There's a race happening right now in almost every service industry. On one side the incumbents have deep domain expertise, years of hard-won process knowledge, and a growing fear that they're about to be automated out of existence. On the other side, well-funded AI-native upstarts are coming in hot, confident that the incumbents' complexity is just inefficiency in disguise.

Both sides are half right.

The incumbents are right that domain expertise matters. What looks like inefficiency from the outside is often decades of edge cases, regulatory nuance, and exception handling that nobody wrote down but that turns out to be essential to delivering a consistent, high-quality outcome.

The AI natives are right that a lot of service work can be dramatically automated. Entrenched processes are often decades old, developed organically without careful oversight, and rely on obsolete toolsets for purely historical reasons.

But automation is not just technology, it’s the application of technology to a process, and successful processes arise only through iteration. Everyone has to tame the edge cases to produce valuable customer outcomes; there is no shortcut, no silver bullet. All else equal, it's much easier to automate an existing process than a hypothetical one, because you're starting with the lights on. Try to automate prospectively, and you'll keep tripping over stuff you didn't know was lurking in the dark - the same edge cases and exceptions that make the service valuable and hard to do well in the first place.

Winning in this environment means blending technology with human expertise, and both the incumbents and the AI natives have a real path to get there. But the structural advantage sits with whoever already has a working process and the people running it, the subject matter experts (SMEs) who hold the process in their heads.

If you're an incumbent, the opportunity in front of you is real. The question is how to actually leverage your strengths.

One Step Back, Two Steps Forward

Most people expect automation to feel like acceleration, as though someone stepped on the gas. Same process, running faster, with machines handling the repetitive parts so the humans can focus on the interesting work. Clean, additive, not especially disruptive.

That is not what happens.

The core activity of automation is deconstruction, because machines don't meet you halfway. Think about a chef cooking the same dish for 15 years. They don’t measure anything, they just know when it’s right, by feel, by smell, by the sound of the pan. Writing down that recipe for someone else (or something else) to follow means breaking the cooking apart into steps the chef never consciously tracked. Exact temperatures, exact timings, exact ratios; even exact movements. What to do when the meat sticks or the vegetables start to char. The end result may be the same, but the experience of creating and cooking from the recipe is vastly different than the original instinctual experience.

That’s what automation does to a process. A machine needs the work to be small, clean, and well-defined. LLMs widen the field of opportunity, but they still benefit heavily from small, circumscribed units of work. Human processes are optimized for humans, and they need significant adjustment before they can be refit for machine labor. High-quality automation means taking the process apart and rebuilding it, piece by piece, while it's running, so that the work becomes machine-accessible and machine-achievable.

This means that for a stretch of time, your SMEs aren’t simply making the call anymore. They're recording it, categorizing it, explaining each decision in structured terms a model can learn from. They're labeling data, or at least providing examples, tuning a prompt, reviewing errors, defining guardrails. The judgment that took two minutes now takes five, plus documentation. The work that felt like an expert flow state starts to feel like a hassle, and on a bad day it feels like data entry.

When the automation isn't deployed yet, the throughput has dropped, and your SMEs are telling you this isn't working, the whole thing looks like it might be a mistake. If you don't know that this is what automation looks like from the inside the temptation is to reverse course, but this is precisely the wrong moment to retreat.

Hold the line, and the work pays off. The expertise that lived in the SME's head is now encoded in the model, in the decision logic, in the labeled dataset that keeps getting better with every run. The experts, freed from mechanical execution, bring their full attention to the challenging work that genuinely needs them. The business gets smarter in a way that compounds, and in a way that is very hard for anyone else to replicate.

Do that a few dozen times, and the end state looks nothing like the original. The process is now optimized for machines rather than humans, which shows up as a series of steps that are smaller, more consistent, and often counterintuitive. You couldn't have drawn it on a whiteboard at the beginning. It only reveals itself through the work.

That said, ignoring SME complaints is also the wrong move.

Subject Matter Experts: Essential and Mobile

A resilient system tends to resist change. Software doesn't care whether you change it, so the resistance has to come from somewhere else, and that somewhere is the people.

