Business Process AutomationMap · Automate · Integrate · Measure

Which business processes are worth automating first?

Start with volume, not irritation. Multiply how often a process runs by how long one pass takes and how many people touch it, and the ranking often surprises the team: a quiet reconciliation three people do every Friday can outweigh the approval everyone complains about.

The best first candidates share four traits. They follow rules rather than judgment, their inputs arrive in a predictable shape, the steps have not changed in months, and they move data between systems that already exist.

• Copying records from one tool into another
• Rebuilding the same weekly report from several sources
• Creating and closing accounts for joiners and leavers
• Matching payments against invoices or stock against records

Workflow tools, scripts or custom code: which fits?

When the systems involved have usable APIs, a workflow tool is usually the cheapest route. n8n suits teams that want to self-host, Zapier and Make are quick to set up for hosted connectors, and Power Automate fits a Microsoft estate. Apps Script covers a lot of Google Workspace work with no new licence.

Code earns its place when the logic is heavy, the volume is high enough that per-task pricing adds up, or the data needs joining and cleaning on the way through. Python and Node.js services with scheduled runs and webhooks handle that. Tools that drive a screen like a person should be the last resort, used only where a system has no API, because they break whenever the screen changes.

How do you measure the saving from automation?

Before anything is built, record a baseline for each process: how often it runs, how long one pass takes, how many people touch it, where it waits, and how often it goes wrong. Agree that method in writing, because a saving measured a different way afterwards proves nothing.

After go-live, measure the same process the same way over a comparable period. Report the hours returned, the change in errors and delays, and the hours the automation now costs in upkeep. Where the result is smaller than the estimate, say so; it usually means an exception path is still handled by hand, which is useful to know before automating the next process.

What does an automation cost to keep running?

The build is only part of the bill. A hosted workflow tool charges for seats or task volume, a self-hosted one needs a server someone looks after, and custom code needs hosting and logging. Budget for all three before comparing options.

The larger ongoing cost is change. Vendors revise their APIs, a field gets renamed in the CRM, a form gains a new question, and an automation that ran cleanly for months stops. Failure alerts mean you hear about it from the system rather than a customer, and the handover document means your own team can often fix it. Where changes are frequent, a retainer that covers maintenance is usually cheaper than paying for each repair.

What should stay manual?

Anything that needs real judgment each time: a credit decision, a complaint that needs a person, a discount outside policy. Automate around the decision instead, by routing the request to the right person with everything they need and chasing it when it stalls, and leave the decision itself with them.

Rare processes rarely pay back. A task that runs twice a month can take years to recover what it cost to build and maintain. Processes still being redesigned should wait too, because automating a moving target means rebuilding it. An honest shortlist names these, which is why a report that recommends automating everything should be read as a sales document.

How is sensitive data handled in an automation?

Access comes after the NDA and is limited to what each automation actually needs: a dedicated service account with the narrowest permissions the job allows, never a person's own login. Credentials are created in your accounts and stay yours when the engagement ends.

Every run is logged, so you can see what an automation read, wrote and changed. Joiner and leaver flows are worth automating for this reason alone, because an account left open after someone leaves is a security gap rather than an admin chore. If a process touches regulated data, such as health, payment or personal records, raise it at the start, because it changes where the automation can run and what it may store.