Premature Automation Leads To Failure To?Perform

Automation is one of the hot buzzwords in business today and there are many tools available to do it. The internet is full of articles telling you that you should immediately go forth and automate everything in your business.

But STOP!

Automation is indeed a huge benefit for many businesses but if you just plough into it headlong it will probably not provide the capacity increases and cost savings you imagine. Indeed you may even struggle to see a meaningful return on investment.

The reason for this is that automation is simply the latest iteration of what computers allow us to do really well — fail at scale! Back in the 1980’s when I spent many teenaged hours in my bedroom programming for fun, my father bought me a red mug one Christmas. In bold white 1970’s style LCD writing it said “To Err Is Human… To Really Foul Up It Takes A Computer” (I subsequently discovered it was a famous quote from Paul R. Ehrlich, but at the time it was just my mug). And the same is true today.

Of course our systems have become a lot smarter and usually now detect the sorts of simple data entry errors that used to lead to thousands of incorrect letters being printed and sent in the middle of Sunday night. And this makes it easy to assume we’re in a bold new future where we can and therefore should automate anything.

AI, though still far from perfect has made enormous steps since it began as a largely theoretical field in the 1960’s and with the advent of adequate computing power feats that took days 20 years ago now happen in fractions of a second (source: I studied my degree in AI and on more than one occasion AI tasks that are virtually instant now took down the campus computer network for several hours)

So it’s seductive to think that with all these improvement in processing power, AI and just the fact that humanity has spent more time screwing things up with computers, that the age of doing the business equivalent of running fast into a brick wall is behind us.

But here’s the thing…

When you look at this through a “process” lens, all of that validation is essentially syntactic. It finds things that violate the rules of system (like data entry errors, missing data etc). But it does not catch the semantic errors.

You can syntactically validate a process with a machine, and indeed nearly all automation has at least some level of syntactic validation. If you try to pass a customer record with no address to a tool like Mailchimp, it’s going to detect that. If you try to use Zapier to integrate two platforms and don’t pass the required fields, you’ll get an error and an alert. Some systems even used AI to look at the values being passed and decide if they look “reasonable”, and with training they can be really effective at preventing events from triggering on incorrect data.

But step back and look at what you’re automating. This is where the semantic errors become obvious.

Unless you just designed your manual business process then it is probably built on some assumptions about how certain steps need to be done. And what’s worse is that an awful lot of business processes in small businesses were never actually designed — they evolved from people finding solutions to problems that had occurred in the past and adjusting how things are done.

all you’ve done is to allow your business to perform a poor process, really fast

On one level that’s fine — when you’ve got human intelligence in the loop then it’s maybe not a big deal because you can (hopefully) trust them to use “common sense” to decide when extra checks are needed or not. But once you automate things, you’re likely to find you don’t get close to the return on investment that you expected because all you’ve done is to allow your business to perform a poor process, really fast.

Depending on the automation solution, you may even find you’re generating additional licence costs from the extra steps you’re undertaking, and you’ll probably be generating additional data to store and manage which actually has zero business value. And you will also have missed the opportunity to put more appropriate controls in place.

Let’s take an example…

Imagine that one time you lost ￡5,000 by forgetting to send an invoice out because after printing it fell off the desk and slid down by the wall, not to be found for months.

This had never happened before but was of high value so you now employ additional controls around each invoice being sent to make sure that it is ticked off as being printed, ticked off again when it goes in the envelope and finally that when the post is collected you tick it off as handed to the post office.

For good measure you then call the customer after 48 hours, confirm receipt and tick that off too.

You trust that if the invoice is of “low value” your staff won’t take too much time on this but for the “high value” ones they will be thorough.

Putting aside that the example of paper invoicing is a little old fashioned (but still pretty widespread), let’s think about how much this actually costs. If the business is sending out 50 invoices per week then it doesn’t need much time before the cumulative effect means it actually costs more to implement the control than the value at risk. And this happens frequently in business processes because humans are excellent at identifying and focusing on things that failed but frankly terrible at estimating probabilities so we almost always put in controls to prevent a recurrence of any mistake that is made, even when the cost of such a control is disproportionate to the risk.

Over time, all these extra controls get “baked into” the process. And often the original reason for them is lost in time.

So now we get to the reason for my slightly old fashioned example. Imagine that company is now moving to an electronic invoicing system that delivers invoices by email with a “pay now” button that allows instant payment through Stripe or GoCardless, right out of their accounting platform.

If they just go ahead and automate their existing process then they will still require their spreadsheet checklists to be filled in at each step. Maybe this can be done automatically, but the fundamental flaw here is that at a non-zero cost, they’ve just automated valueless work — because they’ve built controls for a risk that no longer exists.

Obviously this example is confected and vastly simplified compared to the reality in most businesses but it demonstrates the problem with just charging headlong into automation (and remember all those unneeded automated controls cost time and money to build and maintain)

In this example it’s (intentionally!) glaringly obvious that the right thing to do is to stop and redesign the process before automating it. The intention here should be to work out what the process achieves, not what the process does. But I would argue that you should ALWAYS stop and redesign the process before automating it.

Once you’ve done that and designed a new process, with the right controls and reporting, in the right places, then by all means — automate it as fast as you can! Proper automation of a properly designed process can make a huge difference to both the capacity in your business and the error rates, but as always the carpenter’s adage applies (no not the one about “measure twice, cut once” although I could probably make a comparison there the “6 Ps”)

If you take away one thing from this article it is this. The automation platform can’t tell you when your process is just fundamentally wrong — only a human can do that (for now!)

Design the process properly before using automation to run it fast if you want to see the return you expect. Get help if you can’t do that yourself — in the end it will be well worth it.

This article was originally published on the Cobalt Beach website https://cobaltbeach.com/premature-automation/

Mark Bowden is the founder of Cobalt Beach. You can book a free consultation with him to discuss your automation woes here