The Top Issues in Trying to Launch IA with Cognitive from the Start (and How to Succeed)

Add bookmark

The business environment today is tough, but also full of opportunities. Technology continues to drive organizational change, but now it's about making operations smarter by shifting from humans consuming data to machines consuming data. Farsighted companies are already recognizing the convergence of digital, intelligent process automation (IPA), analytics, machine learning (ML) and artificial intelligence (AI) into new organizational models that leverage these capabilities exponentially.

The question is: how and where to start?

Many businesses still assume the formula of “developing a strategy – finding a vendor – implementing a solution” to these free-flowing, all-encompassing intelligent automation (IA) tools. But that approach doesn’t apply. To get the most out of today's opportunities you need to understand the convergence of technology and process in solving enterprise challenges.

Many organizations embarking down the IA road see their first step as deciding on whether to start with desktop, discrete automation, and work from there – or leverage a more “intelligent” platform approach that embeds advanced cognitive capabilities from the start.

Your decision is driven by your objectives 

What are you trying to solve?

The answer, of course, is not that simple. First and foremost, it depends on the objective. While desktop or point solutions can be used to fix a given problem, that approach should not be confused with platform-driven scalability and digital transformation. The real question, therefore, is whether you are looking for a short-term, robotics-enabled fix – for example, to get around legacy challenges; or whether you are looking to deploy IA in support of a new operational model. The latter approach leverages scale and reach; the former provides cost-effective solutions.

Getting to the right option requires asking a completely different set of questions around how to meet your objectives­. The solution is not so much of adopting a "cognitive” strategy, but of recognizing that your objectives require a more engaged approach around end-to-end automation and transformation. Today, the addition of IA means you are able to choose from a range of benefits that solve your most pressing challenges.

The benefits of IA are irrefutable and cover quantitative as well as qualitative results. For example:

  • Quality: reducing manual intervention, producing work that is free of duplication and errors.
  • Speed: working around the clock, speeding up processing times and throughput and increasing capacity.
  • Governance: supporting better compliance by embedding requirements into automation rules.
  • Security: eliminating human misbehavior and reducing risk of security threats and data breaches.
  • Business continuity: allowing processes to be switched easily and smoothly to other servers, which expedites disaster recovery processes.
  • Talent retention: intellectually challenging work leverages the potential of people.
  • Value insights: visibility and data create business insights into processes.
  • Flexibility: automated solutions delivered 24/7.
  • Agility: scale up/down on demand.
  • Compliance: fully maintained audit trail and enhanced controls.

Moving along the IA continuum

As is the case for most transformative technologies, some practitioners are on the leading edge, some are fast followers, others mainstream adopters, and finally, there are laggards. Those on the front end are recognizing that the data derived from process automation can fuel machine learning to drive a cognitive capability offering massive returns. Almost all IPA providers are now engaged in providing not just execution and orchestration capability, but also an intelligence engine that delivers cognitive decision-making and business process insights. What this means is that both on the discrete, single-process end of the scale as well as the platform-driven end, there exists the ability to engage with cognitive from the start. Even providers that started with simple tools are now developing these to tap into data as a cognitive enabler.

The evolving capabilities of IA tools are best described as moving from their simplest versions to more sophisticated:

  • Robotic Desktop Automation (RDA)
  • Robotic Process Automation (RPA)
  • Digitized RPA/Intelligent Process Automation
  • Machine Learning
  • Artificial Intelligence

Most modern-day solutions incorporate some or all of these capabilities, whereby we see a distinct trend of ambitious providers aligning themselves with the end-to-end process solutions that drive proportionately greater change, based on a server based model that has the advantage of being scalable, more easily governable, and more secure from an IT perspective.

“The data derived from process automation can fuel machine learning to drive a cognitive capability offering massive returns.”

Another way of categorizing the array of IA technologies is according to what kind of process they can help and what kind of information they can process. For example: 

  • Robotic process automation can work with standard processes that are rules based, with structured and predictable data. Traditional RPA represents the first stage, for repetitive transactional type work. Roughly 30-40% of exiting processes are likely to be impacted by RPA according to Gartner. Benefits range from cost savings to improved control to error rate reductions. RPA is limited, however, in its capacity to manage unstructured data, leverage natural language processing, or embed judgment. For most organizations, this is where the key focus of robotics activities is. RPA is best used where process predictability and stability is high and a majority of processing can be performed in Straight Through Processing (STP). By definition, these are standard processes that would traditionally be considered for SSCs or BPOs/ITOs.
  • Robotic desktop automation can dynamically “pause” its automation at points in the process where human judgment or decision making is required in order to move the process forward. RDA is best used in processes that are complex or have dynamic inputs that influence how a process should be executed.

