Top automation pitfalls and how to avoid them

First Tech Credit Union is a San Jose-based financial institution with more than $16 billion in assets. As the eighth largest in the country, it primarily serves tech companies and their employees, but still has a lot of manual processes in place.

“We’re very early in our automation journey,” says Mike Upton, the organization’s digital and technology officer.

First Tech had been deploying some robotic process automation, trying to replace paper forms, as well as using for other automations. But these efforts all fell short.

The first problem was many of the bank’s processes cut across organizational and technological silos. Its existing point automation solutions were often unable to do the hand-offs.

For example, the process to send a domestic wire involved 105 different manual steps. “When we started mapping all that, we realized how many touch points and hand-offs there were.”

So First Tech began a new approach last summer using a low-code automation platform from Pegasystems. The vendor was selected specifically because of its cross-silo capabilities. But having the right technology in place wasn’t enough.

In some cases, even when the processes were well documented, one department might not fully understand how their workflow impacts another team, Upton says.

“The technology is very powerful, but the way people think is very challenging,” he says. “They’re comfortable with what they know. Having to re-imagine, re-engineer and re-think processes turned out to be one of our biggest challenges.”

In addition to the hand-offs, the credit union sometimes had to get everyone’s agreement on whether to automate at all.

“There were challenges getting the different business team partners to agree on where automation could be applied and where they had to have manual controls in place,” he says.

And there were many things that could’ve derailed the project that had less to do with technology and more with business processes, change management, and controls.

“Thankfully, we were able to avoid complete catastrophe,” he says. “But we’re seeing this more frequently as we take on other RPA projects.”

Eventually, the drawn-out wire process was cut to just five steps, saving hundreds of labor hours. The bank also reduced average call handling times by 40% and eliminated all data entry errors by auto-filling forms with relevant case data. That time saving now allows employees to focus on higher-value tasks, and help the credit union grow without needing to add additional staff in a tight labor market.

But First Tech is not unique. Issues like these are common to most companies embarking on automation journeys.

The who, what, and why of automation

According to a 2022 survey by Salesforce and Vanson Bourne, demand for automation by business teams has increased over the last two years, said 91% of respondents. And according to Gartner, the RPA software market grew 19.5% last year compared to 2021, and is expected to grow 17.5% in 2023. And by 2025, 70% of organizations will implement full automation in infrastructure and operations, an increase from 20% in 2021.

But automating a bad process can make things worse as it can magnify or exacerbate underlying issues, especially if humans are taken out of the loop.

In some cases, a process is automated because the technology is there, even if automation isn’t required. For example, if a process occurs very rarely, or there’s a great deal of variation in the process, then the cost of setting up the automation, teaching it to handle every use case, and training employees how to use it may be more expensive and time-consuming than the old manual approach.

And putting the entire decision into the hands of data scientists, who may be far removed from the actual work, can easily send a company down a dead end, or to end users who might not know how automation works, says James Matcher, intelligent automation leader at Ernst & Young.

That recently happened at a company he worked with, a retail store chain with locations around the US.

The retailer approached people on the front lines, and employees and managers working on the shop floors, for suggestions about manual processes that should be automated.

“They ended up with a long list of use cases along the lines of, ‘How do I upload this Excel spreadsheet,’” says Matcher.

But these were minor issues that didn’t scale across the whole operation.

“These were little tactical things that you couldn’t repeat,” he says. “So there was no definitive value coming out the back end of the exercise.”

So they spent six months going to individual stores getting ideas, wasting thousands of hours before deciding on a different approach of putting together an internal lean team, bringing in consultants, and taking a holistic, role-based approach to automation.

“We spent about four months during the persona-based mapping,” says Matcher. “That was quite a rigorous exercise to get right.” Then came two months for designing the technology, and the first use cases went into production three months later.

After all, customers have a wide range of demands and each needs different kinds of help—and different kinds of automation to serve their needs that might involve actions by different employees or different corporate systems.

Other tasks currently handled by employees could be replaced by self-service tools. For example, a customer looking to return a product could start the process on their smartphone app, eliminating the need for excessive manual data entry.

“We went through the process of matching customer personas and employee personas, and got a huge amount of optimizations,” he says.

One key factor to set up the right automations is to match them to the right business objective. For example, companies looking to automate in order to reduce headcount or labor costs might miss the main objective: to improve customer service and grow the business.

