NEW INC. MAGAZINE COLUMN FROM HOWARD TULLMAN

 

If Your Customers Aren’t Brand Ambassadors, You’re Doing It Wrong

As a new business builder, you learn sometimes that it’s not just your competition that stands in your way, but your own customers and their agendas.

EXPERT OPINION BY HOWARD TULLMAN, GENERAL MANAGING PARTNER, G2T3V AND CHICAGO HIGH TECH INVESTORS @HOWARDTULLMAN1

Dec 9, 2025

 

Many years ago, when I was starting my first business helping insurance companies do a better and more accurate job of settling their vehicle loss claims, we began to acquire national brand-name clients. Initially, we would deal only with their local offices or branches, either as a pilot project or because other parts, locations and divisions of the same companies were handled by different managers or administrators. The plan was “land and expand” and we were anxious to grow. But the insurance industry is composed of a million different fiefdoms.  

When you’re an entrepreneur trying to expand your revenues, especially once you’ve demonstrated the real economic value and operational benefits of your products and service, you want to go after the biggest volume opportunities within the given organization. We started our company in Illinois working with State Farm and Allstate, but we knew from the beginning that the home run volume states were California, Texas, and Florida. New York and New Jersey also had great volumes, but they were hyper-regulated and rife with fraud problems. Even back in the 80s when we started, there were only a dozen or so giant insurers that mattered, and everyone knew who they were. State Farm and Allstate were among the top five by any measure, and their claims, volumes and customers were matters of public record. They were the biggest fish in the pond, and you always want to fish where the fish are. If fishing were easy, they’d call it catching.  

In our case, we were delivering—speeding up claims’ operations, eliminating adjuster errors and fraud, and, most importantly, saving the companies serious dollars on each and every claim. Once we started to process large claim volumes, the savings were so substantial that the insurers were actually worried about negative media attention and asked us to change the terminology on our monthly results reports from “savings” to “variances” so it wouldn’t appear to an outside reader that they were shorting their insureds and claimants by settling their claims for less than they were entitled to receive. But by every measurement, using our service was a win-win (more accurate settlements completed more quickly) and we found local supporters and sponsors in all of our customers. We were, however, in for a rude awakening. 
 
We assumed that our local champions would be interested in and excited about our plans to expand to their other offices across the country in the major markets. Expanding the financial benefits we were delivering locally to some of their largest offices would create even larger and more dramatic savings and other efficiencies for their firms. But they weren’t remotely interested in anything other than expanding within their own areas of responsibility and benefiting their own bottom lines. Their bonuses and promotions depend on the results in their own regions and on their own turf. Plus, they loved the service and attention they were getting and didn’t want that diluted by our focusing on our expansion elsewhere. And they made it very clear that going over their heads to pitch the decision makers at the corporate level would be really bad news for us.  
 
So, it was all about Decatur and forget about Dallas. Peoria was fine with them, but Pasadena was a hard pass. As a new business builder, you learn sometimes that it’s not just your competition that stands in your way, but often it’s your own customers and their own agendas as well. It’s easy to find people who will say “no” but difficult to figure out who within any given organization can say “yes.” Our champions often turned out to have cotton in their mouths when their peers from other regions called for references.
 
I encountered another somewhat less obnoxious, but no less costly, version of the problem where the left hand had no clue what the right hand was doing when we worked with a company starting in 2015 called Knowledge Hound that built systems to help large organizations manage and keep track of their own information. CPG companies in particular did countless customer surveys and focus groups over the years and literally didn’t know that some group, division, client or other partner had spent substantial sums of money on research like this and then buried the results in the bottom of someone’s drawer never to be seen again. It wasn’t deceptive; it was just that the various players didn’t understand the immense value that even older behavioral data and time-lapse results could have for ongoing and new projects and products. These businesses didn’t know how to find and employ expensive and important information within their own organizations or how to bridge the data gaps and silos that existed in their own companies.  

It’s going to be very interesting to see how well and how quickly A.I. tools are going to address and remedy this particular kind of problem. It should be one of the most appropriate and easily implemented applications, but the businesses are going to have to understand the critical need to share data and to build small language models of their own rather than getting sucked into costly attempts to boil the ocean with LLMs.
 
