Problems that prevent routine white-collar tasks from being automated as a service

Recently I had a serial-entrepreneur acquaintance send me a form email for a new company he’d started with six others that aims to automate all routine tasks that white-collar workers have:

I have a quick question: When you hear the phrase: “Annoying business processes”, what comes to mind for you?

My company automates these processes. We can handle anything (anything!) that can be turned into a set of instructions. Even if the instructions contain complex logic — we’re totally custom. Even if the labor required would normally take hundreds of hours — it’s the same price.

I responded:

I’ve had this same idea. To be helpful, just in case these are things you haven’t encountered yet, some tricky things I’m aware of are:

1) dealing with edge cases. It’s very hard to get a customer to be able to produce all of the edge cases up-front.
2) dealing with secret/proprietary information. It can be hard to convince a company to delegate / outsource something that requires that the employee know some kind of secret (often customer-related) information.
3) documenting a process in the depth necessary to delegate it can take such a large up-front cost that people may not be willing to pay it. Video recordings are a good way to get around this.
4) the employees whose processes are the easiest to automate are often the most resistant to having them automated, because it’ll either threaten their job security or it’ll require them to learn some new, often more-difficult process, often at the same pay
5) interfacing with intranets / private software / email can be extremely difficult.

Good luck, I think it’s a really good idea!

His response:

Yup, you pretty much nailed it with all of these.  Extremely difficult up front, but they all make this extremely defensible after we solve it. And we do conclude, ultimately, that these are all solve-able. Would you disagree?

My response:

I think it’s probably solvable for some tasks and not others, and so the trick is to figure out what kinds of tasks it’s solvable for and how to efficiently identify which companies have such tasks.

I think #5 may be the hardest problem to solve. The only way to get around it that I could think of is to set up the work environment on a cloud machine and then use autohotkey-like software, but that’s extremely fragile and would need to be updated every time the UI of a needed intranet / web app changes.

Another problem I just remembered is that because not everything can be easily automated, and because non programmers generally can’t tell easily which tasks can and can’t be automated, you can’t just rely on asking an employee or manager which tasks to automate. You actually need to end up watching everything they do like a consultant and pick out the parts that can be automated.

To be honest I think it’s a great idea but too hard, and so what you see successful companies doing instead is finding a single expensive process that many companies have and automating it. That’s what Parker Conrad did at zenefits (dealing with employee healthcare) and that’s what he’s doing at his new company (employee onboarding and offboarding).

Oh yeah and another problem I just remembered is needing to interface with information contained in written English (emails, intranet comments) to automate the task.

His response:

Again, I’m impressed with your analysis. Very good. Companies certainly are specializing on discrete problems, as opposed to the general problem.  That creates a coordination problem: startups struggle to use all of the services available to them.

Am CCing two other partners on the team so they can read your analysis. These 6 problems you’ve identified are all ones we’ve identified too. We’ve already got some clarity on each one,  but it is good to bring them into relief again and reconsider our answers

My response:

Another competitor to think about is Zapier. They seem to be going after the segment of the problem for which every piece of software needed to complete the task is online and can be communicated with via an API. So there may be an opportunity to go after tasks that can’t be totally automated via APIs.

(Just an idea:) Maybe you could explore making it possible to have an API for online or offline software that currently doesn’t have an API: you’d have a cloud machine that would install the software if necessary (if it wasn’t an online service), log in if necessary, and then scrape the page for whatever information was necessary, and provide that information in API form. You could sell it to Zapier, and they could charge the customer a premium to interface with those services. You could also sell it to software developers who want to build on top of those services but for whom creating their own fake API would be too expensive. Your job would be to watch the non-API software and update the scraper whenever the website / offline software changes. Again, just an idea.

Update 2024.01.20 – I’ve just found out about Appian, a service that makes this kind of automation possible. Their approach is to not automate your processes themselves, but rather give you the tools you would need to automate it, which bypasses the issues of data security and accessing intranets:

If you found this interesting, you can follow me on Twitter.