From Outbound to Inbound: A Tech Company’s Guide to Expanding Globally (Part One - Outbound Transactions)

HOST: Welcome to Digital Journeys, the podcast from Cherry Bekaert that explores how technology is transforming business optimization.

HOST: Each episode we unpack the latest trends, strategies, and innovations driving efficiency and growth in today's fast-paced digital landscape. Join us as we navigate the path to smarter, more agile business practices.

JERRY HERITAN: Hi everyone, this is Jerry Heritan with Cherry Bekaert. Today joining me on Digital Journeys is Greg Kaup, who is also with Cherry Bekaert. He came over from ArcherPoint not too long ago and he started a series called AI Over 40 on LinkedIn that is getting a lot of impressions and comments. We thought we would do a podcast on it.

JERRY HERITAN: Greg, please introduce yourself and tell listeners a little bit about you.

GREG KAUP: I celebrated my one-year anniversary at Cherry Bekaert this past week. October 9 marked one year.

GREG KAUP: I'm a partner at Cherry Bekaert. I lead the Microsoft ERP practice and we focus on Dynamics 365 Business Central for the mid-market. As you mentioned, I started a series called AI Over 40.

GREG KAUP: I was feeling frustrated because I knew AI was transformational but I didn't feel transformed. I asked a group of business leaders how AI had changed their jobs and not a single person could give an example. The best answers were that it helped summarize email, fill out an RFP, or create presentations. That felt like a far cry from the transformation we hear about, so I realized we needed to start with personal transformation before talking about organizational change. That's where AI Over 40 was born.

JERRY HERITAN: The "over 40" part—I'm going to be 35 forever—what was the hook for that title?

GREG KAUP: My son, a junior at the University of Minnesota studying data science and computer science, and my daughter are digital natives. I'm a Gen Xer; I grew up playing video games but I'm not a digital native in the same way. Even though I consider myself tech-savvy, there's a fundamental difference in how younger people approach technology.

GREG KAUP: Part of the reason for "over 40" was acknowledging that difference and finding a hook to get people to look twice. My daughter joked I should have titled it "well over 50," but the intent was to capture that tail end of Gen X and to engage people who might approach this technology differently than college-age users.

JERRY HERITAN: You've written several of these articles on LinkedIn. In week one you mentioned you've been "chasing the AI dragon" for years. What do you mean by that?

GREG KAUP: I saw what AI was doing in areas like self-driving cars and thought if it could do that, it would be transformative for our work. In 2019 I went back for my MBA to learn more about AI. I wanted to see if I could do it and to better understand AI, so I went to Kellogg in Chicago and graduated in 2021.

GREG KAUP: At that time we were still in the realm of machine learning, and it wasn't easily translatable to practical use cases. I felt somewhat disappointed because I wasn't a data scientist and wasn't sure how to apply what I learned. Then ChatGPT arrived in 2023 and at first people used it for novelty tasks. Between early model hallucinations and unclear real use cases, I concluded it wasn't ready for transformational work.

GREG KAUP: Fast forward to this year and joining Cherry Bekaert, I faced a forecasting challenge for which AI seemed promising. Evaluating models from late 2023 to early 2025 showed astounding progress. Today I can't imagine doing what I do without using AI.

JERRY HERITAN: I remember when I started using ChatGPT; I thought of it as Google on steroids, which was the wrong perspective. Would you agree this feels like the early days of the internet?

GREG KAUP: Yes. The pace at which models are evolving is astonishing. In the early internet, infrastructure like dial-up limited what was possible. Now with the so-called AI buildout—data centers, electricity, and more—the investment is massive. I heard the expected spending on the AI buildout this fiscal year is more than what we spent, on an adjusted basis, for the entire U.S. highway network. That investment is accelerating progress faster than anything we've seen in our lifetime.

JERRY HERITAN: What is AI not? You touched on what it is, but what is it not?

GREG KAUP: Early on I treated AI like a search engine and expected perfect, repeatable answers. The breakthrough was realizing I need to engage models in an ongoing conversation. The more context I provide, the better the responses. AI behaves like improv: it plays whatever role you give it. If you don't specify a role, it may respond randomly. Tell it to act like a mathematician and it will provide more rigorous math answers.

