Choosing the Right Path for Intelligent Automation

[SouravSaha]

One of the most commonly asked question right now in the Intelligent Automation is - When should we use rule-based automations, and when should we use AI?

Rule-based automation works best for structured, repetitive, and predictable tasks, such as copying data between applications, Excel manipulations, UI automation, performing validations, or scheduling reports.

AI becomes important when dealing with unstructured or dynamic data that requires dynamic decision making, like extracting data from handwritten invoices, categorizing customer emails, quick assistance chatbot.

In short: If a process can be fully defined using “if-then” logic by a human who knows the process, rule-based RPA is the ultimate way to go.

If it involves context understanding, prediction, or handling unknown variables, combining AI with RPA would be optimal way. Because in this case users do not know the variables and dynamic scenarios that automation need to handle.

Would love to hear your thoughts - how do you decide whether a process should remain rule-based or evolve into AI-driven automation?

[Denis__Dennehy]

I totally agree @SouravSaha.

For most heavily regulated industries (finance, health, etc), I would say a rules-based first approach is essential with AI augmenting only for specific blockers preventing end to end automation (such as validating client documents in the KYC onboarding process or reading a specific document where existing OCR tools fail).  

The reason is that audits of highly governed industries like to understand reasoning in business processes, and companies need to prove that every customer is treated fairly and equally - something that can be difficult in black box LLM AI tools which might be prone to hallucinations. AI cannot be used if there is any chance it will treat customer A differently to customer B - which AI can do.  Set rules must be used.

My experience of automation over many years is the most powerful success stories come from process improvement/design people mapping and improving existing business processes, looking at the end-to-end process and how to take a great bite out if.  I know a banking customer that has automated 25% of its workforce using mostly traditional RPA - so scaling is possible.

The problem is most organisations are not very good at scaling RPA, they often do not have the fundamentals in place to succeed (this is where BP royalty @EmmaKK comes in and advertises ROM2..).  Scaling is impossible in an organisation not great at managing change, not building an automation community with advocates, missing a strong and dynamic leader and exec sponsorship, etc.. etc.. etc..  What happens when RPA capabilities do not scale is that the automation platform gets blamed for being too hard and the leaders look around for a new toolset (at the moment that is all AI) rather than understanding why they are not performing as well as other automation capabilities.  If a company is struggling doing automation with RPA tools, they will still struggle at automation with AI tools.

For me AI is very much a door opener to be able to automate far more using end-to-end flow logic, but you should still be starting with the same automation fundamentals and at a process analysis and design phase. 

As a side note, I've also tried using AI tools built into other automation platforms (i.e CoPilot built into Power Platform) and I have found it totally useless and if anything it slows you down - I can use a tool I know well far quicker with a mouse and keyboard than having a conversation with an AI chatbot and then having to correct the mistakes it inevitably makes.  Blue Prism's flow diagram UI still rocks 20 years after it was conceived, mapping out your business process is far better and quicker than AI prompting (whatever the competition might say!).

[Michael_S]

---

@Denis__Dennehy wrote:

The problem is most organisations are not very good at scaling RPA, they often do not have the fundamentals in place to succeed (this is where BP royalty @EmmaKK comes in and advertises ROM2..)

---

Excuse me whilst I prepare...

[SouravSaha]

Totally with you, @Denis__Dennehy!

People are creating too much hype around AI, especially Agentic AI, which still has a long way to go before it becomes truly smart and useful. If we look at heavy industries, only a few are really using AI, mostly when it comes to OCR. But yes, in the future, as AI improves, it may start making a real impact in automation.

[Michael_S]

On topic - this is something I think about a lot. For me, the most interesting space in automation is when a process crosses through RPA -> LLM -> RPA

• NEED & TRIGGER (RPA)
• ANALYSIS & OUTPUT (LLM)
• CHECKS & ACTION (RPA)

That's how we end up with agentic that has governance built in, right? Because LLMs are wonderful at analysis and summary and even connecting concepts, but far too wild (and subject to too much change) to be allowed to go off and take action unchecked. You need something else on either side. 

[EmmaKK]

Well I'm certainly honoured to be referred to as BP Royalty 👑 (thank you @Denis__Dennehy and @Michael_S)

Great question @SouravSaha  – and definitely one of the hottest topics in automation right now.

Rule-based automation still has its place. It’s reliable, auditable, and essential for regulated industries where fairness and consistency matter. AI comes in when you hit unstructured data, dynamic decisions, or blockers that stop you achieving end-to-end automation.

The way to look at it isn’t “RPA or AI,” but “RPA with AI.” If the automation foundations aren’t in place – governance, sponsorship, change management, community – then adding AI won’t solve the problem, it just magnifies it. That’s why we use ROM2 and the EOM framework to help customers decide when to stay rules-based and when to evolve with AI. The maturity model then gives a clear pathway that aligns with each organization’s strategy and risk appetite. Maturity Model 

Practically, the strongest approach I see is:

• RPA for triggers and execution

• AI for analysis and interpretation

• RPA again for checks and final actions

That’s how you progress towards Intelligent Automation (IA) – the combination of RPA and AI with governance built in. AI extends what automation can do, but it only works if the fundamentals are solid first.

Curious how others are setting guardrails for when AI should and shouldn’t be applied?