The Most Valuable A.I. Use Cases as Reported by Searchers, CEOs and Investors

---

Listen to This Blog Post

---

If you’re anything like me, you might be wondering whether you’re under-utilizing the rapidly emerging power and everyday usefulness of Artificial Intelligence. You probably don’t need to look terribly far to hear about how quickly the technology has emerged and how powerful it has become, but I often find this type of commentary to be overly generic. I appreciate that the technology is powerful and is getting increasingly so seemingly by the day, but what I really want to know about are the specific use cases that have proven to be most impactful for my peers.

For this reason, several weeks ago, I commissioned a survey that I sent to thousands of Searchers, CEOs, and investors across the SMB ecosystem. This was an anonymous, one-question survey that asked each respondent for their single highest-impact use case for A.I. thus far. The goal was for these use cases to be as tactical and as practical as possible, such that interested listeners could begin experimenting with similar use cases on that very same day. 

In today’s episode I present the results of that survey, organized into three categories of use cases: Searching, Operating, and Investing.

Before we dive in, let me start with a few general observations based on my review of the raw data:

• First, you might not be surprised to hear that there was a lot of overlap in many of the responses that we received. Not only did we combine and remove duplicates from the raw data, but we also synthesized and summarized what were clearly recurring “themes” across each of our three categories. For this reason, we begin each category by first discussing that primary theme, and then presenting other use cases that were of particular interest
• Second, this isn’t intended to be an exhaustive list that encompasses all of the responses that we received. Instead, I’ve presented the use cases that were either most frequently reported, most creative, or just most interesting to me
• Third: I suspect you’ll agree with me that the most interesting and creative use cases came from the operators. While there were a large number of use cases provided by active Searchers and Investors, I didn’t find many of them to be terribly surprising. Useful and impactful for sure, but just unlikely to come as a surprise for most of you.
• Fourth (and this caveat is likely to become increasingly important over the months and years ahead): I did not use A.I., in any capacity, in putting this list together.

Let’s begin:

Searching for a Business to Acquire

As mentioned above, Searchers provided the most “fragmented” list of answers, insofar as there was a rather long list of rather short responses.

That said, as was the case in all three of our categories, a clear theme emerged, and in this case, that theme was using A.I. to quickly triage investment opportunities to more quickly ascertain which deals deserved further attention & analysis, and which opportunities could wait.

What follows are some example submissions to illustrate what I mean (direct quotes are italicized here and throughout the rest of this blog post):

• “I perform quick, web-based research of a batch of companies from a master list against a predefined set of criteria (e.g., CAPEX profile, revenue quality, industry dynamics, ownership) to determine whether a company should be prioritized for outreach or not”
  
  
• “I give A.I. the CIM, the tax returns and [other documents from the data room], and ask it to pull every key figure (eg: revenue, EBITDA, owner compensation, add-backs, etc.) into a table with three columns: the value, which document it came from, and the page number. I tell it to leave a cell blank if it can’t find the figure rather than estimate it. That turns hours of transcription into minutes, and the page references let you spot-check the data quickly.”
  
  
• “I have A.I. take a first pass at reading, interpreting, and summarizing the deal materials – CIM, financials not in the CIM, industry and company information, and etc.  To make this truly useful when considering a number of different transactions, and to train the AI over time, you need to create a comprehensive prompt template, preferably including a financial data worksheet and a form that the AI uses for output. Otherwise I find that the output is inconsistent in content, structure and focus.”

One example that I found to be particularly interesting was somebody who used Gemini Notebook (previously called Notebook LM) to quickly summarize incoming opportunities in the form of a short, digestible, 10-minute podcast, where the “hosts” focus on the parameters that the user has deemed to be especially relevant. While the substance of this use case is ultimately similar to what I just described above, I did find this particular form factor to be quite novel.

There were several other one-off use cases that were mentioned with a high degree of frequency, including the following:

1. Performing a deep SWOT analysis on potential targets
2. Note taking for meetings & calls (Granola seemed to be the tool that was mentioned with the most frequency here)
3. Quickly reviewing long & dense legal documents for key terms or specific clauses of interest
4. Industry analysis, both basic and detailed
5. Gathering specific data on prospects, often to personalize cold outreach, or to prepare me for an initial call with an owner. One specific response that I found to be particularly interesting said:
   
   “I developed a tool that drafts personalized messages based on reading all available materials on the company’s owner, listening to all podcasts that they’ve been on, reading all news and online reviews about the company, and then connecting any relevant details to my personal life as a tool to create a possible point of commonality or emotional connection between the two of us”
   
   
6. Running a basic competitive analysis. Again, I found one comment to be particularly interesting here:
   
   “If I like a business/industry, but that particular target isn’t for sale (or it isn’t for sale at the right price), then I use AI to find a list of other companies that serve similar customers, often in similar geographical regions. As a result, my knowledge in that particular vertical – both of the industry itself and the participants within it – tends to compound quite quickly”
   
   
7. New industry idea generation against predefined criteria
8. General industry benchmarks for valuation, and various financial and operational KPIs

Before we move onto the most interesting use cases of A.I. as self-reported by the CEOs, I’d be remiss if I didn’t interject to provide Searchers with a note of caution on where they should not be using A.I., based on some recent personal experiences.