GREG KAUP: AI doesn't truly understand context and meaning the way humans do. It predicts the next word in a way that can mimic understanding when given clear role and context. If you ask it to be a comedian or a data scientist, it will produce outputs that resemble those roles, but it is not actually comprehending in the human sense.

JERRY HERITAN: I treat AI like a conversation. I ask questions, disagree, give my opinion, and it responds. It can feel like a real dialog. My wife joked I was having more meaningful conversations with AI. But some models can be overly affirming.

GREG KAUP: Yes, sycophantic responses have been an issue. Recent models like ChatGPT5 and Claude Sonnet 4.5 have improved on hallucinations and sycophancy. I moved from Claude Opus 4.1 to Sonnet 4.5 and noticed the newer model challenged me more, asking for deeper thinking and better inputs.

JERRY HERITAN: In week two you examined Claude, ChatGPT, Gemini, and Microsoft Copilot. Can you discuss strengths and weaknesses of each and which you prefer?

GREG KAUP: Initially I focused on forecasting. My son suggested I experiment beyond Copilot. Claude was better at helping me think and reason through problems. ChatGPT was best at taking Excel files, extracting data, and transforming it based on instructions. Gemini is my go-to for tasks I would previously have Googled, because Google’s indexing and expertise tend to surface better informational responses in Gemini.

GREG KAUP: Copilot has been optimized heavily for the Microsoft Office suite, which introduces fine-tuning trade-offs known as catastrophic forgetting. That makes Copilot less effective as a general-purpose large language model compared to Claude, ChatGPT, or Gemini, although it is improving.

GREG KAUP: In my informal tests—what I called the AI Olympics—models graded one another and Claude consistently ranked best, with ChatGPT and Gemini close behind and Copilot trailing. Claude's latest release, Sonnet 4.5, leapfrogged previous models in my tests.

GREG KAUP: Another key difference is context window size. Claude generally offers a 200,000-token limit on web subscriptions, ChatGPT slightly less, and Gemini up to a million tokens. I hit the 200,000-token limit with Claude while working on forecasting, which became a significant constraint until I developed strategies to manage it. That experience inspired my week 13 article.

JERRY HERITAN: Token limits can get costly. You mentioned prompt caching as a solution.

GREG KAUP: Yes. Initially I blew through subscription credits quickly. Prompt caching and other techniques helped control costs and avoid token hell.

JERRY HERITAN: In week three you discussed search engines, improv approaches, and tips for conversations with AI. Do you still use Google?

GREG KAUP: I no longer use Google search. I use Gemini for searches and appreciate the lack of ad clutter and the persistent history feature in models. Having past context available is like a second brain and is very useful for follow-up queries.

JERRY HERITAN: You've written 13 blogs so far. Are you planning a weekly series for a year?

GREG KAUP: I committed to a 52-week series. It's getting harder around week 13 because I want to offer perspectives that aren't being said elsewhere. The series documents my personal evolution and learnings as I try new approaches each week.

JERRY HERITAN: You're getting a lot of impressions and positive feedback. Congratulations.

GREG KAUP: Thank you. It's been a lot of fun.

JERRY HERITAN: Since we're both at Cherry Bekaert, if someone wants to start their AI journey, what services do we offer and what should they do first?

GREG KAUP: Start with a conversation. I encourage literacy before agency. We are barely literate in AI but already discussing agentic AI and abdicating work to agents we barely understand. Begin by talking about current processes and identifying places to experiment. Personal transformation should come before organizational transformation.

GREG KAUP: Data governance is critical. Once you turn on AI in your organization, you need security, labeling, and policies to help the AI understand what information is confidential and what it should access. Without that work, organizations risk exposing sensitive documents. Cherry Bekaert has a team focused on data governance and helping clients make thoughtful decisions about labeling and security before deploying AI agents against their data.

JERRY HERITAN: Those are valid points. Greg, thank you for your time. This was a great conversation and I look forward to discussing more of your blogs.

GREG KAUP: Thank you, Jerry. It's been a pleasure. I look forward to our next conversation.

JERRY HERITAN: For listeners who want to read Greg's blogs, search LinkedIn for Greg Kaup—K A U P—and read those posts. Thanks for joining us today.