[Both the above would fall into the desktop-based, “traditional RPA” implementations best deployed to solve a pressing execution problem.] 

  • Machine learning uses structured, semi-structured, and unstructured data to create high-confidence predictive and prescriptive analytics that can substitute for human decision making in process orchestration. Note however, that this emerging field has enormous dependency on data availability and curation.
  • When prescriptive analytics from machine learning are combined with process execution capabilities, they form a cognitive solution. The data created from automated process execution, combined with other data sources, enables dynamic context sensing and decision making that enables an entirely new level of STP. Where processes were fragmented to allow for human orchestration and judgment, a cognitive solution can provide both decision making and execution.
  • The use of intelligent chatbots supports user interaction, and improves the customer experience. Chatbots can be powered by a set of rules or machine learning, whereby the latter are referred to as intelligent chatbots and act primarily as an interface between humans and robotics.
  • Finally, narrowartificial intelligence will eventually gain mainstream adoption. AI crosses the boundaries of what is likely to happen and what should be done about it …to… what might occur if… AI produces a deductive analytic that has until now been exclusively in the human domain. While there are a few limited examples of AI that get a lot of attention (and seem to dominate the press) for most enterprises, AI is still some time away. It is worth noting that for the purpose of providing business value, cognitive solutions will be the workhorse of modern enterprise. Most things labeled “AI” are, in fact, limited cognitive solutions centered around a very narrowly defined knowledge domain.

[The above would fall into the category of solutions that leverage data insights for increasingly cognitive outputs.]

Leveraging “cognitive” requires data 

It's conceivable that an organization could “launch” IA at any of these stages. However, many of the IPA failures to date are due to entering at a stage that is too complex and where general data poverty exists. Again, it is important to emphasize that as one moves up the continuum, data dependency becomes more important.

A typical challenge in machine learning implementations is that necessary data is either discarded intentionally by legacy enterprise applications, or can only be found in the minds of operators. Starting with RPA and RDA creates a foundation of data on which more advanced solutions can be built. In essence, each generation builds on the preceding one, without the need to replace each other entirely. In this way a more advanced stage can leverage and build upon the successes achieved at an earlier stage.

“For the purpose of providing business value, cognitive solutions will be the workhorse of modern enterprise.” 

As organizations move up the IA continuum, solutions become more sophisticated, and the complexity of projects increases. Despite the obvious attraction of a more evolved approach, the majority of efficiencies or cost savings are still derived from basic RPA and RDA. However, for those organizations with the foresight to build cognitive and AI-based processing capabilities into their plans, benefits will be exponentially higher. While the tip of the evolutionary pyramid is narrower and more specialized, the inherent value-adds of these activities are greatly multiplied via IA – though often more qualitative than quantitative or financial in nature.

At the very top level of IA is where the enterprise experiences a change in its modus operandi, which can propel it ahead of its competition. However, in the initial phase, the financial benefits of RPA can be used to demonstrate the value of IA and thereby gain support for moving along the curve. In addition, RPA and RDA are a necessary precursor to utilizing more advanced solutions. Think of RPA and RDA as the hands that do the work, while ML, AI, etc. are the brains that cannot, as of today, execute on a process. Unless something changes fast, the brains will need the hands for quite some time.

“Think of RPA and RDA as the hands that do the work, while ML, AI, etc. are the brains that cannot, as of today, execute on a process. Unless something changes fast, the brains will need the hands for quite some time.” 

One concern is that many executives are confusing robotics, intelligence, and autonomous capabilities. In fact, the convergence of robotics and intelligence leads to autonomous solutions. What sets all these activities apart is that information is being consumed by machines, and these machines are able to “learn”, “read”, “think” and “decide” on the optimal course of action. Autonomous systems don’t just operate without human intervention, but will increasingly move human operators out of the workflow. 

To maximize the value of smart automation, organizations should pay close attention to enabling effective interactions between different IPA “generations”. Without the ability to interact with each other, different solutions or platforms will be limited in their influence. The stop-gaps that emerge inevitably mirror the inefficiencies of traditional processing – albeit within an automated environment (which, in turn, is less flexible). True, scaled, benefits of robotics will accrue to those who build holistic automation across the enterprise.