Smart companies don’t silo or sequester their information assets; they share them broadly for the greater good. Spreading the word – like lighting one candle from another – doesn’t diminish the first, it just doubles the illumination for all. 

NEW INC. MAGAZINE COLUMN FROM HOWARD TULLMAN

 

Time to Ask Some Tough Questions About AI 

Has there ever been as much hype about a product that has yet to prove its value for most businesses? That doesn’t mean you shouldn’t take a hard look at what AI might do for you. 

 

EXPERT OPINION BY HOWARD TULLMAN, GENERAL MANAGING PARTNER, G2T3V AND CHICAGO HIGH TECH INVESTORS @HOWARDTULLMAN1

MAR 4, 2025

You just might need AI to help you list all of the issues and concerns that surround AI today.  

One of the most persistent worries in the business community is whether any of the various large language models (LLMs) are really ready for prime time and for widespread adoption by companies looking to incorporate these new technologies into their day-to-day operations.

Even if you are willing to put aside all of the commentary about hallucinations and false references, and the circularity problems raised by these systems blindly ingesting the garbage already being generated by other AI systems and thereby diluting the value and accuracy of their own outputs, you still reach the fundamental question of which version of the “truth” your own people can rely upon. Or even choosing among competing offerings that are now creating and delivering inconsistent and conflicting results.

It’s a very tough choice for the IT department to decide which LLM, if any, to endorse and adopt at this point. While a segregated sandbox to be experimented with wouldn’t cost a bundle (apart from the overhead and personnel time), once any firm tried to incorporate these systems into their own workflow at the enterprise level and install it in hundreds of seats, you’d be talking about a few hundred thousand dollars.

I guess that if you don’t care where you end up, and you’ve got money to burn and want to tell your board that you’re doing something, any road will get you there.

A free consumer offering and a novelty accessed by millions of curious users is one thing. People will try anything for nothing, especially folks with plenty of time on their hands and nothing to lose. But this is not a sustainable solution for serious operators on either side of the equation and – as we have already seen – it’s also not a remotely profitable model for the primary providers, since they lose money on every inquiry.

Why They’re Trying to Get Everyone Hooked on AI

All the big guys are racing to create a viable AI assistant for the little people in the hopes (as has happened in the past) that adoption from the outside in (remember all the ad world creatives using Macs) will eventually dictate which larger solution a given business will adopt. If your people all love Perplexity, you don’t really want to start swimming upstream and pushing some other choice.

The civilian population is already reaching the point of confusion and fatigue because there are at least half a dozen major offerings in the market with more variations and versions coming every day. ChatGPT presently towers above the rest with more than 350 million monthly active users.

But Microsoft, Google, and DeepSeek are already reaching some reasonable levels of scale and it’s never smart to bet against fast followers when they are as deeply entrenched and well-funded as these guys are. Watching Microsoft Teams slowly eat Slack’s lunch is a good indicator of where these things often end up.

Microsoft’s decision to shut down Skype and put the functionality into the Teams package is another good indicator of the old tech rule that winners take all. Remember that Microsoft itself spent $8.5 billion in 2011 to buy Skype to replace its own mediocre video offering.

The AI Race Is Still Wide Open

No one is there yet in the AI race. The main riddle is to make the assistant contextually savvy, and surprisingly Amazon is a player in this race because of Alexa. With more than 600 million Alexa-enabled devices, the world is already comfortable asking Alexa for help. And with new tech, familiarity builds acceptance and comfort rather than contempt. It’s still a “go with what you know” world.

All the major players aspire and claim to be delivering the most accurate and comprehensive responses to carefully crafted prompts. In fact, the demand for prompt architects and prompt engineering  has exploded as it becomes clear that even the best answer is useless if you’re asking the wrong questions.

We’re also seeing a surge in new businesses aiming to deliver industry-specific AI tools like GPT-4o for Law and also startups that offer to help companies build their own small and custom models based on their own proprietary data.  The idea is to avoid the generic overkill and costs of the major LLMs. You don’t have to boil the ocean and burn big bucks every time you need some straightforward answers about your own business and customers.

