Bits on Bots: It’s Time to Automate
Exploring White Spaces in the RPA Market
In our last blog post, we broadly examined what RPA means to us and how it can transform how we work. In this post, we take a deeper look at the white spaces ripe for disruption by automation and AI, tailored to provide insights to entrepreneurs and investors alike. I hope this read is especially useful for an “almost-repreneur” — someone who has the skills and interest to start a company, and is keen to learn about the gaps in the market. With that, let’s dig in. ✍🏻
General RPA software has excelled at doing repetitive tasks quickly and with less error. However, AI is transformative in that it is allowing for machines to handle much more variation in tasks and automatically adjust to differences in the job — departure from how we look at traditional RPA software today. There’s now a meaningful segment of work that humans can’t do well, but RPA can. Few areas come to mind where:
- a human “crunches the numbers” and makes a decision once a day, but a RPA system is able to update its decisions 60 times per second
- a human decision maker struggles to discern and interpret nuanced differences within a colossal trove of information
- decisions that rely on data external to the organization — the shipping congestion at a route passing through Singapore may affect inventory and pricing decisions for a Walmart store in Kansas
To better understand the automation gaps across industries and why it’s hard for startups to target horizontal use cases, we need to analyze categories of work activities, each with varying degrees of technical feasibility. Technical feasibility is a function of an activity’s potential to be automated by adopting currently demonstrated technologies. Of course, considering the technical potential isn’t enough — the actual level of automation will reflect the interplay of the technical potential, the benefits and costs, the supply-and-demand dynamics of labor, and regulatory and social factors related to acceptability.
Notwithstanding, understanding activities that are most susceptible to automation from a technical perspective could provide a unique opportunity to rethink how workers engage with their jobs and how digital labor platforms can better connect stakeholders (Exhibit 1).
As investors, we look for the white spaces or areas where a critical mass of RPA startups has not entered — could be a new market or gaps in the existing market or product lines. Duncker’s 1945 radiation problem — the most famous hypothetical problem in all of cognitive psychology — showed us that analogies in different contexts aid us in our probe for solutions to conceptually similar problems in our domain.
To explore the untapped opportunities around automation, it’s helpful to assess RPA potential in a “reference class” of disparate domains.
Scale 👇🏻 denotes the degree of automation potential with current technology.
⚫⚪⚪⚪⚪
🛒 Salesperson for a Hybrid Retailer
⚫⚫⚫⚫⚪
Predictable physical activities — workers carry out specific actions in well-known settings where changes are relatively easy to anticipate — figure prominently in the retail sector, and are most susceptible to automation.
We already see widespread adoption of automated tools for packaging objects for shipping and stocking merchandise. We even have inventory management software that automates reordering and fulfillment. Still, there are untapped opportunities that can make a retail salesman’s job easier.
Shopping assistance
Advising customers which cuts of meat or what color shoes to buy requires judgment and emotional intelligence. This can’t be replaced but there’s an opportunity for RPA to enhance both the customer and employee experience. Lowe’s introduced an in-store robot, LoweBot, to help customers with way-finding and product information. San Francisco has a completely robotic coffee shop Cafe X. Chatbots respond to customer queries before directing them to the appropriate human, or without even getting an actual person involved.
Marketing
Consumers are demanding more personalized interactions and promotions from brands. You can build out automation to tag customers with specific attributes, and automatically trigger marketing campaigns based on actions they take. For example, if someone hasn’t made a purchase in 6 months, you can create an automation to remind them that it’s time to go shopping again!
Fraud detection
Machine-learning analytics ranks the highest amongst fraud mitigation investments. RPA tools can proactively protect retailers from fraud by sending an alert or requiring further identity verification for online or phone orders deemed high-risk. You can establish high-risk orders based on order value, browser location, or shipping location.
Digital shelves
RPA tools coupled with AI technologies at scale can create digital shelves that display prices, nutrition facts, coupons, and video advertisements — all dynamically updatable from a central source. Eventually, these shelves can be linked to shoppers’ smartphones, allowing an increased level of personalization. Kroger’s retail-as-a-service product is a good initial offering in the space of hyper personalization.
Merchandising
RPA can automate historical analytics and generate predictive scenarios, significantly reducing the time needed to plan merchandise. Similarly, dynamic systems with web-scraping and predictive impact analytics could automate pricing and promotions. Automating these and other time-intensive processes will enable merchants to focus on more strategic activities, creating value for the enterprise (Exhibit 2).
