Automation Planning

Business Automation: Where to Start and How to Plan the First Project

By Arezoo Mohammadzadegan October 28, 2026 8 min read
Business Automation: Where to Start and How to Plan the First Project

When people ask about business automation, where to start is usually the first question, and the honest answer is one process rather than a platform. This guide sets out the order that works for a small or mid-sized company: choose the process, record how it runs today, take a baseline, check the numbers, decide what kind of automation it needs and who will build and own it, then run a small pilot that ends in a clear decision. Each stage links to a deeper guide where one exists.

The stages follow the automation lifecycle in the NSW Government’s Automation Guide, a public guide written for government agencies, scaled down for a private business. Sources are linked where they are used and were read on 28 October 2026.

Choose one process to start with

Most teams have a list of jobs they would like to stop doing by hand. Pick one of them for the first project. A single process gives you something you can measure, finish and judge, and what you learn about your data, your tools and how staff react carries over to the next one.

Good first candidates tend to share a few traits.

  • It happens often, every day or every week, so small savings add up.
  • It follows rules you can write down, such as “if the invoice total matches the purchase order, file it and tell accounts”.
  • The input arrives in digital form, as an email, a form or a spreadsheet row, rather than on paper or in a phone call.
  • A mistake is easy to spot and easy to undo.
  • One person or team owns it and can say what a correct result looks like.

Google’s book on site reliability engineering has a useful name for this kind of work. In its chapter on eliminating toil it describes toil as work that “tends to be manual, repetitive, automatable, tactical, devoid of enduring value,” and that “scales linearly as a service grows.” The book was written about running software services, but the same traits describe a lot of office work, such as retyping order details from emails into an accounting system. It also gives the warning that matters most here: “If human judgment is essential for the task, there’s a good chance it’s not toil.” Work that needs judgment can still be helped by AI, but it makes a harder first project.

How to compare several candidates and rank them is the subject of the next guide in this series.

Write down how the work is done today

Before anything is automated, record the process as it actually runs, not as the procedure document says it should. Sit with the person who does it and note these points.

  • What starts it: an email arrives, a customer calls, or it is Monday morning.
  • Each step in order, and which system or file it touches.
  • The decisions along the way and the rule behind each one.
  • The exceptions, such as a missing reference number or an order sent as a voice note, and what happens to them now.
  • Where the result goes and who checks it.

Spend the most time on the exceptions. The common path is usually the easy part to automate. The cases that do not fit are where projects slip, because each one needs its own rule, a fallback or a person.

The NSW automation lifecycle puts this at the start of its discover stage, asking teams to “discover and identify current process state and verify existing hypothesis.” Writing the process down often shows steps that nobody needs. Remove those first, because automating an unnecessary step only makes it happen faster.

Process map with steps, one decision point and a side path where exceptions go to a person
Map the exceptions as carefully as the main path.

Measure it before anything changes

Take a baseline while the process still runs by hand. Without one, nobody can say later whether the automation helped. The same NSW page tells agencies to “determine baseline metrics to use them as benchmarks for post-automation performance analysis.” For a small company, four numbers are a reasonable minimum.

  1. Volume, meaning how many times the process runs per week or per month.
  2. Time, meaning minutes of staff time per run, measured over a normal week rather than guessed.
  3. Errors, meaning how often a run has to be corrected and how long the correction takes.
  4. Delay, meaning how long the customer or the next team waits for the result.

Write down how you measured each one so that you can measure it the same way after launch.

Check whether the numbers work

With a baseline you can estimate the return before you pay for anything. The time saved is only one side. The other side is what the automation costs to build and to keep running, which includes subscriptions, usage charges, maintenance when a connected app changes and the staff time spent checking failed runs. The guide on how to calculate automation ROI goes through each input, puts a fully loaded cost on staff hours and works through one example from start to finish, including how a drop in volume can turn the result negative.

If the numbers only work under the most optimistic assumptions, choose a different process for the first project.

💼

Calculating End of Service for your employees?

Use our 100% free, private Gratuity Calculator updated for UAE Decree 33.

