Special Feature
Part of a ZDNet Special Feature: AI, Automation, and Tech Jobs

Humans often to blame for automation horror stories

Don't let human error turn your automation dreams into a nightmare. Experts share automation project failures and best practices for success.

istock000015894884smallresize.jpg
© iStock/Daniel Schweinert

Machines can do everything humans can do, only better. At least that's the impression you get from reading stories like this one written my colleague Conner Forrest: Chinese factory replaces 90% of humans with robots, production soars.

Not all automation projects however, go this smoothly or result in such astounding productivity gains. In fact, many efforts to replace humans with computers are spectacular failures. Didn't the movie War Games teach us this when the W.O.P.R. (a.k.a. Joshua) almost started WWIII?

But why do some automation projects work brilliantly and others go all pear-shaped? Unfortunately, we humans are often to blame.

Humans are usually the weak link

"Issues stemming from automation are actually quite common and usually driven by human error," said Matt Goulet, VP of Sales and Marketing at Globalscape. That error can be poor planning, creating an overly complicated process, failing to involve the right stakeholders, or a host of other mistakes.

Paula Tompkins, CEO and founder of ChannelNet, told me about an automation project to install point-of-sale (POS) terminals at car dealers. The terminals, which would collect information and process transactions, were designed to streamline the sales process, collect data on customer choices, arrange and complete financing, and close sales. "The ultimate goal was to increase customer satisfaction and retention," Tompkins said.

This was no small project. The terminals were to be installed at more than 15,000 car dealers around the world. It had a budget of over $50 million (US), and it would take three years to go from the inception and planning stages to staffing and implementing a pilot. ChannelNet's role was to provide the customer and dealer facing application interface. Unfortunately, it was during the pilot that two serious problems surfaced.

The first problem was speed, or lack of it. The POS terminals were extremely slow to retrieve data when it was requested by a salesperson. The problem was caused by a poor human decision. During the project's planning phase, "the internal IT team decided to collect and transmit the corporate data via a company-owned satellite instead of the internet, which would have been faster and cheaper," Tompkins said.

The second problem also revolved around data access and was caused by another bad design choice. The system would only retrieve data if the customer's name, address and phone number were entered first. Neither the salespeople nor their customers liked this approach and refused to use the system.

It might have been possible to overcome the slow data transmission problem, but according to Tompkins the personal information requirement was a "huge dealer breaker" and ultimately killed the project. Despite having spent significant time and money on the project, the system was never fully deployed. The vehicle service part was launched in Europe (using the internet to transmit data this time), but the company scrapped plans to deploy the system in the United States.

Tompkins' example illustrates a large, multi-million dollar automation failure, but problems can also occur in small projects.

David T. Scott, CMO of Livefyre, described an automation breakdown that occurred when he was CMO at Gigya. "We were planning a big presence at SxSW," Scott told me. "We had over 2,000 people join us the previous year and wanted a bigger presence than before."

Scott and team was using a third-party marketing system to automate their customer outreach efforts. The admin running the system was new. She had trained on the system, but never practiced with it. Making the situation worse, the previous admin hadn't properly maintained the data within the system. This lack of experience and messy data combined to create a complicated worksteam prone to mistakes, which is exactly what happened.

Instead of sending an email blast to the company's active list, the new admin accidently sent thousands of emails to people on a "do not email" list. It was a simple mistake, but Gigya had promised not to contact these people again. Needless to say, there were complaints. "They called us, unsubscribed again," Scott said. "Some people even burned us on Twitter."

Best practices for automation success: Address the human factor

To keep an automation project from turning into a nightmare, IT must account for the propensity of humans to unknowingly introduce errors into the process through bad decisions, miscalculations, or just plain old mistakes.

For example, Tompkins believes the dealership POS system failed due to a series of bad decisions made during both the project's planning and execution. "The business owners turned over the execution to a large IT consulting firm and the internal IT department," she said. Unfortunately, the consultants lacked industry experience and the business owners (who did have experience) weren't "materially involved" enough in the project. "If you don't integrally involve the sales operation in defining what they are going to use," Tompkins said, "you won't get the necessary buy in."

Scope was also a problem. "It is simply impossible to launch all at once a global strategic transformation of the sales process that involves a franchised sales and service channel," Tompkins said. "It would have been better to start with a region and eventually expand to a global rollout."

Goulet recommends the following best practices:

  1. Ensure that you have the proper development and testing environments.
  2. Develop and stick to a sensible process when creating automation procedures (i.e. ask questions about what the goal is, understand the desired results, etc.).
  3. Include both infrastructure experts and other IT managers who understand the systems generating or consuming the data or programs being automated.
  4. Implement web-based visibility solutions that will monitor and track project success or failure.
  5. Provide simple, self-explanatory alerts with clear subject lines when things go wrong.

"Ultimately," says Goulet, "reducing complexity can go a long way to solve potentially catastrophic issues caused by automation errors."