Is the automation industry stagnating?

A preview of the Solitude Agent Builder

Companies and enterprises alike have been spending loads on automation over the past decade, and that spending has taken many forms:

• RPA (Robotic Process Automation)
• No Code / Low Code Software
• Hyperautomation (RPA + AI)

While this is great, where are all the gains we supposedly should have attained by way of this automation? We can point to any number of organisations that have published the savings they have made in time and possibly revenue by implementing these tools, but as an industry, we keep missing the mark when it comes to delivering on the promise splattered across every nascent automation company’s hero section: “less time doing boring stuff, more time doing work that matters”.

Despite the success on the surface, we took a deeper look on the forums where the people who actually have to build and implement the solutions for their company congregate: Reddit & Twitter. As it turns out, the developers charged with building and maintaining these solutions all harbour common complaints:

1. So called “Citizen Developers” rarely create robust and useful automations
2. Most automations created are quite fragile and deeply coupled to the underlying process that they’re built around
3. The UX for developing these automations can be quite poor, especially in the case of market leading RPA solutions

So what gives? To understand this better, lets segment these tools by the market they address.

Large enterprises often use RPAs for automation

Large enterprises have software and processes that are very difficult to automate (e.g they have no API surface), or, the existing jungle of red tape, ERP systems and BAU management functions make finding what needs to be automated extremely difficult. In to the rescue comes RPA.

RPA solutions are unique because they are built specifically for these use cases; this software segment focuses on automating mouse movements and keyboard entries. Using an RPA tool goes something like this:

First, record yourself completing a given task, warts and all:

1. Open your ERP,
2. Click around,
3. Enter some customer details,
4. Close it
5. Open Excel
6. Add some details
7. Finish.

Then, play back this recorded routine on a set schedule.

I’ve glossed over the fine details here but it slowly becomes clear why this process is so fragile.

Source: https://www.reddit.com/r/automationmaniac/comments/ymn5jr/one_man_army/

RPA works best when a process is fairly unchanging and extremely repetitive, any change to the sequence defined above would mean that you have to go in and record the entire process all over again. Since theres no API surface (with certain notable exceptions for legacy software, Oracle Developer Server I’m looking at you), this is probably the best solution for now.

That said, much of the frustration arises when the process inevitably changes. The existing work has to be reprogrammed, which is especially painful if it was originally put in place by “Citizen Developers” as they are usually not provided adequate training for building scalable automations that are robust to edge cases.

The gain in productivity is achieved in exchange for the friction of updating the automation over time. Not to mention, the cost of skilled workers or consultants assigned to the maintenance work. Should this skill and knowledge leave the organisation without capturing the information, the value of the automation is significantly diminished.

If an organisation evolves and needs to update their procedures, they will either have to replace the entire automation, or retrain new employees on the automation tooling to accommodate the required changes.

Startups benefit from low/code no code tools to move faster

In a completely different market segment, low code / no code shines where there are APIs openly available that can be used to connect various solutions. Ergo the myriad of integrations the market incumbents like Zapier and others offer.

These tools come in many different flavours, website builders, workflow automation and design-to-code tools to name a few; we’ll focus on workflow automation tools today.

At small-medium sized companies and especially startups that might not be around in the 24 months, there is a heightened urgency to remove any busy-work so that the team can focus on iterating the core product / service and achieving the holy-grail of product-market-fit. In this respect, using low code / no code tools to quickly automate business processes outside of the core business domain such as accounting, finance and and CRM flows, quickly demonstrates an ROI during the early lifecycle of the company. Over time however, these companies begin run into the same points of friction we pointed out earlier, although from a different vantage point.

1. Automations need to be updated every time the underlying process changes or structure of information changes (which in startup land, is very, very often)
2. The customer charged with developing the automation needs to learn how to use the platform, and will likely need technical expertise to integrate a user database for instance (not a task for “Citizen Developers”)
3. Currently tooling is quite inflexible (especially when working with tabular data) and so the user is limited to performing the actions that the automation platform makes available. Anything bespoke has to developed outside the platform, stood up as an API, and connected to the automation platform, putting the customer back at square one.

So for simple tasks, the existing suite of software can be a great fit, but as SMEs grow and mature, they are often motivated to roll out their own in-house solutions as they encounter more edge cases and their processes become more complex due to growing head-counts and increasingly cross-functional demands.

What’s the next evolution of automation for the enterprise?

Looking at existing solutions, we identified 3 key issues:

• Complex automations are too difficult to pull off without technical expertise
• Automations in place are un-reliable in the face of small changes in the process
• The steep learning curve hinders the the time required to see real value from the automations

The team here at Solitude have collectively experienced the pain that comes with building and maintaining these low code automations, especially when the intended audience is non-technical. We wanted to make this experience so much better for customers and really deliver on that promise of letting you engage more complex, strategic work rather than get bogged down in the minutiae of the day to day.

We asked the question: what if we could completely re-invent the way we create and engage with automations from the ground up?

We drafted a short list for what this ideal solution would look like:

• As simple as possible. As a customer, I should be able to construct an automation specific to my needs with a simple description. No faffing around with nodes, drop downs and complex diagrams or screen recordings.
• Zero hassle changes. As a customer, I shouldn’t have to worry about making changes to my process, my automations should change in response to my business, not the other way around.
• Immediate gains in productivity. Post setup, I shouldn’t need to spend weeks and months training a team or department on the platform. The onboarding process should be just as seamless on day 0 and day 10.

Considering these factors, an emerging technology known as “Agents” came out on top as the most suitable approach to finally enable a refreshingly delightful user experience.

Agent technology is a form of generative AI built on-top of LLMs (large language models) similar to GPT-4. A key distinction here is that an Agent is not an LLM, it uses LLMs combined with advanced distributed systems architectures and prompt engineering techniques to do something really novel: take actions on behalf of the user, rather than just answering questions ChatGPT style.

Amazing work by Lilian Weng if you want to learn more about agents at: https://lilianweng.github.io/posts/2023-06-23-agent/

What does this mean? As a user, you can connect your calendar to the agent and direct-message channels like whatsapp, slack and discord, and direct the agent to schedule meeting or follow-up calls with your customers whenever a it recognises positive signal from a customer interaction with your team on a chat channel or a new event is inserted into your CRM.

Better yet, you didn’t have to program any of that, or configure 100s of settings. As a customer, just connect your tools and tell the agent what do, just like you would a new hire. Of course what matters the most is that you remain in the driver’s seat for this interaction, so we’ve made it easy for you to ensure that you sign off on any action the agent proposes before it gets committed.

We have so much more we want to deep dive on, and we’ll certainly get into more details as we begin to release our weekly blog series. If what you’ve read here sounds interesting to you, we would love to have you join our waitlist to be invited to our early adopter’s program at https://solitude.ai

Sources:

1. Why RPA fails to meet expectations: https://www.advsyscon.com/blog/why-rpa-fails-robotic-process-automation
2. Automation doesn’t work if your enterprise doesn’t know what to automate: https://www.forbes.com/sites/forbestechcouncil/2022/03/16/automation-is-not-a-panacea-for-what-ails-the-enterprise/
3. Agents as the next platform shift in enterprise software: https://www.mosaicventures.com/patterns/llm-agents-the-next-platform-shift-in-b2b-software