Process mining, RPA
5 Challenges Automation Rates Face Today within RPA Projects
This is a follow-up in the series presenting key conclusions on the current state of assessing automation potential within Robotic Process Automation (RPA), while also providing insights into its connection to Process Mining.
In our previous article, we explained the types of processes we know, defined RPA automation rates, how they are calculated, and why they don't tell you much when being presented with no context.
Today, you'll learn about the 5 challenges automation currently faces, and how you can add context to your automation rates, thanks to Process Mining. Ready?
1. Bots Can't Automate Processes, Only Tasks
As mentioned in the previous article, “bots” are usually not able to perform processes end-to-end. Automation takes place only within separate, simple tasks (remember Robotic Task Automation?).
There are a few things to be taken into account here - exceptions that happen and aren’t programmed, outliers that need human decision-making or manual input. Say you have order processing, for example: a bot may be setup to process all standard orders below a certain threshold, all the rest need to go through a person to ensure its validity.
So even if you automate 80% of your process, the residual 20% will probably need human input, and therefore can't be automated as this may involve more complex activities.
Therefore we recommend looking at automation rates with a Himalayan grain of salt, and always ask for more context. Before even running an RPA initiative, you should first analyze your processes to get a complete picture and understanding of the “as-is” state.
2. Bots vs. Human Users
Based on the process analysis, you can much better decide which processes are automation-ready and which are not. Process Mining software can play a crucial role in this step, taking event logs from your IT-based systems and mapping out your processes.
Here comes another issue with automation rates, though.
Based on the event logs, e.g. from SAP or Salesforce, you can tell the user who performed the activity. Yet, unless you manually flag it or cluster bot-executed activities, you can't really tell whether it was an employee or a bot, most of the time. So, as a step to include in your data pre-processing exercise, this will need to be taken into account in order to distinguish between all the different technologies and humans active here.
3. High-level vs. Granular View on Business Processes
The current state of Process Mining enables the solutions to track processes on a high level only, based on the event logs from IT systems. As such, you can't say what should be automated, because you don't really know how many and what kind of lower-level steps such an activity involves.
Picture it as submitting an invoice for approval. You receive an invoice to be processed, which includes several steps of filling in a form and uploading an attachment. Once finished, click “Send”, and your system logs it. However, this whole set of activities would most likely, with most systems, be represented only by one activity. Therefore you cannot see all of the actual activities that have to be performed, in order to accomplish this, as you are not able to track all the various steps carried out in-between.
Top level activities that Process Mining software gets from the system do not show the interface level activities hidden from sight. Thus, automation rates can't identify how many of them can be automated.
So stating that the process has 30%, 50% or 90% automation potential might come out as misleading.
4. No Benchmarking in RPA
If someone says that 20% of your Purchase-To-Pay or Order-To-Cash process can be automated, what's the first thought that comes to your mind? “How do you know that”, right?
And that's the right question to be asking.
RPA is still a relatively young field of Digital Transformation. Even though the first numbers are rising to the surface, it's still not enough for solid benchmarking. Some estimates – such as that about 70% to 80% of rules-based processes can be automated – are floating around, but real evidence is missing.
Additionally, even if we had these numbers, no one has disclosed which activities are most often automated in specific processes, and whether it's always those same activities in every RPA project. For example, you can't tell whether the Purchase-To-Pay process in your organization should be 35% automated.
There are no benchmarks in the market that would tell you how many of the process activities can get automated, as well as no one truly being able to tell you what level of an automation rate for such and such process is good, and what level is too low.
So the questions persist. Is 100% the ultimate goal? How much decision-making do you want to leave to bots, with low to no learning ability?
On top of the benchmarking issue, another challenge with automation rates arises: the uniqueness of your company.
Yes, you function in a particular industry which has some common traits and specifications. Yet, every organization is different and unique on its own, even compared to the organizations within your industry.
What you've been through, your clients, your history, your employees, your products and services, your workflows and your processes - all being very specific to your company.
When it comes to automation, all of this plays a serious role. Saying that your PTP process automation potential is 80% because your competitor obtained such a number, is – even considering some overlaps - walking on thin ice.
How Can Process Mining Get Your Automation Rates on the Right Track?
To be clear, no one's saying automation rates shouldn't be used in your RPA initiative. They can give you a glimpse into the current level of process automation. What it lacks, is more context and granularity.
Here, Process Mining software enters the game.
Thanks to the on-screen recording functionality, it brings more context into your processes and activities, enabling you to see the even lower-level steps of an activity, performed either by robot or human. By doing so, you'll get a much better understanding of potential automation candidates.
Furthermore, Process Mining can help you run bot simulation. This means you're not only being told what your automation rate is, but you'll also add deeper context of what the process is going to look like once automated. With knowledge like this, you'll be able to calculate more meaningful and insightful, objective ROI expectations.
So, in order to get true value from the automation rates when running an RPA project, keep in mind what you learned today: be cautious of any number that is presented to you.
Why? Because you can't usually differentiate the activities performed by a bot vs by a human, and you also need to consider the fact that bots can't usually perform activities end-to-end. Current solutions can only track bots on high-level activities. In addition to that, don't lose sight of the fact that your organization is unique and benchmarking in the market is not that solid, at least for the moment.
Have you ever tried calculating your processes' automation rate? If so, how close to reality have they become? Let us know in the comments.
If you are eager to get a better understanding and context to your process automation potential, just drop us a call.
02. 07. 2019