A collection of interviews published last week. Visit our platform for full coverage
Published Last Week
Rollins
Rollins’ stock is down ~35% YTD on fears AI search is impacting its brand equity and organic growth algorithm.
Ultimately, LLMs are recommending a wider range of pest control providers. Unlike with traditional search, these recommendations vary significantly across platforms, markets, and can even yield different results for the same search run on different days. - Jerry Gahlhoff, President and CEO, Rollins, 22 Oct 2026 call
AI search is directly impacting competitor organic lead generation:
AI engines are eating a lot of Google search volume, so optimizing your brand and your business to be present on AI search engines is more important than ever. Personally, a couple of years ago, I was not a huge believer in it. Then I started to see our organic volume from Google fall, as did a lot of other businesses, and saw traffic from AI engines pick up. That is a real thing that is happening. ... It is still smaller than Google search. I would say Google search is still four to five times larger. - Director of Growth at Orkin Competitor
LLMs may prioritize smaller players, closer to customers, even if it has less reviews than Orkin:
AI arguably takes into consideration far more than just reviews. The problem is none of us knows what it looks at and how it chooses to interpret what it is reading. It is taking sentiment. In that first iteration, sentiment was analogous to the review, but now the review is more than whether I gave it five stars or four; it is looking at the sentiment attached to the review. It is also no longer considering that I have 1,000 and you have 100 at the same 4.5. It is very likely that the 100-review company is going to show up first. It is not that it does not make any sense; that is not fair. But it is not a consistent logic that anybody has been able to figure out how to impact, and that is the unnerving thing about it. We do not control it, we do not know what it says, therefore we are not quite sure how to combat it. - VP at Orkin Competitor
Density lifts route profit but does not need national scale.
There is no doubt that the denser your route, the more profitable that route is, by and large. You have less drive time and less likelihood of an accident. Your mileage is down, and you are nicely concentrated. ... Turnover is lower. There are a lot of positives. But route density does not require great size to accomplish. ... Every 800 customers is a fully productive route. I do not need to be $1 million to have 800 customers; I do not have to grow in size. - VP at Orkin Competitor
This interview goes on to explore the risks to Orkin's growth and can be read alongside the following:
Repligen
Repligen is in an interesting position to benefit from biologics manufacturing demand over the next ten years. It has leading single-use columns and valve technology, and is a leading proteinA ligand supplier to chromatography resin providers. Repligen packs anyone's resin and sells ligands, while choosing not to make its own resin given the heavy capital and the risk of customer conflict. Management saw this as the most capital-efficient way to position itself in chromatography:
One, I think overall we still make the best prepacked columns. Two, as much as we would like to sell our own resin, we are not tied to it. If you want your resin in it, we will pack your resin. As people have gotten smarter about resins, what they want and how they use them, that flexibility has meant they do not need to buy into the old GE model. As their process knowledge has increased, they are more likely to say they do not necessarily have to buy the Cytiva ÄKTA system. - Former Vice President of Strategy at Repligen
The interview also explores Cytiva's market share loss in chromatography and how suppliers like Repligen add value to pharma companies:
I think both Sartorius and Repligen are going to start to eat into Cytiva's market share on the chromatography side...two things have happened. One is the average knowledge. When I first got into this, I always thought all the big biopharmas and big end users had a lot of smart people like we did. When we started working with them, I realized they do not know what they do not know. They think they have a lot of really good people, but they do not. They looked at us and asked how to do this. - Former Vice President of Strategy at Repligen
Brown & Brown
Brown & Brown adjusted organic revenue by ~$44m in nine months for individuals who left to join a 'startup broker': $10m in Q1 2026, $18m in Q2 2026 and $16m in Q3 2026. Revenue growth was 0.3% YoY in the last nine months, including the adjustments it was -0.8%. The company is in litigation with the competitor.
If the concern, and we have heard this swimming around out there in the marketplace, that we have lost all kinds of producers and everything else, that is absolutely factually incorrect. - Andy Watts, EVP and CFO, Brown & Brown, 27 Oct 2026 call
This interview with a Former Senior Vice President Commercial Insurance at Brown & Brown covers Florida condo brokerage and producer pay. The executive now works at a competing broker and holds Brown & Brown shares. We walk through broker compensation structures:
I was paid 40% on new business and 30% on renewals. But I developed a system where I took on a protégé, a right-hand man in a team setup. I gave him 10% in the first year, and he helped me get the quotes, line everything up and do the presentation. He got nothing on the renewal, but he got 10% on the new business plus a base salary, so it turned out well for him. I had so many leads that I couldn't close them or do quotes in time, so I was losing business, and I decided I would rather give up some of my commission. Technically, Brown & Brown was giving me 40/30, but in essence I was on 30 all year round, because I was giving away 10% of the first year to my teammate. It was so successful that the office started trying to get the other agents to do it too. - Former Senior Vice President Commercial Insurance at Brown & Brown
And Brown and Brown's company stock plans and talent retention:
For the most part, it is beneficial to Brown & Brown because of the way they structure it: you have to be there three years to vest it, and five years to own it. They have a very long leash on you, which makes it very difficult to leave. When I left, I had close to $1 million in stock, and maybe 20% of it was vested. I was not going to walk away from $800,000. ... I love the fact that they get you vested and put these targets in, and the stock grows tremendously while you are working, but it also keeps you trapped there. - Former Senior Vice President Commercial Insurance at Brown & Brown
Leavers typically try to take their book and then settle after Brown & Brown sues them.
