To get AI value, shift left (and eliminate jobs)

Every telecom executive I talk to has artificial intelligence (AI) initiatives running. They’re investing millions, hiring AI teams, launching pilots left and right. Yet somehow, these pilots aren’t making it to production. The promised value isn’t materializing. Results aren’t hitting the bottom line. For what’s supposedly the biggest business transformation since ecommerce, what gives?

The answer: You’re too scared to let AI actually DO what it’s designed to do: replace human work.

Marc Benioff is also too scared to say the truth out loud.

I know, I know. Eliminating jobs is something no one wants to talk about, not even Mr. Salesforce himself, Marc Benioff. But for this blog, put on your Mr. Spock hat (or ears!) for a moment. We’re going to be objective and really focus on the business results of AI. We’re going to ignore the obvious emotional side of job elimination. If you can handle that concept for a moment, let’s be logical and talk about how to actually extract value from AI.  🖖

To get value from AI, you need to shift left

Forget the feel-good “human-AI collaboration” nonsense you’re hearing at conferences and telling your workers. The organizations winning with AI aren’t playing nice; they’re being brutally honest with themselves—and with their people—about what needs to change.

Let’s use the most widely accepted AI use case as our example: customer support. Everyone agrees AI can handle customer inquiries, but most telcos are implementing it the wrong way, by augmenting humans with AI instead of replacing humans with AI (yes, I said it).

Here’s the methodology that actually works to extract real value from AI:

1. Do the math.
2. Face reality and tell the truth.
3. Keep humans in the loop (temporarily).
4. Shift AI left; shift humans up.
5. Replicate success.

Let’s walk through the steps.

Step 1: Do the math

The reason AI has started in customer support is because it’s an easy, obvious place to target. Tickets, problems, and solutions are already documented in databases ready to be turned into retrieval-augmented generation (RAG) systems. But the other reason is that support teams are FULL of humans. Which means any improvement in productivity almost instantly turns into a linear reduction in headcount and cost.

Let’s crunch the numbers: A large telco typically has 17,500-26,600 customer support agents. At $42,000 base salary each (with loaded cost closer to $60,000 with benefits, training, and management overhead), you’re looking at $1-1.6 billion annually just for Tier 1 support. Add in facilities, technology, and management layers, and you’re easily north of $1.5-2 billion per year.

AI can handle 70-80% of these interactions for a fraction of the cost.

Step 2: Face reality and tell the truth

Once you realize the impact of AI on customer support, reality will hit you hard: “Whoa, we’re going to eliminate a lot of jobs.” And this is where you have to fight the instinct to hide from (or calling it what it is: lie to) your employees about what’s about to go down.

Start with the truth, and tell them how jobs will be eliminated but you’re going to support them through this transition by teaching them how to use AI so they take new skills to their next job. It’s super important to be honest with them. (I could write a whole other blog about how you need to talk about how it’s not just about the company’s gain but also how it’s valuable to them.) Do this step well, and not only will your transformation work, your people will appreciate you being straight with them as opposed to firing them on a Friday afternoon like a coward.

I promise you, they won’t quit en masse. You know why? Because they already know AI is coming. They’re seeing the chatbots get deployed, watching the pilot programs, dealing with increasingly sophisticated automation. What they want is leadership that’s straight with them about the plan and is actually invested in their future.

Step 3: Keep humans in the loop (temporarily!)

Now comes the execution phase: implement AI, put it into production, and have humans in the loop to watch it perform. Remember, your goal is to build an AI system that is autonomous, not a knowledgebase to help support agents answer issues faster.

Resist the urge to build a “human-assisted AI” system—one that requires humans to check quality—or a system that humans get answers from only so they can continue to interact with customers. Build an AI system that humans will temporarily monitor and improve, but that will eventually act on its own.

