How to start your own robot revolution with intelligent process automation (iRPA)

Business Challenges

For years, central banks and leading economists have been pointing out that poor productivity gains in major economies is holding back wealth creation. Ten years after the last major financial crisis, flat economies, austerity & process outsourcing have become the new normal and populations are getting restless. People want to feel like progress is being made and that their children will be wealthier and healthier than they are now.

But how do we achieve the growth needed to bring back economic progress?

The modern office worker’s time is already filled with emails, form filling, reconciling, reporting, document writing, presentation prep and meetings and in some of the world’s largest economies, jobless figures are historically low and now are falling. If these trends continue, these economies will be at capacity or inflation will re-emerge. 

So, coupled with the pull from technology maturing, and the push from the economy running low on human resources, I think it is time for an office robot revolution. Like the robot takeover of factories late last century, where jobs manning the production lines switched to programming, supervising and fixing production robots, ‘intelligent’ office robots will likewise take over and automate many office processes. 

So what services are out there to help me automate my manual activities and processes now? 

There are already numerous AI/ML services and algorithms in common operation that we use often, possibly without realising it; smart phone users are likely familiar with speech and face recognition and effective translation has been available for a number of years. And who doesn’t use internet search tools?

In a business process context, businesses should prepare for a rapid uptake of these smart algorithms, as there are many opportunities for automation to free up staff to work on other, higher value tasks.  The list below highlights some key examples of smart robotised processes that HCLTech is working on with customers today:

• Image Scene Recognition – AI can easily be trained to recognise objects and text within an image with similar precision to a human – only tirelessly and at much higher speed. We are working with a client to rethink their service delivery and customer service processes to include street operative image capture. Currently, re-visits to sites cost a significant % of their revenue. In the future, the site image capture step will:
  1. Help operatives notice when they have missed an item to be picked up
  2. Provide evidence of how the site was left
  3. Provide data for smart algorithms to further analyse and assess sites to flag issues offline
• Optical Character Recognition (OCR) - We are helping a utilities client apply OCR to the process of commissioning a new meter. Instead of form filling details by typing on a touch screen, in the new process an image is taken of the newly installed meter.  OCR reads the meter reading, meter ID number and automatically enters them in the corresponding fields.   This reduces costly errors, site re-visits and also provides evidence of the state and status of the meter at the time of installation.  As per the Image scene recognition example above, a wider-angle image shows the site was left clean & tidy
• Topic Detection – Unstructured data is no longer a barrier to automating processes.  Many companies use email to send and receive important documents - like invoices and purchase orders or have lots of information to sift through such as CVs in recruitment.  We are working with clients to use AI & ML to classify emails, documents and communications in order to queue and process them using or robot or “bots”.  The queuing is important because this can be sampled by humans as well as the end processing.   For example, a classification bot thinks it has found an invoice, and puts it in the invoice queue. The invoice ingestion bot then reads it and inputs the information into ERP system and reconciliation takes place.  A sample of 5% is then sent to a human for checking. Perhaps a new client is using a new unknown format, so the topic detection bot needs retraining; or a new format doesn’t make sense to the ingestion bot and that needs additional examples and some mapping rules. When fixed, the bot gets iteratively better
• Face and voice print recognition – This technology is not just for high end smart phone users. Face and voice recognition are now available to make computer human interaction more natural. Machines can recognise who is operating a process and adjust seamlessly. We are working with customers to use their field devices to recognise the operative, thus removing tedious repetitive log-ons and improving security.  We’re also working with a plant manufacturer to optimise equipment and machinery settings automatically to recognise and interact with elderly people.  This more than anything can move us towards machines that ‘just work’ because they recognise and react to the operator.

Unstructured data is no longer a barrier to automating processes.

At the moment, these intelligent algorithms require significant training data -  for example, the customer needs to provide both images with traffic cones and barriers and without in order for the machine  to distinguish between sites that have been left clear and those that have not.  But in the future, we expect pre-trained algorithms to become available, and customer trained algorithms to be valuable in their own right for their trained data set.

Sounds like the AI/ML time has come, how do I get started?

There are 5 key enablers to AI/ML that must be in place to leverage intelligent process automation opportunities:

1. Data, data, and more data – AI/ML are not programmed in the classic “if/else” sense.  Rather they are trained on datasets to recognise patterns; be they a voice, a face, a product on a shelf, a word or a brand logo.  So having data to train and then test an algorithm is key.  Don’t throw away data!  But make sure the data you keep has enough information to derive meaning. For example, if you have pipe pressure telemetry data, make sure you know where the reading was from, what kind of pipe it is in, and which valves are close by, etc.
2. Cloud Platform – AI/ML typically requires ‘big compute’ - from both processing power, in-memory and large data storage capacity.  It is uneconomical for most organisations to host ‘big compute,’ so buying services from Amazon, Google, Microsoft or SAP – the latter incorporating business services and packaged AI/ML into a mature ERP solution - is the way to go
3. Adopt an agile mindset and working practices – while the Agile (“think big, start small, scale fast”) methodology has been around for a decade or so, many organisations still plan and execute in a fairly rigid hierarchical waterfall fashion. Implementing intelligent process steps should certainly be approached in an agile manner.  But also the operation of AI/ML processes - where results are not certain and throw out the occasional curve ball (‘seaming delivery’ for cricket fans) due to their statistical nature – lends itself much more to Agile ways of thinking and working with delegated responsibility to think, act & fix
4. Clear Strategy - while projects can be Agile and exploratory, it always pays to have a corporate strategy that the top level both buy into and drive throughout the organisation.  The top brass should have a decent grip on where their costs and inefficiencies are, so are best placed to point to the areas where intelligent algorithms should be applied.  Avoid high risk areas for early projects, and consider established packages with intelligence built in - such as modules from ERP providers
5. Skills & Learning & Change Management – for any change to be successful, it takes the people involved to learn, understand & implement and for the staff in the organisation to adapt and adopt the new process, skills and roles they perform.  Related to adopting an Agile mindset, this is probably the most difficult to achieve and takes continuous motivation and energy.

For any change to be successful, it takes the people involved to learn, understand & implement and for the staff in the organization to adapt and adopt the new process, skills and roles they perform.

It’s been fun writing up my thoughts on iRPA and the coming intelligent computer revolution and I look forward to providing updates in the future on where we’ve got to and where we’re heading next.  I’m set to retire in 2040 and I’ll be sure to look back to see where we are. I leave you with a reminder summary of the benefits of RPA.

If you have any comments, queries or corrections, I’d be delighted to hear from you in the comments below

Summary of the benefits iRPA aims to achieve