(This column originally appeared in Forbes)
You can’t visit a tech website nowadays without the words “AI” all over the place and I’m as guilty of this as anyone else (see my headline above).
AI is the tech buzzword of 2023, even though it’s not really a word. A year or two ago it was the metaverse. Before that it was a myriad of tech terms and acronyms ranging from IoT and AR/VR to robotics, 3D Printing, autonomous vehicles, machine learning and Web3 as the tech industry desperately seeks the next iPhone or big thing from which to re-invent itself. Maybe AI is the next second coming or maybe it’s over-hyped or will be replaced by the “next big thing” in a year two.
Regardless of the hype, AI does have a lot of potential and it’s spawning countless startups and billions from investors, VC firms and software providers to realize that potential. We’re just at the beginning of this. Over the next few years, every company that has anything to do with technology will be rolling out their next generation of products that is “leveraging AI.” It’s good marketing.
Except for one problem: AI has a big Achilles heel and unfortunately no one seems to be talking about it. The problem isn’t with the technology. It’s with what the technology relies on: data.
For AI to do its job it needs to use data. For AutoGPT, the likely next generation of ChatGPT that promises to automatically perform tasks without conversation, it will have to leverage the information available in a database — or multiple databases — to perform those tasks. For Google’s Bard to create email campaigns or productivity tasks based on Gmail or Google Docs it will be reliant on the information in those sources. Salesforce’s Einstein needs good customer relationship management data to track sentiment and opportunities.
I write a lot about CRM because my company implements CRM systems. Many people in my industry are excited by all the new AI-based automations coming from the software vendors in this industry. Salesforce is already way ahead of the curve with recent announcements touting their AI offerings. Other big technology companies like Microsoft, Meta and Amazon and many CRM software providers that cater to smaller companies like Hubspot, Zoho and Pipedrive are announcing rollouts of their new features that are using AI. And I’m just naming a few. AI is creating mass giddiness across all software sectors — from CRM to accounting to HR — not only because of it’s promise but because it’s a shiny new buzzword that can be used to excite customers and — most importantly — keep them paying their monthly fees.
But, like cars that need roads, all of this software needs data to do its job. And unfortunately the data at most businesses — big and small — kind of…well…sucks. Don’t believe me? Just talk to a few of my clients and they’ll openly admit this.
Even at my largest clients, data is everywhere — spreadsheets, emails, document management storage sites, CRM systems, accounting systems, HR systems, inventory systems, order systems, even manual files. Matching orders received with invoices often spits out errors. Building email campaigns is a crapshoot. Automating things like sending out overdue notices or confirming orders received oftentimes confirms with the wrong recipients. Even doing a simple thing like creating a list of customers to send a Christmas or birthday card takes a mass investigative effort.
Did we really just wish that guy a happy birthday? Didn’t he die last year? Did we just send our big client a Christmas card? But they’re a Jewish deli! You get the point.
Every company I visit I find that their data is anywhere between “not great” to a “total mess.” Fields are incomplete and haven’t been updated in months. People who changed jobs month before are still listed as working at the former company. Inventory balances lag behind physical counts. Billings are done days after products are shipped. Monthly cash reconciliations reveal dozens of missed deposits or disbursements. Are we really going to rely on AI tools to “automate” functions when the data is this unreliable?
That’s the problem. Building the iPhone took time. But it couldn’t succeed until networks and an app infrastructure was built around it. Creating all these wonderful AI tools is pointless unless the data they’re leveraging is leveragable and yes I just made that word up.
Some of this problem can be addressed by future iterations of AI tools that don’t take just one source of data at face value and instead are built to validate the information by checking multiple places. So a person’s birthdate in a CRM system can be validated by mining an external government system, assuming access is granted. Or before sending out an overdue invoice notice that could upset a good customer who normally pays on time, a payment history at the bank as well as the accounting system can be checked. Are software developers taking these internal controls into consideration as they’re rushing out their latest AI treats to their customers? Are they building in the ability to check multiple data sources before taking action? I’m doubting that.
But ultimately they will. However, it’s going to take some time. Which is why I’m betting that many of the software pundits who are glamorizing AI are going to scratch their heads and wonder why more companies — particularly smaller companies — are resisting the magic. My clients are no dummies. They’re not going to trust automation just for the sake of it, particularly if it causes more problems than it solves.
There will be solutions to this. And my smartest clients who want to take advantage of these tools so that they can cut overhead and get things done more productively will recognize the importance of a clean and complete dataset and invest the needed resources into making it clean and complete. Hopefully they’ll be doing some of the things I suggested in a previous piece I wrote. I’m optimistic about this. But I’m also cautious that in the meantime the Achilles Heel of AI isn’t getting the attention it deserves.