One other interesting new startup, Avatar Buddy,  builds low-cost, task- and role-specific “buddies” for sales and support people, as well as experts and digital twins for educators, which provide real-time assistance and direction to folks in the field.

But all these conversations tend to return to the core issue, which is: How is a buyer supposed to evaluate and decide between these many alternative tools when even extensive, comparative tests are inconclusive or contradictory? There’s very little credible guidance so far; the players keep updating their solutions and moving the measurement goal posts.

Which means that for the foreseeable future, if you want to hold your nose and jump into the pool, you’re probably best advised to follow Yogi Berra’s classic advice: When you come to the fork in the road, take it.     

HOWARD TULLMAN JOINS LISA DENT ON WGN RADIO TO DISCUSS A.I.

 LISTEN TO THE SHOW HERE

LISTEN TO THE SHOW HERE

NEW INC. MAGAZINE COLUMN BY HOWARD TULLMAN

 

You Need to Get Real With AI and LLMs

Trying to harness all the world's knowledge to create a sales lead or a new product will only send your company down a variety of rabbit holes. Industry-specific models are emerging that will help you narrow your focus. 

 

Expert Opinion By Howard Tullman, General managing partner, G2T3V and Chicago High Tech Investors @howardtullman1

Sep 10, 2024

Everyone is talking about ChatGPT, LLMs, and AI, and they all want to know about the opportunities and risks these new tools and technologies represent to makers, markets, and, of course, mankind. Most of the conversation seems to be quite high-level and mainly strategic. You don't hear much about the practical, operational, and tactical concerns that any entrepreneur who's thinking about building a new business based on these tools should be addressing. You can spend your time building castles in the air, but, as Thoreau said, the most essential task is to put solid and sustainable foundations under them. Otherwise, you've built nothing of substance or value.

There's an enormous wave of new AI-focused startups enabled by chat-derived interfaces, but they suffer from two debilitating deficiencies.

First, they are sitting on top of too much, rather than too little, information. Even if you employ the world's best prompt engineers, they aren't miracle workers, and will soon report back that there is little likely to be gained by attempting to broadly interrogate vast and largely irrelevant stores of information. You need to fish where the fish are, rather than in the entire ocean. The fact that access to the generalized Large Language Models (LLMs) has been commercialized and simplified isn't a reason to waste your time and effort, because it won't get you to where you need to be. It's exactly the same as the old story about the man looking for his lost keys next to a streetlamp--because the light was better there. 01:49

The somewhat encouraging news is that at least a portion of the latest entrants are now starting to offer, market, and fundraise based on variations of a single theme--the successful implementation of industry-specific inquiry systems designed to interrogate one or more of the LLMs that have been built by the four or five tech major players. The idea is that they will build inquiry tools that limit and focus their tasks only to those portions of the universal datasets that relate to a given industry, and use and incorporate distinct terms and particular language.

Building these "industry-specific" overlays is actually one of the first cases of a grudging recognition of the obvious fact that asking any general LLM a detailed question about your specific business is a fool's errand, very much akin to attempting to boil the ocean. You'll get back vague, broad, and useless pronouncements (with the occasional hallucination) and not much else. A proprietary LLM built upon an underlying dataset that relates to your specific area, interest, business, or industry is the only smart approach. This will save time, money, and your technical resources, and will be far less costly and much easier to develop in-house rather than through third-party vendors.

The second major concern for many of these new players is that they don't remotely have control of their own destinies because, at best, they're mere renters of the powerful LLMs that underlie the entire industry infrastructure. These, unfortunately, can be altered, limited, withdrawn, or priced in ways that effectively destroy the operation and the value of the businesses that depend upon them.

We have seen this movie many times in the past, perhaps most recently and glaringly in the various sectors of the digital ad economy. That's where startups and even well-established companies awoke one morning to discover that Facebook or Google or Amazon or Apple had abruptly shut off their oxygen, and their vital traffic, by shifting some criteria, algorithm, or other categorization, rendering them effectively invisible on the web.     

But a far more telling and direct example of the "platform" problem is the computer gaming industry, where what began with hundreds of startups aiming to build computer games ended up a few years later completely dominated by Xbox (Microsoft), Nintendo, and PlayStation (Sony). They became the only players in the space because they built and owned the gaming platforms on which every game (regardless of who built it) needed to be licensed, with fees and royalties paid to the platform owners.