The SMEs are not wrong to find this process uncomfortable. They experience a worse version of their job, in service of something they can't yet see. What they can see is the thing they think of as their valuable skill - the actual doing of the work - starting to look obsolete.

While the discomfort is real, the grind is temporary. Once the model is trained and validated, it takes over the work the experts used to do by hand. They move to reviewing the edge cases the model flags, then to auditing a sample, then eventually to something else entirely. The process normalizes. Before long nobody thinks about that phase at all, because it just runs in the background.

The experts never get the original process back. They get elevated past it, and the job changes for good, ideally toward applying their real skill set with more leverage and creativity than they had before.

The SMEs are also not wrong to find the permanent change uncomfortable. They bear the brunt of the uncertainty, and it doesn't reliably break in their favor. Cutting costs and cutting jobs is a loud theme in the discourse, even though the companies getting this right are using automation to unleash their experts rather than discard them. When you ask people to trust an opaque process without offering a compelling picture of their future, it's no surprise that they imagine the worst.

The friction that makes an expert want to leave is the very thing that makes them worth keeping. Start by recognizing what's actually scarce: your process knowledge is hard to replicate, but it lives inside your SMEs, and your SMEs can walk. They are the one asset that is both essential and mobile, which is exactly what makes them the prize. The AI natives understand this too, and the adaptable, expert people who can navigate this transition are precisely the ones they're trying to hire away from you. Your advantage is real, but it's contingent on holding onto your people.

The SME Playbook

SMEs aren’t the entire story. Standing up significant automation also takes a capable technical org, clear metrics, committed leadership, and a host of other institutional reorganization. But without engaged SMEs, the rest of it won’t get you very far. To that end:

Protect your top SMEs. The valuable people in this transition aren’t simply the most senior or the best at the current process, they’re the ones who pair deep expertise with tolerance for the process changing under them. These are the people to build around. They’re exactly who the AI natives are also trying to identify, so they’re the hardest to keep and the costliest to lose. Protecting them is about more than compensation. It's making sure the people absorbing the brunt of the transition are recognized for their effort, informed about where it's going, and rewarded when it pays off.

Bring them along, don't just mine them. The automation process depends heavily on their input, which makes it dangerously easy to treat them as a knowledge source to be extracted: pull out what they know, feed it to the model, move on. That’s a short-term strategy; you'll get the data but hollow out the person and ultimately lose them. The alternative is to make them genuine partners in the change. Invest in them, give them opportunities to learn how to work with engineers, expose them to how the automation actually gets built, allow them a real hand in shaping it. They don’t need to become engineers, but some upskilling benefits everyone. The question every expert is quietly answering for themselves is whether they're helping build something new and exciting, or being managed out of their own job one documented decision at a time.

Anchor to the outcome, not the tasks. Experts often identify with the work itself - specific tasks, manual review, the particular way they've always done things - and those tasks are exactly what's going to change or disappear. Help them re-anchor to the things that don’t disappear: the client's outcome, and the expertise that produces it. Delivery is the ongoing work of getting the client a result they trust, and it outlives any individual task inside it. The SMEs’ judgment, their relationships, their ability to make a client feel confident and taken care of, none of that gets automated away. It gets more leverage, and as automation frees their time the attention each client receives should deepen, not thin out.

Be transparent, even when you can't promise an outcome. You're asking your people to trust a process whose end state can’t be fully described, and to spend months making their own jobs feel worse in service of something they can't see. Nobody can honestly promise them what their role looks like in two years. What you can do is be clear about what the transition involves, honest about why it matters, and explicit about what you value in the people who come through it. Trust is earned through actions, so commit to them out loud, and then follow through.

The Running Start

The AI natives aren't wrong that the work can be automated, and the best of them will get there. But they have to build the working process first, against real clients and real edge cases, before they have anything worth automating.

None of this is easy. You're going to be making hard calls, and you're doing it while running the business and fending off competitors who move fast and carry none of your history. But the incumbents who get this right won't merely survive the transition. They'll come out of it holding something the AI natives can't easily build: a proven process, a roster of experts who chose to stay, and a service that gets better and cheaper at the same time.

This race isn't won with the fanciest technology or the most expensive dev team. It’s won with the durable, compounding advantage that comes from weaving together tech, process, and people.