To paraphrase Warren Buffett’s timeless advice to investors for business technology buyers, “never invest in a technology you cannot understand.”
Given the complexity of modern technology, this is not easy advice to follow. Clarifying and understanding current technology trends amid perpetually accelerating growth and constant change is daunting to say the least. In particular, intelligent automation (IA) technologies have never been more hyped, numerous, or popular. This landscape is a minefield for businesses and teams seeking to apply new technologies wisely, and side-step the high likelihood of digital transformation project failure.
Heroically, the Shared Services and Outsourcing Network (SSON)—a community of experts that focuses on shared services, global business services, and outsourcing—attempted to make sense of the IA tech space, which includes artificial intelligence (AI) and machine learning (ML), in their recent report: “The Top IA Technologies Driving Business Resiliency and Digital Excellence in Shared Services.”
This article offers an overview and analysis of the SSON’s report, including explanations and findings on:
Perhaps surprisingly, the SSON’s report emphasizes that despite being decades into digital transformation, paper is still very much with us. Documents, both digital and physical are everywhere, and growing, representing a massive amount of business data that is difficult to access and use. This is because document data is unstructured, meaning it lacks a clear manner of organization, making it difficult to process and analyze.
Enter intelligent document processing (IDP), an AI-based data extraction method used to glean information from documents of all shapes and sizes. The research describes the value of IDP to reduce legal and financial risk by serving as the “front door” of the enterprise: “It considerably reduces the volume of manual processing, while at the same time improving accuracy and control.”
Yes, IDP is a “game changer.” But the report misses the bigger picture: unstructured data processing (UDP). Instead, the report delves into explaining in detail the nuances of optical character recognition technology (OCR), which is at the core of many IDP solutions, which the SSON essentially describes as OCR + machine learning (ML.
Forget documents for a moment, unstructured data, which includes things like images, video, text, biometric data, chemical structures, and more, makes up 80% of all business information, and it’s growing faster than any other data type.
Unfortunately, as the report points out, IDP is a slow, complicated approach to handling this data. The report explains: “IDP implementation is typically a multi-year endeavor with a carefully constructed plan to release the core capabilities, and then release increasingly complex components over time.” For context, the report adds: “It is worth noting that IDP implementation, along with the attendant integrations, are significantly more complex than RPA.”
Unsurprisingly, their survey found that two-thirds of businesses are at least a year away from even getting started with IDP:
The SSON describes IDP solutions as both specialized and verticalized, hence the required templates and implementation complexity. However, the report fails to highlight existing AI-powered solutions that allow users to process any unstructured data type (including documents), often without learning to code. Self-service, no-code unstructured data processing platforms like super.AI are far more flexible and cost effective than convoluted legacy IDP tools.
The term “AI” has been thrown around so much that its meaning has become vague, often leading to confusion around the specific technology the term refers to. Despite this, the SSON’s research found that nearly a quarter of businesses have started in-house AI departments, complete with data scientists, data stacks, and data infrastructure. That sounds… expensive. SSON would be doing it’s audience a service by pointing out the existence of no-code AI.
No-code AI for unstructured data processing serves as a barrier-free and cost-effective path to realizing the value of AI for business, and specifically for intelligent automation—which is the focus of the report. Overlooking this option may steer a lot of well-intentioned people, eager to leverage intelligent automation such as AI to solve complex problems, to invest in limited and needlessly complex solutions.
What SSON’s report does point out is the question of AI performance, explaining that “while there are many performance criteria useful to the engineers and data teams that develop the models, the user generally cares mostly about accuracy and error rate.”
For zeroing in on the issues of accuracy and errors (and for providing a mini-course on ML theory!) the report’s author, Lee Coulter, Chair of the IEEE Working Group on Standards in Intelligent Process Automation, deserves a huge shout out. Companies building their own AI/ML models run aground on quality and error rates (not to mention on cost and time, too). But what if you could guarantee both? “Assembly line” AI ensures quality and guarantees output accuracy, in part by enabling internal and external human-in-the-loop. This includes post-processing of false positives and negatives by human experts, which is then used to improve the underlying model through continuous learning.
