So you’re interested in cognitive RPA? You’re not alone. Almost half of enterprises implementing RPA in the next six months will be doing so with RPA solutions that offer cognitive capabilities.
But just as the processes you can automate using cognitive RPA are more complex, implementing the software itself can be tricky. It’ll take longer than traditional RPA, and it will probably cost a little more, too, because you’ll need more powerful hardware. But the results will be worth it, since because of its capability to process unstructured data, cognitive RPA actually accelerates the already-significant returns that enterprises get from RPA—achieving ROIs of up to 300% in just months, according to McKinsey.
To help you out, we’ve put together a checklist of things you should do to succeed with your RPA deployment.
STEP 1: Select the right processes to automate first
Don’t be too ambitious when you start cognitive automation. Ernst & Young found that as many as 30% of initial cognitive RPA deployments fail—typically due to common mistakes that could be easily avoided. So pick a process that allows you to take a few baby steps and fall a few times without bringing down important enterprise operations. Semi-structured documents such as invoices, POs and balance sheets are a great place to start.
STEP 2: Build a solid business case
Automate a process that will give you a decent return on investment (ROI) within a reasonable timeframe. PwC found that the payback period for a cognitive automation initiative in the financial services industry can be as fast as six months. Leading areas of high ROI were operations, finance, and IT.
What’s the way to ensure ROI? Pick processes that involve sufficient number of people—at least 10 employees in developed countries where labor costs are high, or at least 50 employees from developing geographies. Also make sure that you’re processing sufficient number of documents—at least a few hundred a day—or are using the system for multiple processes to make it worthwhile.
STEP 3: Get buy-in from senior management and IT
Many cognitive automation projects start with business functions such as finance or HR, or within shared-services operations that provide businesses with a range of services ranging from legal to accounting to payroll. Leaders within these business units recognize the potential value of RPA, and eagerly jump in to reap the benefits they’ve been hearing about.
But before embarking on a cognitive automation initiative, make sure you have buy-in from both senior executives and IT. Businesses have found that winning a champion in senior management greatly increases the chances of getting the resources needed to make the project a success. If your business already has an RPA center of excellence (CoE), that is the ideal place to start.
Step 4: Follow the 80/20 rule
Use the Pareto principle to keep your initial cognitive automation projects manageable. Also known as the 80/20 rule, the Pareto principle estimates that, for many objectives, roughly 80% of the success comes from just 20% of the effort. So if you’re trying to process invoices in a global enterprise, don’t attempt to automatically process invoices from every language. Stick to English and Spanish at first, and you’ll reach 80% of your processing goals. Likewise, choose one department or division to start with, rather than implementing a cognitive automation program enterprise-wide. Start small with a pilot and fine-tune it before you attempt to scale.
STEP 5: Select a representative set of documents to “train” the cognitive system
Cognitive computing is data hungry. The more data you give it, the more accurate the results will be. For example, if you want to automate invoice processing, provide a good representative sample of invoices from various vendors, as the layout and even the field labels can vary greatly from company to company. Generally, several hundred if not thousands of documents should be provided to the system to ensure success.
STEP 6: Scale and expand
After you’ve succeeded with your initial implementation, gradually introduce even more complex processes—for example, moving from semi-structured to completely unstructured data. So rather than simply responding automatically to customers’ email requests for payment status, you could program a cognitive bot to manage emails from a vendor support portal, where the queries are more varied, and the interactions thus more complex.