Calculate Gratuity
100% Free Tool by ArtinWebs

For teams in the UAE, the free AI ROI calculator takes salary, visa, insurance and workspace costs in dirhams. Enter your own figures rather than relying on the example it shows when the page loads.

Decide what kind of automation the process needs

Not every process needs AI. There are three broad options, and the simplest one that does the job is normally the one to choose.

  • A fixed workflow follows the same steps every time. It suits structured input and clear rules, such as copying a paid invoice from one system to another.
  • A workflow with one AI step stays fixed but uses a model where something has to be read or sorted, such as pulling the order lines out of a free-text email. The guide to intelligent automation in Dubai covers when that AI step adds value and when plain automation is enough.
  • An AI agent decides its own steps within limits you set. It suits open-ended requests where the path changes from case to case. The guide on how AI agents for business work explains when an agent beats a fixed workflow and which tasks to hand over first.

The tool comes after this decision, not before it. If you are looking at workflow platforms, what n8n is used for and how it works shows how a workflow is built from a trigger and a chain of steps, which is a useful picture whichever tool you end up using.

Decide who builds it and who owns it after launch

There are three common routes. Someone on your team learns a no-code or low-code tool, you hire a freelancer, or you hire an agency. Any of them can work. What matters more than the route is that three questions have answers before work starts.

  • Who owns the accounts? The automation platform, any AI model account and every connected login should be in your company’s name, so the automation keeps running if the builder leaves.
  • Who receives the alert when a run fails, and who fixes it? Connected apps change their logins and data formats, and a workflow that nobody watches can stop without anyone noticing.
  • Who is allowed to change it, and how are changes tested before they go live?

The NSW guide’s rules of success make the same point at a larger scale, asking agencies to manage bots as IT assets with access controls and security checks.

If you go to an outside builder, how to choose an AI automation agency in Dubai and what to ask lists the questions to put to an agency before you sign, including account ownership, hosting and support after launch. Most of them are just as useful when you are talking to a freelancer.

Check the data rules before you connect anything

An automation moves data between systems, and that data often includes customer names, phone numbers and order history. In the UAE, the government portal’s page on data protection laws says the personal data protection law, Federal Decree-Law No. 45 of 2021, “prohibits the processing of personal data without the consent of its owner,” with exceptions, and that it applies to processing through electronic systems inside or outside the country. The same page also lists the DIFC’s separate Data Protection Law, DIFC Law No. 5 of 2020.

Wherever you operate, list which personal data the automation will read or send and which outside services it passes through, including any AI model, then confirm that this fits the rules that apply to your business and what your customers agreed to. This is general information, not legal advice.

Run a small pilot with a clear decision at the end

Build the first version for the common path, leave the exceptions with a person and test it before it touches real customers. The NSW lifecycle says solutions should undergo “rigorous testing to ensure they work as intended before production deployment.” For a small company that means running it on real past cases first, then side by side with the manual process for an agreed period, with someone comparing the two sets of results.

Agree the success measures before the pilot starts and take them from your baseline, for example minutes per run, the error rate and the share of cases still handled by hand. At the end, make one of three decisions. Keep it and widen it, fix named problems and run the pilot again, or stop. Stopping is an acceptable result, since finding that out cheaply is the reason to run a pilot.

If the automation uses AI, keep a way to switch the AI part off. The NSW guide says organisations should be “ready to remove any AI part from their automation at any time,” and the same habit protects a small company when a model starts giving poor answers. After launch, keep measuring the same four numbers so that the comparison with the baseline stays honest.

Where to go next

For the cost side in full, read how to calculate automation ROI. If an outside builder is likely, the questions to ask an AI automation agency are worth reading before any first call. The next guide in this series covers how to choose which process to automate first when you have several candidates.

If you already have a process in mind, estimate your own numbers first. Then, if you want a second opinion on the plan, book a free 30-minute strategy call and bring your notes on how the work runs today and the baseline you measured.

Arezoo Mohammadzadegan
About the Author

Arezoo Mohammadzadegan

AI Programmer & Digital Marketing Strategist at ArtinWebs (A M H R MARKETING MANAGEMENT L.L.C). Specialist in Artificial Intelligence development, AI agent programming, n8n automation workflows, and digital transformation. Based in Dubai, UAE.