Despite all my years of experience in the automation industry, and despite having an inside track on software sales, I still reckon that it is a Herculean task to choose with confidence when confronted with the claims made by most vendors in the space.

Even the use of terminology can be misleading; it’s abbreviation city, buzzword bingo. Most vendors claim to have the best technology, the greatest domain experience, the fastest return on value and, of course, the highest degrees of accuracy. And yet, when you actually get to see the technology in action, or start to scope projects, it’s often only then that you start to confront the reality.

The other frustration I have is the number of vendors that try to pass off 3rd party technologies as their own, when really, it’s simply white labelling. So, what drives this behaviour? It’s gaps in their solutions. And the biggest gaps being covered up relate to the ability to deal with real-world data (often referred to as un-structured data) because, guess what, 80% of the data businesses need to process is classed as ‘unstructured’.

The next biggest gap is variation. How many customers or suppliers supply or request data in exactly the same way? And don’t mention the other Achilles heel of any number of solutions; their inability to deal with major languages. Don’t get me started on the need to do this whilst actually delivering the levels of accuracy that were promised.

Getting the solution You Need

So how do you get the solution your business needs and avoid another failed project because you have been oversold?

The first thing I would say is don’t fall for the buzzwords. Almost every company bangs on about ‘AI this’ and ‘ML that’. Don’t get me wrong, these are very important tools that are absolutely necessary in any good automation platform, but they are also the terms most overused by companies that simply don’t have what they’re claiming.

It really baffles me how some people get away with trying to present technologies like RPA and OCR as artificial intelligence. Just to be clear, there is absolutely nothing wrong with these technologies. When used in the right way they’re fantastic. But people are waking up to the fact that they are often being oversold. RPA Bots are not intelligent and they can’t deal with unstructured data [examples of unstructured data include data in non-standard formats such as emails, letters, reports, claims, forms containing handwriting, stamps, images, signatures etc – Ed.]. OCR also has its place, but again it’s not intelligent. It’s a 30-year technology that struggles to deal with the variations found in most business processes.

At the other end of the spectrum there are multi-million dollar, super-complex, all-singing, all-dancing AI systems. They may yet help cure cancer or fight climate change, but for most business processes they’re the equivalent of taking a sledgehammer to crack a nut. All too often they take far too long to deploy and, before the project is finished, the business requirement has changed or disappeared entirely, and all that remains is a huge bill. You’re left trying to justify a multi-million pound spend on something that’s no longer needed, not a position any of us want to find ourselves in.

Asking The Right Questions

So, what should you do?

Put simply, look for a solution that does what it say’s on the tin. ‘That is far easier said than done,’ I hear you say. And you would be right. But there are questions that you can ask that should help you get the right solution for the task in hand.

  1. Can the solution really deal with un-structured data? It may not be an issue now, it will be in the future as you expand the project. And don’t forget, 80% of data typically handled by organisations is classed as unstructured.

  2. Can it deal with high degrees of variation including languages? If you are a global company, you need a solution that can support your global operations. Even if you’re not, but have international ambitions, you don’t want to be held back by your technology.

  3. Ask for a real-world demo from a use case that has actually been deployed rather than one prepared by the marketing department. Even if it doesn’t match your use case exactly it will give you a far better view of the capabilities.

  4. Ask for ballpark costs early. This will help avoid wasting lots of your precious time on something that doesn’t stack up financially. A simple way to do this is to ask for likely costs for low, medium & high complexity processes. This will then you give you a range that allows you to make some provisional calculations and determine if you are going to get the returns that warrant further investigation.

  5. Run a pilot with real-world documents – real-world documents help produce real-world results. This, coupled with mutually agreed success criteria, will give you a far better view on whether it’s the right solution for you.

  6. And, last but not least, be wary of ‘Market Place Pre-Built Models’ or ‘white label’ offerings, for the simple but very important reason that you don’t want excuses along the lines of ‘it’s not our tech, it’s theirs’ when you have a problem in production.

So, in summary, there are some great solutions out there but, as ever, it’s buyer beware. Don’t believe everything you read. When buzzwords are flying around you could get stung.