VP Strategy & Operations @ Prescient AI | Helping omnichannel brands to optimize spend | Co-founder Don't VLOOKUP

I didn’t come up through marketing.

I came up through finance.

Investing, M&A, diligence across hundreds of omnichannel brands.

I then became an operator, managing and deploying budgets directly and making the pitch to finance for more. 

The same questions came up every time:

Where should the next dollar actually go? How can we scale profitably? 

Marketing was always the largest variable line-item that went chronically underutilized. 

ROAS looked great in decks.

Cash flow and lifetime value told a different story.

The real issue wasn’t growth.

It was efficiency of working dollars - marketing, inventory, and capital tied up in the wrong bets.

So teams default to decisions without a more forward-thinking lens: 

• Optimize to what could be tracked with clicks

• Undervalue upper-funnel channels

• Cut spend that *looked* inefficient

• Then wonder why demand fell apart weeks later

That gap is why I joined Prescient.

Most tools explain yesterday yet ignore the reality that conditions on the field are constantly changing.

We focus on how dollars actually move the business - across channels, over time, and beyond last-click.

Not perfect attribution.

Better decisions that break down the barriers between marketing and finance to invest with confidence.

Will Holtz

I am an Associate Professor of Marketing at the University of Maryland's Robert H. Smith School of Business. I popularized customer-based corporate valuation, a "bottom-up" approach to valuing firms by assessing the value of their customers. My other areas of research include data fusion, data privacy, missing data problems, machine learning, and causal inference.

My research has been accepted and published in top-tier academic journals such as the Journal of Marketing Research, Marketing Science, Journal of Marketing, Journal of the American Statistical Association: Theory and Methods, and Annals of Applied Statistics. My work has also been featured in media outlets such as the Harvard Business Review, Wall Street Journal, FT, Fortune, The Economist, Inc, USA Today, Barron's, CBS, CNBC, Slate, Business Insider, and CFO Magazine. I have won numerous awards, including the Don Lehmann Award, Robert J. Lavidge Award, MSI Alden G. Clayton Award, ISMS Award, Shankar-Spiegel Award, and Gary Lilien ISMS-MSI-EMAC Practice Prize. I was also a finalist for many others, including the Paul Green, John A. Howard/AMA, JM Hunt/Maynard, and MSI H. Paul Root Awards.

In 2015, I co-founded a predictive analytics company, Zodiac, where I served as the Chief Statistician. Zodiac was acquired by Nike in March 2018. I subsequently co-founded Theta to commercialize my work on customer-based corporate valuation.

Before joining the University of Maryland, I was an Assistant Professor of Marketing at Emory University's Goizueta School of Business. I hold a Ph.D. in Statistics from The Wharton School of the University of Pennsylvania. I also earned a BSc in Economics with a concentration in Statistics and Finance from The Wharton School and a BAS in Systems Science Engineering from the School of Engineering and Applied Science, as part of the Jerome Fisher Program in Management and Technology at the University of Pennsylvania.

Daniel McCarthy

Claude has less than 0.5% of the AI assistant market. Gemini has about 12.5%.

But Claude generates 40x more revenue per user than Gemini. So despite having 25x fewer users, Claude actually brings in 61% more total mobile revenue. Holy moly!

How is that possible and what does it tell us about the future of AI competition?

In a new paper with Maxime Cohen, Eddy Hage-Youssef, and Daniel Sokol, we used worldwide mobile app data (iOS and Android, May 2023 through December 2025) to map the competitive dynamics of AI assistants. The findings challenge a lot of conventional wisdom.

The biggest surprise: the "AI war" framing appears to be wrong. We tracked 15 major model launches across platforms, and every one coincided with total market growth. When OpenAI launches, Claude grows. When Anthropic launches, ChatGPT grows. Competitive launches expand the category. ChatGPT lost market share but over that same period of time, it grew tremendously. A rising tide lifts all boats, and we've seen a veritable flood of people into this category.

All three have striking different strategic positions in the market:
--ChatGPT: Scale play, with ~79% market share, first-mover brand, and low-to-moderate monetization
--Gemini: Ecosystem play, with ~12.5% share, and almost zero direct revenue. Google treats it as a loss leader to protect its advertising business
--Claude: Differentiated, premium niche with ~0.5% share but 3x and 40x higher monetization per user than ChatGPT and Gemini respectively. Where ChatGPT and Gemini go wide across modalities (text, image, video), Claude goes deep on text-based tasks like coding. Small but intensely loyal user base.

This framework also helps explain the Anthropic-Pentagon standoff that's dominating the news this week. Anthropic's willingness to walk away from a military contract makes more strategic sense when you understand their market position. Their differentiation, which includes a more cautious orientation towards safety, is very much a part of their brand identity. Diluting it by accepting terms that conflict with their brand could cost them far more than $200M. It is more expected that OpenAI would step in given its 'serve the whole market' position. 

The question is whether this coexistence lasts. When growth slows, these strategies face very different vulnerabilities, which we discuss in the piece.

Full paper linked below. And short discussion slides for class (and other) discussion. Curious what folks think!

Associate Professor of Marketing at the Robert H. Smith School of Business, University of Maryland, College Park

I was recently interviewed by Ben Sherry at Inc. Magazine about how I use ChatGPT's Deep Research for market research, and I wanted to share a few tips that have made a big difference for me when using DR in general:

1. Talk at it! ChatGPT has the best speech-to-text I've found anywhere. Just walk around your office for five minutes and talk through what you need. Start, stop, start again. It's the only LLM that lets you do that (which to me is mind boggling). It's so much easier and less tiring than type type typing away (also contributes to your step count, which is a plus)

2. Ask it what you should be asking. One of my favorite moves is to say: "given everything I just told you, and my goal, what do you think I should even be asking you, including things that are important that I might not have mentioned?" It's great for catching stuff that you might have missed.