Today we're seeing virtually the exact same thing happening with LLMs. The main LLMs are controlled by the four or five usual tech suspects, who have already become gatekeepers and toll takers for user access. There's really no way to avoid or escape them--but, as noted, there's some modest consolation in the fact that, for many years to come, most businesses won't need access to such enormous and unwieldy datasets.

The bottom line is pretty clear. Every new AI startup that is dependent on, and sits upon, one of these tech giants' platforms for its operations is a tenant at best, running a business subject to the whims, competitive considerations, extortions, and other demands. These startups can be cut off in an instant. It's never smart to build your business on someone else's real estate.

New INC. Magazine column from Howard Tullman

 

The AI Hype Machine is Running on Empty.

After dumping hundreds of billions of dollars into AI startups, investors are discovering that the payoff to date has been extremely underwhelming. 

EXPERT OPINION BY HOWARD TULLMAN, GENERAL MANAGING PARTNER, G2T3V AND CHICAGO HIGH TECH INVESTORS @HOWARDTULLMAN1

JUL 23, 2024

 

In a column in January I noted that in the practical world of business, where real results matter rather than hype and bragging rights, the smart players were starting to back away from their substantial commitments and investments in generative A.I. tools and projects. Especially the guys who write the checks and keep score. Yeah, they were all still talking a good game, but fewer and fewer of them were putting their money where their mouths were.  

The main reason seems to be that the near-term prospects for seeing concrete growth and improvement in revenues as opposed to cosmetic reductions in admittedly overbuilt headcounts aren't very encouraging. In many cases, any paths to eventual bottom-line benefits weren't even apparent because the operating costs of these new large language model (LLM) engines are so high that the businesses were spending serious capital dollars to generate digital dimes - if they were lucky. Compelling, substantive use cases for these tools as opposed to novelties, chatbots and toys have been few and far between.

What has really been emerging is the fact that, in addition to AI being ridiculously costly and resource intensive, after all the manipulation of the underlying data is done, you still need to hand off the output to a human being to actually get something done. Instructions aren't the endpoint of virtually any process - whether it's manufacturing, medicine, or movement - it's real-world implementation and execution by people that ultimately gets the tasks done.   

Things might be getting done faster, but it's by no means certain that the outputs are better. And it's absolutely clear that these outputs aren't new or innovative because they're ultimately constrained by the limits of their training data to making what amounts to best guesses at what's next based on what's happened in the past. You still can't Google the future. And if you're not smart and sharp enough to ask exactly the right questions in your prompts, you get garbage for an answer. There's no big prize for even having the best answer to the wrong question.

Bumping the speed and the scope of analysis or review may create some efficiencies, but these "improvements" don't add "intelligence" until the outputs are evaluated and employed by the human end users. No one's willing to turn these systems loose until their results, findings and conclusions have been vetted and fine-tuned by humans. Hallucinations might be a more polite term than lies or fabrications by the machines, but ultimately no one is going to trust them with our lives or our livelihoods any time soon.

There's still talk about the next generation or newest black box that will work some kind of magic that not only scales but shrinks costs as well, but there's no evidence that it's anything more than a pipe dream about a new version of Moore's Law. One of the flaws in this analysis is that the underlying foundation of Moore's Law is that experience is gained in production over time, which enables exponential enhancements in the circuitry. Sadly, to date, it's clear that in the GPT world we're interested bystanders at best and, while it's fascinating to watch, we rarely learn much of anything beneficial that will let us improve the process. Nor is there evidence that simply by adding more data and more computational power, we do anything to improve or expand the outputs so that they become self-effectuating and autonomous.   

Interestingly enough, we are finally starting to see that even the shameless hucksters and promoters on Wall Street are taking a hard second look and changing their tunes from rabid generative A.I. boosterism to a far more tentative endorsement that smells more defensive than aggressive. Research reports, press conferences, and presentations from most of the leading financial firms, led by Goldman Sachs, are beginning to observe and report on the empirical evidence in the field, which suggests that they may have completely misunderstood what's happening with this latest technology. Two key things are becoming obvious and each of them is largely contrary to the speeches and spiels we've been hearing from these guys for the last two or three years.

