Where AI operations management actually helps: Justin Reilly of Tucows

Most conversations about AI operations management start with what the technology can do. Justin Reilly starts with what it cannot, which is why this episode is worth the time.

Reilly is chief product officer at Tucows, the company sitting at the intersection of connectivity from domain names to Ting wireless and Ting fiber. He is also the first guest on Between Two COO's who is not a chief operating officer, which was deliberate. The angle is what operations looks like from the product seat, and where the two functions actually meet.

Before Tucows he ran product and customer experience innovation at Verizon, where the company was working to become less of a finance and operations driven telco and more of a product company. He founded several companies before that.

The road in, by way of basketball and machine learning

Reilly expected to play basketball overseas after college. His body had other plans, the career ended earlier than intended, and he graduated without the plan most of his classmates had.

He was at Penn at a useful moment. Venmo was being incubated, Warby Parker was starting, Invite Media was about to be acquired by Google, and Milo was on its way to becoming part of Half.com. He got into social ad tech early, worked at a commerce social network and then a digital products company, and consulted between exits.

Verizon came next, during its digital transformation, where he helped put the AI foundation in place.

Then he left to work on chronic illness with machine learning. His sister had been dealing with an illness for years, and the problem he wanted to solve is specific to what he calls invisible or non-linear illnesses: acute healthcare does not fix them. A physician sees one snapshot while everything else in a patient's life stays invisible. Working with Dr. David Sontag at MIT, the effort was to plot a longitudinal health journey and use machine learning to fill in the gaps, on the premise that what happened inside someone's body over 30 years is fairly deterministic and fairly predictive of the future.

Then Toronto called. Reilly's first reaction to the name was that he thought he had bought a domain from them once. What made it work was the open internet position at the core of the company's values, which he describes as standing against a lot of the prevailing trends in big tech. He moved from Brooklyn to Toronto.

He is candid about the transition cost. Healthcare and telecom rhyme more than you would expect, both full of interoperability problems and the same underlying operational challenge of getting a human to follow a new workflow, whether that human is a physician or a customer service agent. But he showed up speaking clinical product language that nobody at Tucows understood, and it took about three months to put the acronyms down.

The last 30% of the stack

Reilly's most useful frame for operators is about where the last fifteen years of digital transformation money actually went.

Picture the technology stack in thirds. The top touches the customer. The bottom is as far back office as you can get, the systems of record. Nearly all the investment has gone into the first 70%.

You can see the result in almost any large company. Customer service has a good interface, there is a new website and app, and meanwhile someone in finance or field service management is working in something close to unusable.

What surprised him is that the strongest of the next generation of product managers are moving toward that last 30% on purpose. It is not shiny work. It is a billing system that affects five internal stakeholders, all of whom have terrible interfaces and real inefficiency in how they run core parts of the business. Reilly's read is that people are finding genuine value in solving those problems, and that some companies have made the unglamorous internal layer a legitimately interesting place to work.

Bad customer experience is a symptom, not a cause

Ting is known for customer care, and Reilly's explanation of how you get there has nothing to do with the care team.

His claim is direct. Most bad customer experiences in this space trace back to bad business decisions. The phone tree you have to punch numbers through exists because call volume is manually triaging a product decision that was wrong in the first place. He points at companies carrying call center cost centers in the billions, and connects them to earlier choices like putting millions of customers on one to one pricing with thousands of SKUs, so that every interaction is complicated by construction.

The upstream fix belongs to product. Make the thing simple enough that the people who call have real problems a human should solve, or are calling because they want to, which he considers an equally good outcome for lifetime value.

He is equally skeptical of applying new interfaces where they do not belong. Chatbots got treated as a universal solution. Not everything needs a conversational interface. Logging in, seeing a bill and paying it with Apple Pay does not require a conversation. Choosing a router or a phone might benefit from back and forth and some discovery. And fairly quickly you reach the point where a person should pick up.

Underneath it he sees an industry that has made itself complicated. Strip it down and telecom is a device, provisioned service, and a bill. The complexity was added over the years through tech, bloat and vendors.