👩🏻💻 Data Scientist in a Supply Chain Company
⚫⚫⚫⚪⚪
For companies with goods to move, data science and RPA are interrelated and mutually reinforcing. RPA improves data accuracy. Data science extracts value.
When it comes to supply chain analytics, data accuracy has been overlooked for years, and variability is the likely cause for bad analytics and poor satisfaction.
Let’s explore an example. Audited freight invoices often have rate tolerances in place. Immediately, the tolerance makes the freight payment data suspect because it could have a $5―$10 “cushion” built into the price.
Too many manual processes cause variability in the data. The shipper might be manually tendering the shipment to the carrier. Or, the carrier might have a manual process that they have to follow to invoice the shipper.
All of these issues cascade into the analytics platform with the data telling the wrong story.
Thanks to data science, we are getting the right and accurate data from more sources. Rather than using information to sound alarms when there are “exceptions” — problems with an individual shipment or in the supply chain―we can use it to prevent them. I see the marriage of RPA and data as a means to manage the future, not the present. 🤝🏻
Reduce dwell time
Accurate data allows RPA platforms to focus on the real exceptions and reduce your dwell time because you are working on the real shipments that didn’t get picked up. With the right data, predictive and prescriptive analytics provide real tangible insights. This will minimize processing errors by reducing human touch points, and help meet customer demands during peak seasons.
Meaningful visualizations
Visualization tools have sharpened the look of end-user RPA interfaces in recent years. But does the underlying data point you to where you are having a problem today, or report the news of where your freight was yesterday? Smart visualization and customizable analytics dashboards on top of RPA tooling should show actionable insights based on accurate data.
Seeing color-coded dots on a map is cool, but does it impact your bottom line?
The dot may show you where your late shipment is. Accurate data will show you a potential issue before it’s even late. RPA dashboards can help you eliminate the late shipments rather than show ongoing “visibility” to the problem.
Real-time insights
RPA platforms need to enable complete operational monitoring and sourcing of real-time business intelligence — without the installation and management overhead. There’s a need for intuitive, plug-and-play interfaces and a search and analytics engine that lets you play with large volumes of data quickly and easily.
Knowing which shipments weren’t picked up on-time allows adjustments to be made and ensures on-time delivery for your retail customers. Timeliness can be critical in your inbound supply chain since visibility that your hot inbound shipment is 30 minutes out can help you mobilize your warehouse staff for faster unloading and processing of a backorder. In the final mile of your supply chain, timely updates manage customer expectations.
👷🏻♀️ Contractor at a Construction Site
⚫⚫⚪⚪⚪
Activities such as operating a crane on a construction site require greater flexibility than those in a predictable environment. Naturally they are more difficult to automate today but it doesn’t hurt to dream a bit. 💭
A typical commercial construction project runs 80% over budget and 20 months behind schedule. RPA has an opportunity to address these shortcomings and complete construction projects faster, safer and cheaper.
Request for proposal production
RPA can assist with several parts of the RFP process, including creating an estimate, gathering supporting documentation, and populating at least part of the proposal.
Order equipment
RPA can automatically do this to guarantee new employees are fully prepared for work and wearing the correct gear. This is critical as workers are exposed to health and safety risks on a daily basis, and regulatory compliance is a necessity.
3D printing bots
RPA tools interfacing with these printing robots will enhance additive manufacturing and create unprecedented scaling opportunities for on-demand and custom projects. Being able to print prefabricated parts on site through an easy to use dashboard will help obviate the need to transport large materials, lower overhead costs, cut down on project completion times, and allow for more nuanced project changes. If a customer doesn’t like how a particular prefab turned out, it’s a simple tweak with a minimal loss of resources and time.
👩🏻⚕️ Nurse & Administrative Assistant at a Hospital
⚫⚫⚪⚪⚪
It’s difficult to automate activities that require direct contact with patients, be it by healthcare professionals, or registered nurses. Empathy can’t be automated. For positive healthcare experiences, patients must feel safe and supported, and that’s what human caregivers will always do best. RPA will be best used as an adjunct to medical professionals to take care of patients with lesser healthcare needs and allow human doctors to spend more time with patients that need the attention.