From what I have seen of other agents leaving, the norm is that they take their book, Brown & Brown sues them, and they settle on a negotiated price that is fair. That has been the standard. This is the first time I have seen someone asked to start from scratch and honor the non-compete, and that is what I had to do. - Former Senior Vice President Commercial Insurance at Brown & Brown
This interview can be read alongside the following:
Google TPU
Alphabet called TPU system sales "a meaningful contributor" to Q3 2026 Google Cloud revenue of $30.7 billion.
On TPU systems and margins, we do not break out margins for individual products or infrastructure components... Given the price performance benefits of TPU systems relative to competitors, growing demand for accelerated compute and the scarcity of accelerated compute, we are very focused on continuing to improve margins as we scale the business. - Alphabet, 28 Oct 2026 call
Management said that TPU software work aims to "make portability seamless". This interview explores which workloads are best suited for GPUs vs TPUs:
To answer your question of whether you would use a GPU or a TPU, it comes down to a lot of different trade-offs. NVIDIA products are highly adaptable and reusable, in the sense that they have such a strong software stack that you can buy the product and use it in so many different ways. You can rapidly iterate on your software development, training and testing for your LLM or whatever your application might be. That is why we have seen this explosion of open source models. NVIDIA built the moat and everything else, then people built software on top of it, and it has all been working very well. Google, on the other hand, built something highly specialized, and they are continuing to try to make that highly specialized thing more generalizable and usable. Specialized in the sense that the software stack is very custom. Building a model takes more time and effort and is more complex than with some of the alternatives in the NVIDIA space. Once you build that model, it will be highly performant and highly optimized, but building it will take you time and usually cost you a lot in terms of human effort. - Former Technical Program Manager, AI Infrastructure at Google
NVIDIAs ecosystem remains an edge:
I saw they just announced, I think, TorchTPU a few weeks ago, which is supposed to be a translator. I think that is going to help for sure, but it is still not native. The observation I am making is that there have been more than five years of very strong development in the NVIDIA ecosystem, which is what everyone is basing things on. If you are going to switch from NVIDIA products to TPU, you need a strong reason to do that. - Former Technical Program Manager, AI Infrastructure at Google
On the other hand, a former acclerator buyer at Anthropic contradicts the size of the software barrier. Note, this is from Anthropic's perspective, not the enterprise perspective. The interviewer asked how important it is that NVIDIA is easier to configure when making buying decisions:
It is a bit important. It used to be a lot more important than it is now, because both the Neuron SDK for Amazon Trainium and the XLA compiler for Google's TPU have become much more performant. Traditionally, NVIDIA was a turnkey solution with full native PyTorch support. If a researcher had written their training workloads in PyTorch, they knew it was going to work. The same was true for an older chip versus a newer chip: the software doesn't change, because CUDA takes care of translating it into domain-specific kernels. The latest hardware gives you faster results, but the older hardware also works. This is becoming more and more prevalent on TPUs as well, so now it is not a major restriction. Especially from a frontier lab perspective, you have custom containers and custom scripts anyway, so it is not a big consideration. - Former Senior Executive, AI Accelerators & GPUs at Anthropic
It's also interesting how TPUs were only used for internal workloads which may have allowed NVIDIA to extend its lead:
I can't remember when they really started doing that. I think they tried to commercialize TPUs externally on Google Cloud back in the early 2020s, because I remember meeting the product manager who was responsible for that, but the nature of it being an internal asset and tool that was leveraged internally made external adoption much harder. It is as if NVIDIA had built the GPU and, for the first 10 years of building GPUs, decided to only let people inside NVIDIA use them. I don't want to say that is why NVIDIA got a 10-year head start on Google, but that is the nature of what happened. - Former Technical Program Manager, AI Infrastructure at Google
This can be read alongside:
© 2026 In Practise. All rights reserved. This material is for informational purposes only and should not be considered as investment advice.