No matter which approach – be warned…

Many of the barriers to IA integration lie firmly within the existing IT landscape. Defining an end-to-end solution and then leveraging the appropriate enabling technologies against it is a complex process and requires consideration of target operating models, tools, and underlying data. In combination, these implementations are powerful. But where the necessary integrations are lacking, problems emerge.

The lure of lights out processing has been dangled before practitioners for two decades, and although intelligent automation is getting us closer, there is still plenty of work ahead.  Even if the technology wasn’t an issue, many organization still remain siloed making it difficult to gain agreement across functional areas on process ownership and design for an IA program.

The key hurdle for enterprises right now is integrating solutions into their existing environment. Perhaps the greatest stumbling block is the fact that many corporations are still struggling with the rollout or upgrade of their existing ERP platform – so touting the "next level" of intelligent automation to build on top of that only highlights the current mismatch between expectations and deliverables (although, in terms of bridging the gap between ERP and applications, simple RDA/RPA are good solutions).

Additional challenges concern the lack of bandwidth, and whether the business case really supports the desired outcomes – and let’s not forget that ERP providers themselves are busy developing additional capabilities, like SAP’s S4/HANA, which already incorporate many cognitive services through Leonardo and eventually, perhaps, may replace some of the innovative solutions that are making headlines today.

The decision around where to invest will also, to a large extent, be guided by where the enterprise has spent heavily on IT to date, reflecting its underlying commitment and priorities. Other factors include whether the knowledge to implement the new applications exists in house, along with the appetite for disruption.

Finally, with the future of what we might still call digital transformation firmly focused on cloud-based services, the question of server-based IA may itself become redundant or expose a corporation to future risk. The speed and the agility of the cloud is a huge attraction, especially in contrast to ERP’s regular and clunky maintenance and update requirements. Today's enterprises want quick solutions that integrate fast. IA is part of this, but the vehicle through which it is delivered will need to be as flexible as the solutions themselves.

 ______________________

Tips on starting with intelligent automation

As with any change management, preparedness and planning are key. Here are some suggestions:

  • Start by talking to someone who has actually had the benefits delivered. Don't just rely on a vendor's promises. Get proof
  • Consider the total cost to build, maintain, and even grow your robotic aspirations. The total cost of ownership is far more than the original cost of the bot. Something that seems cheap in the early stages may not end up cheap in the long run if it requires lengthy and complex workarounds.
  • Do not assume you have to automate the “as is” process. Take a critical review and consider how robotic automation may consume data differently to humans. Many manual steps can be eliminated completely and don't even require automation.
  • Before you build up a large center of specialist experts to run a robotic initiative, consider how many people you actually need. There are three basic models for an Intelligent Automation COE: Inside, Hybrid and Outside. Decide what your company wants before you begin.
  • Consider the scalability of your automation even if in the short term you have no greater plans. It's important to be aware of the limitations of scaling within a siloed environment. Desktop practices cannot be scaled.
  • Consider how the planned automation strategy allows you to exploit innovations from your incumbent ERP vendors. If you are locked into one version of your ERP, because you are locked into one version of the user interface and your robotic IP is tied up in the relevant keystrokes, mouse moves, etc. generated from the current version, you are effectively locking in obsolescence. Consider the implications of updating your ERP. For example, most of SAPs productivity improvement is in the user interface. This means you may find yourself in a bind as desktop innovations are integrated into the underlying systems.
  • An end-to-end automated solution connects the process to the technical functionality you need beneath it. The separation of business process and rules from the technical approach allows you to execute journals in Oracle or SAP upgrades if your process does not change, because you are just updating the parameter that tells you where to source the data. You cannot change the underlying standardized APIs because everything else would break.
  • Every time you have a maintenance upgrade, automation maintenance must be considered. Without good planning and “unitization” of automation, updating configuration can be cumbersome. 
  • Machine-to-machine options are generally more stable than the user interface
  • Start small, and grow your capabilities over time.
  • Look at automation for things other than cost, such as CX, quality, consistency, speed.
  • Consider establishing a pilot program that draws on the strength of an experienced partner. Professional services firms are acquiring experience with technology providers, and the specialized skill set needed to make RPA processes work in tax, accounting, administration and other specialized workflows.
  • Involve your team. Employees hear about the threat of robotics replacing white-collar workers but the real-world experience has been that robotics allow workers to focus on more valuable tasks and help employers build higher-functioning teams.
  • Make RPA a business-led initiative, not an IT-led initiative. Bots can be configured within the teams, close to the business processes and expertise required to make them work.

  ______________________

[inlinead]

 

 


RECOMMENDED