Michael's comparison is to Zappos, where product decisions removed enough routine friction that human to human contact could be an investment rather than a cost center. Reilly agrees with the synopsis and declines the credit, noting he had been there two years and the mobile business started in 2012 and the fixed business in 2015, built by a team focused on simplicity from the beginning.

His favorite illustration is a team member who was one of the most prolific contributors in the telecom device community on Reddit, answering strangers' questions about their phones because he liked the subject. Once he was in direct customer conversations, people began calling and asking for him by name, including people who were not customers. Some of them switched, on the theory that if this is who picks up the phone, the service is probably worth having.

An honest audit before any model

Asked how operators should think about AI beyond the customer facing layer, Reilly starts with an inventory rather than a technology.

Take honest stock of everything your organization does. For each thing that happens between humans and systems, ask what comes out of it and whether there is room for a tighter feedback loop. Then ask whether the improvement comes from better process and communication, or from automating the step, or simply from having better visibility into what it is doing.

His model for where the technology is reliable is bowling with the bumpers up. Models are strong when the subject is clear and bounded. A conversation about a flight, changing a seat, asking about miles, stays in the lane. Open it up to anything and it gets much harder, which is why walking up and talking to a general assistant remains a real challenge. Internal operations tend to have bumpers by nature, because you generally know what a system is supposed to do and what an operations person does day to day, and that is where the useful optimization lives.

His second point is the one he thinks will separate good operators from busy ones. All of this produces an enormous amount of data, and knowing which of it deserves attention and which is noise becomes the actual skill for COOs and chief data officers. He describes teams getting drunk on the sauce, spinning up churn models and warehouse optimizations that turn out to be confirmation bias or noise, and ending up less efficient than they would have been without any of it.

What does not automate, in his view, is the creative and human part. The nuance of inspiring a specific person to do something differently is close to impossible to hand off.

Hire well, then get out of the way

Reilly's leadership approach is deliberately plain: hire the smartest people you can and get out of their way. He describes his job as working for his team. Set the vision and be crisp about it. Give people room to run inside sensible swim lanes, leaving some ambiguity on purpose, because people bumping into each other produces healthy tension. Then remove what is blocking them. His observation is that doing the first two well means much less of the third.

He avoids elaborate progress tracking frameworks, and explains why with a mistake he has made before. A misused asset or the wrong person in a seat cannot be fixed with process. That is a talent problem wearing a process costume. Process is comparatively easy once you have strong people, and the remaining difficulty is getting them to follow it, since strong performers rarely want to color inside the lines. He would rather have that problem.

Scale changes the job. Tucows more than doubled its workforce over a couple of years and is now north of a thousand people. What Reilly treats as his most important task is knowing his people well enough to understand where they are coming from in a given interaction, so that a person leaves a hard conversation feeling accountable and supported rather than just criticized.

What they got wrong about remote

Tucows is remote first and has hired an enormous number of people who have never met in person. Reilly's honesty about the early missteps is the most useful part of this section.

His well intentioned mistake was trying to recreate water cooler conversation virtually. Scheduled coffees, informal hangouts. What it actually did was stress people out with more screen time. His summary of the feedback is that people wanted to get off the screen and go pet the dog.

So he walked it back, on the reasoning that people were probably working more than before because the commute had disappeared, and that unstructured time should be theirs. What replaced it was smaller: leaving a little space for humanity at the start or end of a call, even a tactical one.

The rest is deliberate design. Purpose build the in person moments across the year, at whatever frequency fits the organization, from annually to weekly. Learn something real about each person and let it show up naturally in conversation rather than staging a get to know you exercise.

And then the thing he thinks is underrated. Shared hard work builds trust faster than social time does. The late night call, the weekend triage, the kid who runs into frame, the joke that lands because everyone is sleep deprived. Coming out of something difficult together, with a few scars, does more for a relationship than the coffee chat.

The problem he never expected to solve

Michael's standing question is about the operational problem you never saw coming. Reilly's generalized version involves a large scale product that had cleared its regulatory hurdles and was ready to go.

The thing that nearly killed it was whether the people who would appear in the product liked the photos of themselves. The group had a collective ability to refuse, so one or two holdouts could sink the whole thing nationally.

