can compare outputs before switching over.

A narrow pilot that works becomes the template for the next one: the connections to your CRM, inbox and accounting tool already exist, and your team has seen it work. A broad rollout that stalls becomes the reason the next idea gets no budget.

One person, not a committee. They do not need to be technical. They need authority to make decisions about the process, access to the people who do the work, and some protected time during the pilot.

Bring in outside help for the specialised parts: choosing the tools, connecting systems, designing the AI steps and the guardrails around them. Keep the internal owner focused on the process and on getting the team to use the result.

• Buying before deciding. A chatbot licence or an AI writing tool bought for a vague problem ("we need to be more efficient") ends up unused.
• Starting with the hardest process. The most painful process is often painful because of exceptions and politics. A first project needs to succeed cleanly.
• No baseline. Without the before number, the result becomes a matter of opinion.
• Staff pasting client data into public AI tools. Before anyone experiments, agree which tools are approved and what data can go into them.

Say you run a car rental company in Dubai with 80 cars, a booking site and a WhatsApp line. A week of listing what repeats shows:

• Answering the same WhatsApp questions (deposit, insurance, airport delivery, documents) many times a day.
• Checking returned cars for fines and Salik charges and adding them to the customer's bill.
• Chasing customers to upload licence and passport copies before pickup.

The document chase scores best: it happens on every booking, the rule is clear (no documents 24 hours before pickup, send a reminder), and bookings already live in a system. The success sentence: "Every customer has documents uploaded before pickup without staff chasing."

The pilot runs on new bookings for a month. Once it works, the frequently asked WhatsApp questions become the second project, with an AI step answering from an approved list and handing anything else to staff.

Businesses that start this way tend to keep going, one process at a time, and some go a long way. A founder's AI chief of staff that books meetings, captures every call, files the notes and chases documents before each meeting is a good example of what several connected automations add up to. An AI email sales agent running inbound sales for a digital publishing business across 63 properties is another. Neither started as a big programme. Both are made of specific processes, wired together.

We start with immersion: time with the people doing the work, watching how it actually flows rather than how the org chart says it does. Each repeated process gets an hours figure. Then we rank them by return against effort to build, and say clearly which ones to leave alone.

You leave with a short, ordered list and the first project scoped: what it does, what success looks like, what it touches and how it would be built (n8n, Zapier or Claude Code). You can build it yourselves, or with us.