The SSON report refers to RPA as its own, separate trend, while also mentioning that the technology is moving to platforms that handle entire process life cycles. That means RPA is rarely just structured task execution anymore; providers are branding their solutions as full-service intelligent automation platforms. The rest of the technologies listed in the SSON’s report, such as IDP, AI, and natural language processing, are actually just typical offerings or integrations included in intelligent automation platforms.
Zooming out, the report also considers business process automation (BPA) platforms reveals a market filled with confused buyers:
And it seems BPA platforms aren’t too sure where their strengths lie, either:
“BPA providers bring the capabilities discussed here (OCR, IDP, ML/AI, and NLP/G/U) and more together in an integration platform. The depth of integration among the different components is an important factor to consider. It takes time to fully connect the varied technologies in a meaningful way. It is also important for BPA companies to sort of “declare” where they will compete for native/ internal capability leadership. There are far too many places requiring major investment to be a leader in more than a handful."
The implied cautionary tale here is that while businesses need the full suite of end-to-end process automation tools, beware of intelligent automation platforms that may be marketing themselves as more capable than they actually are.
On a somewhat unrelated note, oddly subsumed within the BPA section of the report, there is a highlight on the importance of no-code and low-code capabilities (NCLC). While we’re glad to see NCLC mentioned, in reality no-code is 100% a major technology movement that encompasses nearly all technologies; it’s certainly not a footnote to BPA.
Natural language processing and related technologies are essential to the unstructured data processing toolkit. While offering immense value, NLP is more likely to be useful when embedded within automation solutions rather than as a stand-alone application.
So while the SSON report shows healthy honesty in sharing that more than a quarter of businesses haven’t “the slightest idea” how stand-alone NLP would be useful to them, that fact in and of itself isn’t surprising. (Perhaps with this group as inspiration, the report dives into a disambiguation of terms relating to NLP.) And we see that once NLP is put in an automation context, more than three quarters of survey respondents identified NLP as central to resolving additional gaps and problems in their automation workflows.
To better illustrate its utility and value, NLP should be contextualized within the broader concept of unstructured data processing. Use cases that involve natural language processing cooperating with other methods of unstructured data processing, such as vector similarity search, make it possible to build intelligent chatbots and automate e-commerce product listing quality checks.
To tie together the plethora of intelligent automation technologies, the report ultimately argues that all of these technologies are the fast-evolving building blocks of a digital labor revolution.
“Digital labor in 2021 is approximately the same as offshore labor in 1991. One possible exception; that of speed. Digital labor is advancing its capabilities year on year at an amazing pace. It is time for a mindset change. RPA and IA are not just novel technologies. They open the door to digital labor which will permanently change corporate operating models as surely as shared services and outsourcing did.”
While a sizable majority of businesses “get it” and are beginning to build a digital workforce in some way, the diversity of approach is fascinating, and the honesty of the 16% that “aren’t even sure what that means” is refreshing. It will be interesting to see how perceptions evolve in subsequent research.
If you, like most of us, are looking for clarity on intelligent automation technologies, this report is invaluable. The SSON’s report is helpful reading for anyone attempting to grasp the broad and constantly evolving landscape of intelligent automation technology, and it’s author deserves applause for the effort to explain, disambiguate, and shine an honest light on intelligent automation technologies in 2022.
The report includes admirably accessible content to make sure you really understand what’s under the hood with these technologies before (or after) you make the move to invest in solutions for the promise of the business advantages they offer (something Warren Buffet would approve of!).
However, as helpful as this report can be, the overarching themes of unstructured information and process complexity are left as undercurrents, implied but not dealt with head-on. For so many of the intelligent automation platforms and technologies the research mentions, the shared pain is unstructured data, whether paper, document, images, or language.
A no-code, unstructured data processing ‘tech stack’ already exists, and it’s ready to integrate with your existing investments and automate complex processes previously considered immune to automation. Check out the following resources on unstructured data processing to learn more about the technology, and don’t hesitate to reach out to super.AI to learn how UDP can benefit your specific application scenario.