The solution came from a colleague who walked into his office the next morning and said his son had picture day at school, so why not do picture day. Photographers shipped around the country, everyone gets a new headshot, and he had already done the math on cost. It worked, the objection evaporated, and people ended up with headshots they used elsewhere.

The lesson Reilly draws is that the blocker on a technical product was not technical at all, and no amount of product rigor would have surfaced picture day as the answer.

The 5 things I took away from this conversation

1. Bad customer experience is downstream of a decision someone made years ago. This reframes the whole cost center argument. If your call volume is high, the useful question is not how to handle calls more cheaply, it is which product or pricing decision is generating them. Ops teams inherit these problems and rarely have standing to fix the cause. Reilly's point is that they should ask anyway.

2. The unglamorous 30% of the stack is where the operating leverage is. Everyone funded the customer facing layer. The internal systems your finance, ops and field teams live in every day got skipped. That gap is now the highest return work available, and it is a genuine advantage that it still looks boring to most people.

3. Put bumpers on your AI use cases. The bowling analogy is the most useful mental model I have heard for scoping this. Bounded, well understood workflows are where models perform. Open ended ones are where demos work and production does not. Internal operations happen to be full of bounded problems, which is exactly why that is where to start.

4. Knowing which data to ignore is the actual skill. Reilly expects the differentiator for operators to be discarding noise, not generating analysis. A team can spin up a dozen models and end up slower and more confidently wrong than before. That is an uncomfortable thing to say out loud in an AI conversation and I think he is right.

5. Process cannot fix a people problem, and trying is expensive. Reilly names this as a mistake he has personally made. When someone is in the wrong seat, adding tracking and ceremony buries the symptom and burns the team's patience. Get the talent right and process becomes the easy part.

FAQ

Where do AI tools for operations management actually work today? In bounded, well understood workflows. Reilly's framing is bowling with the bumpers up. When it is clear what is being discussed and the range of outcomes is constrained, models perform well. Open ended, go anywhere interactions remain hard. Internal operations tend to be naturally bounded, which makes them a better starting point than general purpose assistants.

What is the biggest risk in AI operations management? Acting on noise. Reilly's warning is that these systems generate an enormous amount of output, and teams get attached to models that are really reflecting confirmation bias or randomness. The result can be less efficient than running the business without the modeling at all. Deciding what to ignore is the discipline.

How does product work connect to operational excellence? Directly, in Reilly's view, because most operational load in a customer facing business is created upstream by product and pricing decisions. Simplifying what the company sells reduces the volume of problems that operations has to absorb, which is why he treats simplicity as an operations strategy rather than a design preference.

What should a COO understand before partnering with a chief product officer? The customer, in specific terms. Reilly recommends the jobs to be done framing: what is the customer hiring this company to do. Once every function understands that, it informs decisions in operations, people and finance. Without it, teams argue about why something is done a certain way while missing the downstream customer impact.

How do you build trust on a remote team that has never met? Reilly's answer is to stop simulating the office and be intentional instead. Drop the mandatory virtual social time, which mostly adds screen fatigue. Leave a little human space at the edges of real calls, deliberately design a handful of in person moments across the year, and recognize that working through something hard together builds more trust than scheduled socializing does.

Also mentioned

• Tucows, Ting Internet and the Ting Towns fiber build in places larger providers skipped
• Verizon, where Reilly led product and customer experience innovation during its digital transformation
• Dr. David Sontag's clinical machine learning work at MIT CSAIL, applied to longitudinal patient histories
• Warby Parker, Venmo, Invite Media and Milo, the Penn area startup ecosystem Reilly came up in
• O-RAN Alliance and open radio standards, the shift making telecom networks behave more like software
• Jobs to be done, the framework Reilly recommends for understanding why a customer hires your company
• Zappos, Michael's reference point for treating customer contact as an investment rather than a cost

Listen to the full episode

Justin Reilly on Between Two COO's

Between Two COO's is hosted by Michael Koenig. Subscribe on Apple Podcasts, Spotify, or wherever you listen.