Automation is one of those words that means everything and nothing. RPA vendors promise it will replace half your workforce. Workflow tools promise it will save hours every day. Process mining vendors promise to find the inefficiencies you didn't know you had. The reality is more modest, more practical, and more useful than any of the pitches suggest.
What You Need to Know
- Automation works best on high-volume, rules-based, repetitive tasks. The more judgement required, the less suitable it is for automation
- Start with process documentation, not tool selection. You can't automate what you don't understand
- RPA (Robotic Process Automation) is useful but brittle. It works by mimicking user actions on existing interfaces, and breaks when interfaces change
- Workflow automation (Zapier, Power Automate, custom) is more reliable for most use cases than RPA
- The ROI calculation must include maintenance costs, not just time saved
Three Types of Automation
The automation space is confusing because vendors use the same words to mean different things. Here's a practical taxonomy.
Robotic Process Automation (RPA)
RPA tools like UiPath, Automation Anywhere, and Blue Prism create software "robots" that interact with applications the way a human would. Click this button. Copy this value. Paste it there. Navigate to the next screen.
RPA is useful when you need to automate a process that spans multiple systems with no API. If someone is spending three hours a day copying data between a legacy system and a modern one, and there's no integration available, an RPA bot can do that work.
The downside: RPA bots are fragile. They interact with the user interface, and when the interface changes, the button moves, a field gets renamed, a new step gets added, the bot breaks. Maintaining RPA bots is an ongoing cost that vendors tend to understate.
30-50%
of initial RPA projects fail to meet ROI expectations, often due to underestimated maintenance costs
Source: Ernst & Young, RPA Survey 2021
Workflow Automation
Tools like Zapier, Microsoft Power Automate, and n8n connect systems at the API level. When X happens in System A, do Y in System B. An invoice is approved in the finance system, automatically update the project status in the PM tool. A customer fills out a form, automatically create a record in the CRM and notify the sales team.
Workflow automation is more reliable than RPA because it uses APIs rather than UI interaction. It's also more limited, because it requires the systems involved to have APIs. But for most modern SaaS tools, they do.
Process Mining
Tools like Celonis and ProcessGold analyse event logs from your existing systems to build a map of how processes actually flow. Not how they're supposed to flow according to the process document, but how they actually flow in practice. The gap between the two is usually where the automation opportunities live.
Process mining is the most sophisticated and least commonly adopted of the three. For organisations with significant transaction volumes and complex processes, it's genuinely valuable. For smaller teams, it's overkill.
Where to Start
Don't start with a tool. Start with a list.
Step 1: Identify Candidates
Survey your team. Ask: "What do you do repeatedly that doesn't require thinking?" The emphasis on "doesn't require thinking" is important. Tasks that require judgement, interpretation, or careful decision-making are poor automation candidates.
Good candidates:
- Data entry between systems
- Report generation from standard queries
- Notification routing (when X happens, tell Y)
- File management (rename, move, archive based on rules)
- Status updates across tools
Poor candidates:
- Customer communication that requires empathy
- Exception handling that varies by context
- Anything where the rules change frequently
Step 2: Document the Current Process
Before you automate anything, document exactly how it works today. Every step, every decision point, every exception. You'll be surprised how much implicit knowledge is involved in processes that seem "simple."
The biggest mistake in automation isn't choosing the wrong tool. You end up automating the mess and making it faster.
Mak Khan
Chief AI Officer
Step 3: Simplify Before Automating
Often, the act of documenting a process reveals unnecessary steps, workarounds for problems that no longer exist, and complexity that could be removed. Simplify first. Then automate what remains. Automating a bad process just gives you a faster bad process.
Step 4: Calculate Real ROI
The ROI formula for automation isn't just "hours saved times hourly rate." Include:
- Implementation time (building and testing the automation)
- Maintenance time (fixing it when things change, which they will)
- Error handling time (dealing with the cases the automation gets wrong)
- Training time (someone needs to understand how the automation works)
3x
the typical ratio of estimated to actual maintenance cost for RPA implementations in the first year
Source: Deloitte Global RPA Survey, 2021
If the automation saves ten hours per week but requires five hours per week to maintain, your net benefit is five hours. That might still be worth it. But the decision should be based on real numbers, not vendor projections.
Step 5: Start Small
Pick one process. Automate it. Run it alongside the manual process for two weeks. Fix what breaks. Then roll it out. Resist the urge to automate ten things at once. Each automation is a new system to maintain, and you want to build confidence and capability incrementally.
Our Practical Recommendation
For most mid-size organisations, workflow automation (Zapier, Power Automate, or similar) handles 80% of realistic automation needs. RPA is worth considering only for specific legacy integration scenarios where APIs don't exist. Process mining is worth considering when you have enough transaction volume to make the investment meaningful.
Start small. Prove value. Scale what works.