Back to Basics: Ops in the Age of AI
Each year, companies hand over billions of dollars to consultants in exchange for cost-saving ideas. Clients want to know which initiatives they can implement to reorganize, offshore, outsource, and automate their way to more profit. They crave benchmarks, best practices, and tools.
My former firm and I were happy to oblige. We published articles about the latest case studies. We built frameworks to turn ideas into projects. We knew that making money in client service is about giving companies what they want, not necessarily what they need.
Most of my clients were in service industries — banking, insurance, health care, and retail. I was surprised by how few had adopted basic operational practices. They lusted after new, expensive, risky ideas while ignoring old, cheap, proven ones.
A similar trend is playing out with AI. Companies are rushing to innovate their way to profitability. Most of what they’re doing won’t work. Let’s take a trip down memory lane to understand why.
Old School
I began my career at GE, where the Finance and Operations ran the place. To rise through the ranks, you had to learn how the company made money and what you could do to make more. It was a similar story when I worked in private equity.
These days, “Ops” is often looked down upon in management consulting circles. The cool consultants do strategy and digital work. Ops is what you do if you’re not clever enough to solve complex problems. The question is whether solving complex problems delivers money to the bottom line.
Most companies and consultants have lost sight of what creates value in complex organizations. They’re so obsessed with finding quick wins that they ignore the basics. Time-tested operations principles are the closest thing to the laws of physics in business. You can’t outrun math, and no amount of innovation compensates for poor execution.
I could write a book about all the operational practices service companies ignore at their peril. In this article, I’ll focus on the most obvious: bottlenecks and variability.
(Un)stuck
Companies are only as strong as their weakest link. Exceptional leaders understand this instinctually. They focus their time and attention on what my friend calls “the closest alligator to the boat.” Sometimes that’s filling a critical role on the management team. Other times, that’s dealing with an activist investor or getting a new product out the door.
Most leaders, including many CEOs, lack this discipline. They try to solve every problem simultaneously, fragmenting teams and failing to run any issues to ground. The wheels usually stay on the bus, but nobody has time to fix the flat tire.
It doesn’t have to be this way. Eliyahu Goldratt wrote The Goal more than 40 years ago. In it, he describes the Theory of Constraints (TOC), where every organization has a bottleneck (constraint) that limits performance. Goldratt offers a simple playbook for maximizing profit:
Identify the system’s constraint
Exploit the constraint (maximize its output without extra investment)
Subordinate everything else to the constraint (align processes to support it)
Elevate the constraint (add capacity or investment if needed)
Repeat the process (once one constraint is resolved, another will emerge)
The Goal — purchased by 6 million MBA students and counting
Almost every worker on a manufacturing floor understands the TOC and how to remove bottlenecks. Yet, that knowledge hasn’t permeated service businesses. Insurance underwriters don’t consider their work part of an integrated production system. Medical doctors rarely focus on reducing their inventory of frustrated patients waiting for appointments.
I recently worked with two venture capital (VC) firms seeking to deploy AI in their investment processes. Both wanted to begin with the same use case — producing investment committee (IC) materials. Associates spend thousands of hours creating models and pitch decks. I understand why VC firms think this is an opportunity.
There’s just one problem. Associate capacity isn’t usually a bottleneck. You could double the number of Associates and have little or no impact on the throughput and quality of deals. The bottleneck is typically the Managing Director (MD).
Assume you have four Associates and a single MD. Each Associate can prepare one IC deal per week. Meanwhile, the MD can only review three deals weekly. Adding an AI Associate doesn’t increase throughput, but adding an AI MD does.
Impact of adding an AI Associate (left) vs. AI MD (right)
To be clear, I’m not talking about delegating investment decisions to AI. I’m talking about using AI to automate MD rather than Associate tasks. An “AI MD” can provide feedback on early investment theses, pressure test critical assumptions, and draft diligence agendas. It could also kill bad deals early so the team doesn’t waste time on them.
Automating Associate work increases inventory at the front of the funnel. It does little to improve the throughput and quality of deals. To get the most value from AI investments, you need to focus on bottlenecks.
All Over the Place
There’s a chart every consultant loves to create. It shows a breakdown, by quartile, of team performance. The takeaway usually says something like, “You’ll save millions if you can get your bottom quartile performers up to the median.”
Look Familiar? (Source: SQM Group Inc.)
Mathematically, the consultants are correct. The challenge is that you can’t snap your fingers and close the performance gap. Even well-run companies have significant gaps between bottom and top quartiles. Think about how useless you were during the first few months of your job.
When I worked for GE, we had a massive enterprise program dedicated to stamping out variability. People look at me funny when I say I was a “Six Sigma Master Black Belt.” I’m convinced GE used those titles to make it difficult for us to leave.
At its core, Six Sigma is statistical analysis applied to business. The goal is to reduce variability as defined through the eyes of the customer. But why care so much about variability?
Averages don’t matter if you’re on the long tail
Think about your last horrible customer service experience. Did you wait too long on hold? Were you routed to the wrong agent? Was the person on the other end of the line unable to solve your problem?
Hundreds of things can go wrong during a service experience. A company can solve 99 percent of problems during the first call, but you don’t care if you’re in the 1 percent. More importantly, you don’t care if you’re in the 99 percent a dozen times in a row but slip into the 1 percent just once.
Standard deviations, not just averages, shape service experiences. And what’s the most significant source of variability? The performance gap between your best and worst employees.
Since leaving GE, I can’t remember seeing a standard deviation metric on an executive dashboard. If you want to know why service interactions are annoying, it’s often because companies aren’t measuring what you care about.
Fortunately, early research shows AI is exceptional at reducing variability in human performance. For example, Boston Consulting Group (BCG) found that high-skilled consultants using AI performed 17% better on tasks. More impressively, low-skilled consultants improved 43%, nearly matching the high-skilled group.
BCG Study Results (Source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321)
I’ve written about this phenomenon from an education perspective and won’t rehash the details here. The point is that variability matters, and AI is exceptional at reducing variability between humans. If you’re looking for AI use cases, don’t simply focus on the averages.
In Plain Sight
Business fundamentals change less from year to year than most leaders realize. If you want free advice from a former consultant, stop chasing shiny objects. If you’re unclear about how you make money and what you can do to make more, AI isn’t going to help you.
AI is cheap, skilled, cognitive labor. The price point is the main difference between AI and what you have today. If I gave you 500 bright interns, would you know what to do with them? Solve that problem, and adopting AI will be far easier.