3. Ask for sources it couldn't find. DR will tell you where it pulled from, but ask it what it wanted to find but couldn't. Then you can run a second DR pass targeting those gaps.

4. Do a first pass with thinking mode. Run your results through the thinking model to get a quick first pass at "the answer". If the thinking model catches something you didn't think of, you can now feed that back in and update your original DR prompt (!), which you couldn't do before. Then you don't have to wait the 30 minutes or whatever and kick off your next 30 minute run then.

5. Ask it to explain things like you're an idiot. This is my favorite one. I'm not the sharpest tool in the shed, so I always ask it to talk to me like I'm an idiot. It's amazing how much clarity you get when you strip away the fancy words and inside baseball and just ask for the simple version.

Hope some of these help. The tool keeps getting better and it's worth experimenting with. Best out there for this sort of work. Enjoy!

Open article: Inc. | Tech and Business News for Entrepreneurs and Leaders by inc.com

Inc. | Tech and Business News for Entrepreneurs and Leaders

So every year in my customer lifetime valuation courses at University of Maryland - Robert H. Smith School of Business I'll have games where the winning student(s) pick the genre of my next song. It has run the gamut: rap, rock, reggae, and one year, even dark cabaret (apparently it's a thing!).

So this year one of my undergraduate Interdisciplinary Business Honors (IBH) Program  section picked ... heavy metal. I hadn't made a song for Blue Apron yet. 

One thing led to another and voila, "Burn Rate": https://lnkd.in/e3xxijkA

Banger! 😄 

Those of you who have taught with the Blue Apron case might get a laugh out of your students with this at the end as they leave the room. Captures the narrative arc pretty well I think. And some inside baseball about latent class models, patterns of misprediction, …

Additional resources:
--Blue Apron lecture slides: https://lnkd.in/e3iVNBiH
--Discussion of what happened since, and a peek at Wonder who bought them - good thought starter: https://lnkd.in/ecytdk4Z

More resources below. 

Enjoy! Don't ignore that burn rate - when you listen to this song, other companies might come to mind. Timeless themes here.

Ellie Kyung Carl Mela Walter W. Zhang Peter Fader Bruce Hardie Bryan Bollinger Eric Schwartz

Interesting article from The Information on OpenAI's financials.

A few months ago, I wrote about OpenAI's compute margin for paying subscribers being around 68% (link below). That number is real, but it only covers paying users (i.e., it's the margin for paying subscribers).

This article reports the *company-wide* adjusted gross margin at 33%. Why so much lower? Because it includes all the inference costs from serving hundreds of millions of FREE users. Those users generate zero subscription revenue but still burn compute every time they send a message.

Personally, I'd think about costs differently. The free tier is an acquisition engine. Those costs are really CAC. And the new ad-supported free tier should help offset some of that. Once someone converts to paid, the margin goes to the 68% number. 

So the real question is: what's driving the weakness? This was really interesting to peel back the onion on. The paid margin went from 45% to 60% from 2024 to 2025. So that all else equal would INCREASE the overall margin, but it fell from 40% to 33%. The "free user drag" on margins increased by 22 percentage points, from 5% to 27% of revenue (they don't have a well-defined "free margin" because free users generate no revenue). Net is -7pp, which matches the decline from 40% to 33%.

And when you decompose that 22pp increase in "free drag", it's almost entirely driven by the cost per free user going from roughly $0.70/year to $4.00/year, a ~6x increase. The mix shift (free:paid ratio going from 18:1 to 25:1 over this time period) is secondary.

The intuitive reason for this is in 2024, free users mostly got GPT-3.5. In 2025, they got GPT-4o, image generation, and much richer features. OpenAI made the free product dramatically more expensive to run, in addition to being so good that it brought in a ton of additional (more expensive) free users. 

So the real story seems to be less "growth in free is too fast vs growth in paid" but instead "free users are getting a much more expensive product." That's a deliberate choice to invest in the acquisition engine, and it's "working" (WAU tripled), but it's expensive and they really need to get more of these free users to convert to paid, or decrease the effective cost of the free plan through ads.

This business has the revenue growth profile of a software company but the cost structure of something much more capital intensive. Like a pharma company, they have to keep investing heavily in the next generation of models. But UNLIKE a pharma company, there's no patent protection. They have to keep duking it out with Anthropic and Google every single day.

Net-net, not the best financial update for OpenAI. Margins revised down, breakeven year pushed out. Worth keeping in perspective.

I made a spreadsheet showing the underlying margin analysis. You can find it here:

Open article: OpenAI Boosts Revenue Forecasts, Predicts $111 Billion More Cash Burn Through 2030 by theinformation.com

OpenAI Boosts Revenue Forecasts, Predicts $111 Billion More Cash Burn Through 2030

Marketing Professor focused on digital customer behavior, analytics, and AI

The University of Maryland's Marketing Department is hiring tenure-track faculty at the Assistant Professor (with some experience) or Associate Professor level.  The Smith School of Business is a great home for researchers interested in studying the impact of AI in business.  Great colleagues and fantastic location.  Please reach out if you're interested in learning more.

Open article: Open Rank (Assistant or Associate Professor) by umd.wd1.myworkdayjobs.com

Open Rank (Assistant or Associate Professor)

taplio

The Battle

Will Holtz

Daniel McCarthy

Followers

Posts

Engagements

Likes

Performance ScoreDaniel McCarthy's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Will Holtz

Daniel McCarthy

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Daniel McCarthy's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreDaniel McCarthy's LinkedIn Performance