A new era of automation is here
Robotic Process Automation, or RPA for short, was one of the breakout B2B software technologies of the last decade. It has grown into a large software market, currently valued at $5bn according to Forrester, and remains one of the fastest-growing software categories in enterprise.
RPA is so popular because it involves using “bots” to automate high-volume, low-complexity, repetitive tasks usually done by employees. The basic case for RPA is clear: these bots free up employees to focus on more value-add (and frankly, more interesting) tasks, and reduce overall costs for businesses. They can run 24 hours a day, avoid human errors, and handle huge volumes of tasks.
Tasks that are ideally suited for RPA include processing transactions, manipulating data, responding to queries, and communicating across systems. Back office functions quickly proved a natural fit for the technology, with banks implementing RPA to route and respond to complaint emails, and healthcare firms using RPA “bots” to help in processing and analysing millions of supplier invoices.
For a long time, RPA looked like a silver bullet for companies trying to operate around messy legacy tech set-ups. It has seen significant uptake in highly regulated areas such as financial services, the public sector, and energy — sectors that are often reluctant and/or unable to make large-scale changes to their systems. (Our portfolio company, FlowX.AI, offers another great solution to this massive and widespread issue of building on top of legacy tech stacks).
However, RPA has also received some blowback. It has been criticised as a “band-aid” to avoid painful, but necessary, digital transformation. Customers complain about the bots being brittle, and that hours need to be spent maintaining them instead of building new automations. Let’s take the example of an insurance company using RPA to reconcile claims forms. If the style of data changes, say a new line is added into a form, then the bot will not complete the task. RPA bots do not have any “intelligence” to respond to the “dead end”. Instead, an employee will have to reconfigure the task.
Fig. 01
Today incumbents and disruptors alike are building new solutions to address these issues, largely based on GenAI capabilities, and we believe this development presents a major opportunity in the space.
The opportunity: GenAI & next generation automation
The whole world is now turning to Gen AI, as we explored in our recent blog post. As a result, enterprises are beginning to see a new type of automation: “next-generation” autonomous agents.
These “next-generation” agents are AI bots that can plan out and execute on user-provided objectives using large language models (LLMs) like GPT-4. These bots complete and add new tasks, prioritising their workflow based on the results of previous tasks. They can call on long-term and short-term memory, using old queries for context and storing previous results. This means that instead of stopping at a “dead end” caused by new data, these bots can “learn” from mistakes and adjust their series of tasks.
Fig. 02
Open source communities are now building their own agents for everything from writing code (GPT-Engineer) to ordering pizzas (HyperWrite). And to us, these agents look like a direct challenge to traditional RPA: why bother programming rules into a bot when you can just specify a goal to an agent?
A new generation of automation
Here at Dawn, we are excited by the next generation of disruptors in process automation. These players often have the advantage of being designed as AI-first from the beginning (even GenAI-first), and are more agile at integrating the latest innovations.
Some complement a RPA-centred automation process, for example, with better document processing or code-native platforms. Others are centred on autonomous agents to deliver “next-gen process automation” — often starting out in non-regulated use cases such as eCommerce.
Notable European companies that we’ve seen already disrupting this space include:
- Robocorp: Code-native & open source RPA platform, with a vision to combine Python’s flexibility with the ease of low-code through GenAI code interpreters
- DeepOpinion: Enterprise workflow automation with cutting-edge NLP
- Automaited: Highly flexible automation for any task, with self-learning capabilities
- Workfellow: Next-generation process excellence through plugging into a wide array of workflows
- Levity: AI text and documentation processing for emails, surveys, customer support and beyond
- Invofox: AI data entry with an initial focus on finances like bills and invoices
- Go Autonomous: eCommerce-focussed automations in areas like quotations and sales orders
- Virtuoso: QA and testing automations to reduce maintenance overhead, using NLP and AI
- Workist: Order processing automation with AI for B2B transactions
Fig. 03
The incumbent response: Incumbents & “intelligent process automation”
For obvious reasons, GenAI is also front of mind for RPA incumbents building out their platforms. UiPath and Automation Anywhere are responding to shortcomings in traditional RPA in two ways: they are trying to own more of the automation workflow end-to-end and they are adding GenAI-enabled features.
End-to-end automation workflow:
Process mining platforms such as Celonis, and traditional business process management players including SS&C and IBM, are building out their process automation capabilities. The potential value in owning all of automation is so large that it has also attracted new arrivals.
“Intelligent Process Automation” aims to make the bots more resilient by cleaning its inputs and monitoring its outputs. RPA still remains the core of this process. It starts with better mining and mapping the repeatable processes in organisations that make up the next set of use cases for automation, competing against incumbents like Celonis. It then includes better ingesting data with computer vision, optical character recognition (OCR), and handwriting recognition (HWR). Data can then be processed with clearer analytics and “next action” required. After process automation, this workflow includes a suite of testing and monitoring tools around making sure automation continues to work as planned.
Meanwhile, UiPath and Automation Anywhere have added OpenAI chatbots and connectors to leading LLM providers. Celonis released an “Intelligence API” for its data to be used more easily in automation models.
Fig. 04
Customers are excited by these new developments, but have some reservations to adopt GenAI-enabled solutions, especially in regulated contexts. They expect guardrails on the range of possible outcomes and the quality of outputs. Plus, they want to be able to monitor progress, and occasionally involve a human in the decision-making loop.
This ever-expanding automation stack [see slides above] can also sometimes become unwieldy for enterprises to manage. This is even more likely when coordinating multiple siloed automation platforms and processes that involve many handoffs between human and RPA bots. It can increase the total cost of ownership and reduce overall capacity to build new automations. C TWO, our portfolio company, looks to address exactly this issue: its orchestration platform allows developers to manage and scale their automation infrastructure through a single vendor-neutral module.
We still believe there is so much more to come in automation. We have done on-the-ground market research with enterprise customers, and they all reported still being in the early stages of their journeys to automating workflows. We believe that as much as 70% of the potential dollar value from automation use cases remains untouched, and we expect the next generation to help unlock this value.
Category-defining automation companies are being built right now in Europe, and we would love to hear from you. So if you’re building in this space, please get in touch anytime on shamillah@dawncapital.com and owen@dawncapital.com.