First, there are entire industries where the ultimate impact of these kinds of tools will be largely immaterial - construction is a good example. Even Goldman Sachs suggests that only around 6% of the fieldwork in construction and extraction businesses will be automated and the productivity improvements would most likely simply be a wash for the additional costs. There may be some augmentation but even those tasks will continue to be directed and executed by onsite workers. Fast food, customer service, and transportation will be other areas where it will be very difficult - without sacrificing the quality of engagement and experience - to dramatically reduce personnel. We've already seen all of the major QSR players, including McDonalds, take steps to back away from some of their initial AI implementations.

Second, the most likely jobs to be eliminated in large numbers through substantial task automation (30%-to-50%) are NOT likely to be the low-paying positions (no collar and blue collar), which require physical labor and direct interaction with customers and co-workers. Instead, white collar and new collar (knowledge workers) positions including administrative jobs, legal work, financial analysts, marketing, and writers and editors will take the hit. Unilever in Europe is already leading the pack in this workforce pruning.)  The two critical defining characteristics of the targeted jobs will be that (1) it is very difficult in many of these cases to directly measure productivity and (2) senior managers looking for easy cuts and economies with only passing concerns for content, originality, innovative analysis and quality will happily trade out these positions for machine-created material that may well be drawn and lifted from other similarly situated creators.

When you look closely, as everyone has finally begun to do, the only conclusion you can make is that unless you're Nvidia and basically producing the picks and shovels for this industry (and largely without competitors), there's unlikely to be very much there there. And what is there will inure (as usual in tech) to the biggest of the big guys. As the saying goes, when the elephants start dancing, the grass takes a beating.

 

NEW INC. MAGAZINE COLUMN FROM HOWARD TULLMAN

 

The Future Belongs to Prompt Engineers

Although the platforms that will run AI will largely belong to big tech. There will still be plenty of opportunities for startups to live in this new world. 

EXPERT OPINION BY HOWARD TULLMAN, GENERAL MANAGING PARTNER, G2T3V AND CHICAGO HIGH TECH INVESTORS @HOWARDTULLMAN1

MAR 12, 2024

In my next life, I think I'd like to be a prompt engineer. If you don't know what that is, I'm not really surprised; most people don't.  But you should learn because in the next decade these folks are going to be among the most valuable, strategic, and in-demand employees in any company.

Prompt engineers learn how to think like a machine. Your business will need to find and hire these folks if you plan to be competitive for the same reason that most NFL football teams have more data scientists on their rosters than they do quarterbacks. Data is the oil of the digital age, and its use and proper application will drive every kind of company in the future.

If you plan to scale your business, you can be sure that you'll need to augment your team's own actions and decision making with machine-driven technologies to be able to match the speed, recall and reaction times of the competition. In the digital world, speed kills in a good way. Prompt engineers are the humans who will translate our queries and be our primary interfaces to the existing and constantly emerging massive AI knowledge systems, artificial neural networks, machine-learning environments and large language models (LLMs). These AI engines are now being built by the country's largest tech companies and rapidly deployed around the globe. As you might expect, the prime players are the usual suspects and the only ones who can afford the required investments while, of course, our government itself (unlike China's) isn't even in the game. Once again, it's gonna be a winners-take-all world.

There's also a powerful and discouraging reminder here of the early days of the computer gaming industry, where dozens of entrepreneurs thought they'd build their own game machines, interfaces, and programs. Over a relatively short period of time, these dreams were crushed by Sony, Nintendo, and Microsoft, each of whom developed a dominant platform and basically required all the other industry players to develop games that would run on top their platforms.

While stupid and greedy venture investors will invest and lose billions betting on AI nuts-and-bolts startups, it appears entirely likely that the real battle for platform dominance is largely over.  The playing field will be owned and operated by the same half dozen or so tech giants--Apple, Amazon, Meta, Alphabet, et al. -- that already own search, the desktop, our phones, email and messaging, video, and large slices of the Internet itself.