As AI gets better, everyone from leadership to frontline agents sees the results and knows it’s working. It will require continuous monitoring and adjustment by an AI engineer—just one; not hundreds of people. Newer models will come out, new use cases will emerge, and customer behavior will evolve. Your AI engineers are there to catch errors, flag improvements, and train the system, but always with the clear understanding that their role is to keep AI acting on its own, continuously improving. This step is all about fast iteration and improvement. 

Remember, you’re building a customer support factory, not a better knowledgebase for humans.

Step 4: Shift AI left; shift humans up

At this point your AI now handles password resets, billing inquiries, service status checks, and plan changes without any human intervention. Once you hit 90–95% correctness on solving a particular problem, it’s time to remove the humans—and shift left. The work moves from expensive human agents to cost-effective AI systems that operate 24/7 with consistent quality and instant response times.

And what do you do with your humans? Some former tier 1 agents who understand both customer needs and AI capabilities become your AI training specialists and customer experience strategists. Your tier 2 technicians who used to deal with escalated issues are now focused on the most complex technical problems, enterprise customer support, and service innovation. These humans should continue to come up with ideas for shifting more and more work to the left.

And as an awesome side benefit, there will be talent that absolutely embraces and LOVES AI. Promote these people to other areas so they can help implement these concepts throughout the organization. They are your future AI engineers. 

Separately, this is also the point where it’s time to start exiting employees to realize those savings we identified in step 1.

Step 5: Replicate success

Now you know how to get value from AI. What other functions and groups can replicate this pattern and achieve similar results? Network operations monitoring alarms and outages. Billing operations processing routine disputes and adjustments. Collections issues for non-paying customers. Renewal notice automation. Each of these areas has the same characteristics that make customer service perfect for AI: repetitive processes, structured data, and human-intensive operations ripe for automation.

Go get the value!

How we did it at Totogi and Skyvera

How do I know it works? I’ve already been through the process at Totogi and Skyvera. We followed our own methodology in our own customer support group. Two years ago, we had employees managing 100% of our support tickets per week. Our goal: getting AI to solve 100% of our tickets.

1. We did the math: AI solving 100% of our tickets would make a significant impact to our ~100 person headcount; most people would not be needed if we were successful. At an average cost of about $50,000 per person, that would be significant savings. We knew we wouldn’t eliminate every position, but the majority of our team would likely leave our organization.

2. We faced reality and told our people: We explained that our goal was to have AI handle 100% of our incoming tickets and that we’d train them to work with AI systems. No one quit. They helped us build it.

3. We put humans in the loop temporarily: We started by building our AI agents and putting them into production on real tickets. We had human agents monitoring the AI’s answers and giving feedback to our AI engineers on what was missing and what could be better, as well as providing new ideas. And it worked. Every week, our AI improved and successfully solved more and more tickets.

4. We shifted AI left and humans up: Once AI hit 90%+ accuracy on routine tickets, we removed human oversight. Our people moved to complex problem-solving and AI system optimization. And yes, we eliminated about 65% of our customer support headcount in the process. Not everyone left the company; some moved to finance and others moved to operations where they used their new experience to implement similar solutions for those functions. They knew AI would work, knew how to do it, and helped light a fire across the organization to implement AI everywhere.

The results: Today, our system resolves 78% of tickets with no human intervention. We’ve been working toward this goal now for two years, and every month we keep improving what AI can do. 

The methodology works. We proved it.

Time to lead

Getting real value from AI isn’t about finding the perfect use case and it’s definitely not about protecting every existing job. It’s about systematically identifying where AI can replace human work, being honest about the transformation with your teams, and reinvesting your human capital in higher-value activities.

The telcos that figure this out will have dramatically lower operational costs, consistently better customer satisfaction scores, and human resources focused on strategic work that creates competitive advantages. The operators that don’t will find themselves competing on price while their AI-enabled competitors deliver superior service at structural cost advantages they literally cannot match. It’s time to decide which kind of telco you want to be.

Recent Posts

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.