Moving from static websites to dynamic patient portals
To bridge the gap between the provider and the web-savvy patient, many hospitals have revamped their static websites into fully functional, two-way patient portals with dynamic user interface (UI) screens. One fundamental reason why RPA fails is when the target UI changes but the bot is leveraging the screen-scraping technologies for the old UI screens. Current technologies have failed due to prediction latency and execution lag.
Sophisticated visual perception and matching algorithms are needed for “screen scrapers” to pull data from a patient portal’s dynamic user interfaces. This capability will empower digital health workers to find the right data for transmitting prescription refills or lab reports. If RPA tools can handle screen positioning more effectively, it will also enable chronic care patients to quickly obtain answers to their questions. Further, RPA tools can be integrated with electronic medical record (EMR) software to cater to patient requests for selective access to their medical records. With the increase in EMR adoption, patient portals will be a big part of the push for enhanced continuity of care, and improving the patient-provider relationship.
Data from patient to process
One, RPA tools can be leveraged for auto-filling patient health form templates. Nursing assistants, for example, spend about two-thirds of their time collecting health information. Two, RPA tools armed with intelligent automation can aid in data collection for healthcare processes with sparse data sets.
From interpreting chest x-rays to identifying eye diseases, the domain of transfer learning and active learning have found their significance in a variety of standard medical tasks but a few pressing challenges still remain. How much of the original task has the model forgotten? Why don’t large models change as much as small models? These advanced ML techniques need to deliver production-grade performance to transfer knowledge across domains and query the data the model wants to learn from.
Bot orchestration
Recent developments in conversational AI have blurred the lines between RPA bots and conversational bots. While both approaches have traditionally been different, there’s merit in them working together and not thinking too narrowly about the opportunity — layering chatbots on top of RPA creates the potential to force a more guided, rather than a conversation-driven flow of activities. When there’s a chat element to a process, think chatbot, and if the process is highly administrative, think RPA.
🚸 Teaching & Administrative Staff at a University
⚫⚪⚪⚪⚪
Some of the hardest activities to automate today are those in education that involve managing and developing people or that apply expertise to decision making, or creative work. For these activities, characterized as knowledge work, rather than replacing humans outright, RPA will need to augment humans and change the kind of work that we do.
Reduce bias in grading
Instructors still grade most homework, and projects. Well-meaning instructors take every possible step to ensure that grades reflect merit and not the personal characteristics of the student or the arbitrary whims of the grading process. But to bias is human. Can RPA help to reduce bias in subjective assessment?
Current ML models don’t have a recursive mechanism to process new data and course correct. Such models, like the teachers themselves, are susceptible to the bias introduced in the initial training data that over time can produce unexpected results and make the decision-making system unreliable. Improving the current models, and leveraging RPA for grading will allow teachers to spend more time outside of the classroom, and less after-school time going over quizzes and homework.
Single screen design for notifications
This might seem counter-intuitive but the UX of education platforms need to be crafted to look more like a UNIX-based system rather than an Apple iOS — best systems will be simple designs devoid of attention-grabbing graphics. RPA tools to notify students and faculty about meetings, events, course enrollment, and low attendance, via a single screen over multiple screens will help in easier navigation and tracking.
Accessible e-learning platforms
Edu-tech platforms aren’t catered to users with less apparent sight deficiencies related to dyslexia or color blindness. If RPA tools can be used to gather accessibility metadata and to interpret student needs, the platform can choose the most adequate educational resources so that the student can understand its content.
These are just some of the areas in which AI can help with RPA. But that begs the question — is this still RPA or an entirely new category? This is more of a philosophical question. RPA has gargantuan applications in the enterprise market, of which the largest tranche still accrues to basic screen automation. Once you go beyond that, you’re now looking at Intelligent Process Automation.
Now, as for an analysis of RPA in specific industry verticals or of the interplay of RPA and AI? That’s another post for another time. 🧘🏻♂️
RPA has opened the gateway for augmenting white collar automation and AI — enough to generate hockey-stick growth excitement for investors. While there’s plenty of real-world complexity in implementing RPA, I hope this post gives a window into a knowledge structure for recognizing deep structural commonalities to your current problem in different ones. This is precisely the skill that sets the most adept problem solvers apart.
If you know of great entrepreneurs looking to raise their first or second round of funding, or of people on the fence about starting a company, who just want to brainstorm ideas in the space, I’d love to hear from you on LinkedIn or Twitter. ⛹🏻♂️
References: HCP Live, McKinsey, Stitch Labs, Technavio