Matcher says he recently saw this happen with another client, a manufacturing company looking to reduce the number of customer service representatives with automation.

The business unit started the automation process last spring, then went back to the CFO for additional funding to continue the project in the summer because they were able to free up several thousand person-hours.

“And the CFO says, ‘I don’t see any adjustment in your headcount in the new budget,’” says Matcher. “‘Where’s all the money I spent?’”

In fact, the customer service reps used the time they saved to cross-sell and up-sell customers, and double their revenues.

“Ultimately, the gross margin level is more beneficial to the organization,” says Matcher. “But if we hadn’t shown the bridge between the two, they probably wouldn’t have continued the automations in that domain. They would have looked at the ROI and stopped the program.”

The when and where of automation

When it comes to automation, people become more important, not less. Forgetting this can be a big mistake.

“You have to put in a lot of conscious thought,” says Sanjay Srivastava, chief digital strategist at Genpact.

By automating simple, repetitive processes, enterprises still need human experts to handle complex and unusual cases, requiring upskilling. But more than that, automation can enable new business activities. For example, someone working in accounts receivable may spend less time generating routine invoices and more on solving customer problems. But they’re also in a position to recognize that a customer is spending more than usual and may be ready to buy additional products or services than they were before. That will require a different set of skills.

“The operating model has to change, and that’s a bigger question about business management,” says Srivastava. “We all know it’s easy to get software implemented, but it’s hard to get business outcomes achieved. There’s a big journey between the two and we mustn’t fool ourselves that we’ve achieved results just because we got the software.”

There are only so many productivity gains remaining to be made, he adds. But there’s unlimited growth potential in finding new business opportunities made possible through automation.

For example, companies can use automation to improve existing service offerings. “If you improve the stickiness, you improve the durable advantage and your competitive position,” says Srivastava. “And that gives you the ability to have a more sustainable business in the long run.”

Next, you can use the improved relationships with customers to expand products and services or create new ones, and to cross-sell customers.

“Then, you’ve expanded your revenues,” he says.

In addition to that, there are also other downsides to removing humans from the loop prematurely.

Many models, for instance, require human supervision and training to fine tune and improve them, says Craig Le Clair, VP and principal analyst at Forrester Research, Inc., and author of a recent report about the perils of automation.

In December, for example, Hertz agreed to pay $168 million to settle disputes related to false theft reports. Customers would return a car late, Hertz’ automated systems would report the car stolen, and the next person who rents that car would be arrested for vehicle theft.

There were more than 360 legal claims filed against Hertz by customers related to such false arrests. In one case, according to law firm Pollock Cohen LLP, a NASA employee was pulled over, surrounded by police with guns drawn, and arrested in front of co-workers.

“Here’s an area where they removed humans in the loop prematurely,” says Le Clair.

Another example of too much automation was Zillow’s plan to value homes with AI and make purchase offers they called “Zestimates.” When the offers came in too high because, say, there were undisclosed problems with leaky basements, human sellers would jump on it, and Zillow wound up with too many bad bets. When the AI erred the other way and offered payments that were too low, buyers would naturally go elsewhere to sell their house.

“If you had a crack in your foundation, the algorithm wasn’t going to pick up on that,” says Le Clair. “You certainly can solve a lot of problems by keeping humans in the loop.”

So companies can have too much confidence in data and algorithms, he says. Just look at the online chatbots without human backups.

“You don’t have an easy escalation,” he says. And when there are humans online to take over when problems arise, the systems often lose context of the conversation and the customer has to start over with the agent. “So we don’t do human well, and that’s a critical element as we move forward.”

In fields like medicine and finance, there are regulatory restrictions to automation, says John Carey, MD in the technology practice at consulting firm AArete.

“There will be a high watermark for some automation because the legal framework needs to evolve,” he says.

But even in industries without heavy regulations and compliance requirements, companies should keep an eye on ethics and standards when it comes to rolling out automation, especially when new technologies like OpenAI’s ChatGPT are making AI tools dramatically more intelligent and capable. “These smart tools are fantastic,” says Carey. “But the challenge for us is to be aware that they are double-edged swords. We have to figure out how to use and leverage them in ways that are legitimate, and build solutions that are ethical for clients and end users.”

Data Center Automation, IT Leadership