Everyone else will be remitted to running on top of these platforms, and as licensees or "partners," with the gatekeepers of these major systems. As the industry giants continue to build out the underlying infrastructure for these processing environments, the most likely and potentially profitable opportunities for entrepreneurs and new business builders will be in developing industry and market-specific tools and applications that make use of the capabilities of the LLMs and other versions of the machines, rather than trying to create new versions of the systems and machines themselves.

This is actually good news in one respect:  little guys will still have a place in the ecosystem. The successful smaller and more agile players will be more organized around supporting business operations, exploiting their industry and market knowledge, and enhancing business logistics, rather than on building costly and super-technical programs that take years to develop and are likely to be copied and overrun by extensions of the established players' offerings. In addition, the capital costs of entry for specific solution suppliers will be considerably less, as will be, at least initially, the cost of attracting and retaining scarce talent.

Amazon of old was a good example of this business versus bits distinction. Today, they call themselves an A.I.-driven tech biz, but at the beginning, Amazon was all about logistics, location, execution and speed. Their tech was okay, but the competitive edge was their aggressive leadership, industry knowledge base, powerful analytics, and people who were hungry, competitive, and scrappy workhorses rather than kids, academics and computer jocks. They quickly came to know the basic book business better than the big guys in the space and beat them at their own game. Also, in fairness, the development work, and advances that Amazon has made with Alexa in terms of voice recognition, interpretation, and conversation continuity have helped dramatically move the A.I. needle forward. That work remained largely under the radar for many years - whether by design or media inattention.

Knowing what's important to ask and how to frame the right questions of the new machines will be the most critical competitive skill set in the new A.I.-enabled economy. That's because, as often as not, it's harder to shape and design the precise question than it is to eventually find the correct answers. The skills required to do this job well are far more practical and qualitative talents rather than the more quantitative and purely technical ones we ordinarily associate with computer scientists and engineers.

Anyone who has ever asked Google a question and had the unsettling experience of being told in response that there are "about 1,250,000 search results" knows that limiting and narrowly stating your inquiry is the key to achieving any simple and useful answer. In much the same way, the current LLMs are way too much of a good thing and need to be tamed and bounded. In fact, at the moment the best way on the web to get the right answer to a question is to post the wrong answer and wait for the good Samaritans and the trolls to weigh in.

Smart prompt engineers will iteratively fashion and input the "prompts" or plain English questions, which will ask the generative machines for increasingly precise and detailed answers, solutions, directions, evaluations, statistical relationships, and other responses. This will all be based on the machines' compilations, interpretations, and discovered connections, which it will theoretically draw from literally all of the digital data and accumulated knowledge in the world. The latest advances have enabled the machines to retain the content and context of earlier inquiries and incorporate those requests into the continuing series of prompts, which has made the iterative process somewhat easier and more consistent.

What's especially attractive about becoming a prompt engineer is that anyone can learn the job and - as far as I can tell - not only don't you need an extensive technical background or an expensive education, you simply need a lot of common sense, an extensive vocabulary, the ability to constantly iterate and tighten down queries on a wide range of subjects, and a passion for problem solving, crossword puzzles, Scrabble and Wordle. Readers, writers, liberal arts grads, and kids straight out of high school are all welcome. It's about aptitude and knowledge, not college -- what you know, not where you go. It also helps to be hyper-literal and anal as well, but that's not essential. Remember that these people are trying to learn to think like the machine they're interacting with.  

And even if you (or your kids) aren't looking for a new career, and don't have access yet to the latest and greatest tools, you should still spend a few minutes experiencing these back-and-forth conversations. I recommend trying Microsoft's Copilot which is free, readily appended to its suite of Office products, and couldn't be easier to use. It invites you to "ask it anything" and it's an interesting trip down whatever rabbit hole strikes your fancy.  And more than a little addictive. You can start to build your own company-specific questions and also do a little DIY experiment to see how these systems can make sense out of, and better organize, your business's own data, historical information, and customer input about their experiences in order to provide support for and augment the performance and behaviors of your team.

Having instant access to the world's storehouse of accumulated information, literature, and knowledge at your fingertips is a very exciting and empowering feeling.  One which will give you a clear idea of why so many people are thrilled, awed, and scared by the possibilities, opportunities, and challenges of these new tools. AI is all about careful curation and filtering the flood at this point. Prompt engineers will be steering the ship and leading the way.