Artificial Intelligence: Seduction Vs. Reality

SVP at Kepler Group, overseeing the company’s product, technology and data services teams.

Shutterstock

All the marketing behind artificial intelligence today reminds me of the push for the cloud (never worry about infrastructure maintenance again!) and big data (kiss concerns about structuring your data goodbye!) just a couple years ago.

Sure, we all benefited from moving to the cloud and getting smarter about working with structured vs. non-structured data. But implementation was never as seamless as the tech sales team said it was going to be. It also took more than just dumping data into the cloud to realize any substantial business benefits. AI is no different than all the technology hype that came before it.

Will AI fundamentally change the way we work and live? Yes. Can your company point an algorithm at your data and just push a button and have magic come out the other end without doing any legwork? Probably not.

I want to be clear: AI has made tremendous strides over the past few years thanks to the ever-increasing amount of data and ever-decreasing cost of computational power. But it’s far from the effortless panacea being pushed for most companies by the collective technology marketing machine, and I worry sometimes that we’re all being seduced into believing we can get all this benefit without any work or critical thinking on where and how to use it.

Fundamentals of decision making and technology solution design still apply to AI as much as any other capability from the past couple years. The thoughtfulness of humans, the quality of the starting data and the state of the infrastructure are just as critical, if not more important than the AI algorithm or automation process being considered.

For example, according to Deloitte’s 2017 cognitive technologies survey (download required), the most common application for AI across businesses is process automation. This includes robotics and software processes such as transferring data automatically from email and call center systems to central data warehouses. In terms of human thoughtfulness, however, AI can’t really help you decide if a process should exist in the first place. Sometimes, the best solution for a legacy practice is to remove it completely rather than trying to automate it with technology. After all, is anyone really looking at that weekly report your team has been pulling together because of that one-off fire drill request from a senior client a year ago?

Speaking of automation, AI also can’t guarantee that legacy COBOL (ugh) reporting system you’re stuck with from that last acquisition can actually export a data feed to your data warehouse. According to Deloitte’s study, 47% of respondents felt the top challenge with emerging AI technologies was the difficulty in integration with existing processes and systems.

Another common use case for AI we’ve come across among our clients is for predictive modeling. An example would be identifying which prospects are most likely to respond to which advertisements. Running prediction algorithms is the easy part. Getting clean data that can be used with an algorithm is the difficult task.

There’s a saying among data scientists: “Ninety-five percent of the work is just getting all the data ready for the 5% of the time you get to do the modeling.” Sure, algorithms can help fill in the blanks for partially missing data, but these algorithms can’t do much when only a handful of people on your sales team are consistently logging the important details of their 9:30 a.m. offsite prospect meetings into the CRM system.

I’m sure this all makes me sound like a modern Luddite and AI skeptic. Quite the contrary. We’ve used AI to help our clients achieve some incredible results for their marketing campaigns. Last year, we built an automated, AI-powered targeting model that helped one of our Fortune 500 clients improve their cost per acquisition by 30% vs. non-targets. But this wasn’t as simple as pushing a button. We needed data scientists to figure out which models were optimal for the use case. We needed a group of business analysts to help understand how to control for sales cycles unique to the client and an entire tech team to build the infrastructure to let the algorithms run autonomously.

Like other companies, we’ve implemented countless AI-powered processes and have been exposed to a number of our clients’ AI efforts over the past year. The proverbial juice is usually worth the squeeze, but it takes a good deal of critical analysis to figure out how to get the most out of AI, and it’s rarely as easy as just pushing a button to integrate it into the business. I look forward to the day when integrating AI is truly seamless and all our major problems are solved. But until then, there’s a lot of work to be done.