Robotic Process Automation (RPA) is software that mimics how a human interacts with computer applications. Instead of clicking through screens yourself, RPA bots follow programmed steps to complete repetitive tasks like copying data between systems, filling out forms, or processing invoices.
RPA works well for high-volume, rule-based tasks where the steps never change. For example, many companies use RPA to extract data from emails and enter it into their ERP system, or to copy customer information between a CRM and billing platform. The appeal is that you don't need to replace your existing software or build custom integrations. The RPA bot just uses the software you already have.
However, RPA has important limitations. These bots break when anything changes in your user interface. If a vendor updates their portal layout or adds a new field, your RPA process stops working until someone fixes it. RPA also struggles with unstructured data like written descriptions or exceptions that require judgment.
The bots can only follow predetermined rules, so they'll fail or produce errors when they encounter situations outside their programming.
How is RPA different from AI agents?
RPA follows a fixed script that mimics human clicks and keystrokes. If your invoice portal changes its button layout, the RPA bot breaks. AI agents, in contrast, understand the intent behind the task.
An AI agent can read an invoice regardless of format, determine what action is needed, and adapt when systems change. Think of RPA as following a recipe exactly, while an AI agent understands cooking.
What types of tasks work best with RPA?
RPA excels at repetitive, high-volume tasks where nothing changes. Copying the same fields from emails to spreadsheets every day, downloading reports from a portal on schedule, or moving data between two systems with fixed formats are ideal.
However, if your process involves reading contracts, making judgment calls about vendor quality, or handling suppliers who send information in different formats, RPA will struggle. Those situations require actual comprehension, not just scripted clicks.
What breaks RPA implementations?
The most common failure is UI changes. When a vendor updates their website, adds a field, or changes a button label, your RPA bot stops working. You need a developer to reprogram it. System updates also cause issues.
If your ERP gets a new version with different menu structures, every RPA process touching that system needs updating. Even small changes, like a vendor switching from one PDF format to another, can halt automation.
How long does RPA take to set up?
Simple RPA processes take 2-4 weeks to build and test. Complex workflows spanning multiple systems can take 2-3 months. But setup time is just the beginning.
Count on spending 20-30% of your time maintaining these automations. Every software update, vendor change, or new exception case requires reprogramming. Some companies find they spend more time fixing broken bots than they save from the automation.
What are the real costs of RPA?
License costs run $5,000-$15,000 per bot per year for enterprise RPA platforms. You also need developers to build and maintain the automations, typically costing $50,000-$100,000 annually depending on automation complexity.
Hidden costs include system downtime when bots break, data quality issues from bots misinterpreting screens, and the opportunity cost of staff time spent on RPA maintenance instead of higher-value work.
What happens when RPA encounters an exception?
RPA bots have no judgment, so they either follow a pre-programmed rule or stop completely. For instance, if an invoice amount doesn't match a purchase order, you might program the bot to flag it for review.
But if the bot encounters a supplier format it hasn't seen before, it will simply fail or worse, enter incorrect data. Someone needs to monitor these failures constantly and fix them manually.
Zamp addresses this by using AI agents. Unlike RPA, Zamp's agents can read unstructured documents, understand context, and adapt to changes without breaking.
When a vendor updates their portal or sends data in a new format, Zamp agents continue working. Activity logs record every action for transparency, and the "Needs Attention" status lets agents flag items requiring human judgment instead of guessing or failing silently.
Can RPA handle documents like contracts or invoices?
RPA struggles with documents unless they're perfectly formatted and identical every time. If all your invoices come as PDFs with data in the exact same positions, RPA can extract it. But real business documents vary.
Suppliers use different formats, handwrite notes, scan at different qualities, or change their templates. RPA requires someone to manually program extraction rules for each format variation. When a supplier changes their invoice template, the RPA breaks until someone updates the extraction rules.
Do I need to replace my existing software to use RPA?
No, that's one advantage of RPA. The bots interact with your existing applications through their user interfaces, just like a person would. You don't need APIs or custom integrations. However, this surface-level approach is also a weakness. Because the bots don't actually integrate with your systems, they're fragile.
Any UI change breaks them. And because they only see what's on screen, they can't access the deeper logic and data validation that proper integrations provide.