Specifically, in my opinion, you should not use A.I. to prepare your primary investor-facing fundraising document (often called a Confidential Information Memorandum, or a “CIM” for short). To be clear: I’m not suggesting that you shouldn’t utilize A.I. as a supporting tool in preparing the document, but I am warning you against the idea of largely outsourcing the production of your CIM to A.I. I’ve seen this done more times than I care to count recently, and here’s what I’ve learned from those experiences:

1. It’s usually pretty obvious. It may not be 6 months from now or even 6 weeks from now, but for the time being, it’s usually pretty apparent
   
   
2. If you over-rely on A.I. to prepare the document, there is likely to be data contained within that you don’t have a solid grasp of yourself. If and when investors ask you about that data, it’s not a particularly good look for you to not know how or why it was included in the deck, or what the primary takeaways ought to be
   
   
3. Putting this document together manually – the old fashioned way – is one of the best available tools to learn about the business and the industry yourself. If you sweat over every sentence, every chart, and every figure, you are providing yourself with the best possible education about the company and the industry prior to assuming the CEO role.
   
   
4. In addition to that, producing the document manually will also position you to handle investor FAQs far more efficiently and effectively, as there likely won’t be anything contained within that you haven’t personally touched. Remember, when investors are evaluating whether or not to invest in any given deal, we’re also evaluating the person (or the people, as the case may be) behind the transaction, who will be most responsible for its ultimate success. One of the tools within that evaluation is the Q+A process, where the prospective buyer is able to demonstrate her grasp of the business and its prospects. If you can’t answer a meaningful percentage of the questions because you’re not entirely sure how or why certain data was presented, it doesn’t tend to be terribly confidence inspiring for your investors.
   
   
5. It can be quite error prone. Again, this may not be true forever, but for now I’ve seen A.I. produce many errors that likely wouldn’t have been present had the document been put together manually. I do find A.I. to be a great tool for checking for errors in a draft version of a document, but it can be considerably less precise when putting the document together in the first place.

Operating a Small Business

Among CEOs, the number one self-reported theme was as follows:

Using A.I. to connect data from disparate internal sources (think CRM, accounting system, project management system, inventory management system, etc.) into a single dashboard, updated in real time, to give a comprehensive view of the business (which is something that would otherwise necessitate long & complex integrations, if those integrations were possible at all). Some illustrative comments within this theme included the following:

• “Automating dashboards and trackers from multiple pieces of software that don’t have native APIs to integrate. Previously it was a lot of work downloading data, scrubbing & preparing that data, and then creating something easy to interpret”
  
  
• “We combine customer interviews, meeting transcripts, competitive research, sales information and internal documents, then use AI to identify recurring themes, contradictions, missed opportunities and potential risks. The value isn’t having AI make the decision. It is dramatically reducing the time required to get from a large amount of information to the handful of issues that deserve human attention”
  
  
• “We built out an agent that lives inside of Microsoft Teams and allows all parts of our organization to access a centralized knowledge base and see live data from our ERP, CRM, and other internal software tools. The use cases can look as varied as: (1) our sales rep checking if we have enough inventory for a certain project he’s working on; (2) a customer service rep pulling up a technical specification for a specific product; (3) a shipping clerk asking for the ETA of a specific shipment. What’s been amazing is that it lives within Microsoft Teams, so people from across the organization are able to use it very quickly and intuitively. They are able to skip the usual scavenger hunt of finding data across different internal software tools or even calling other people within the company. They have one place to ask, and then it produces an answer reliably.”

Beyond this central theme, 8 additional use cases were provided by the CEOs that I found to be particularly interesting:

(1) Compliance

“We use LLMs to conduct document reviews (in our case, behavioral health treatment plans) for contract compliance and errors/omissions before submitting to our funding sources (healthcare insurers) so that all submissions include 100% of required criteria with no administrative errors.”

A small personal interjection here: Based on my own experience on the Board of a Behavioral Health company, I can personally attest to how detailed and time consuming it is not only for these reports to be created in the first place, but for them to be created in a fully compliant way, such that your payors don’t have a reason to withhold payment on account of simple documentation errors. If you’re operating in an industry subject to very specific documentation requirements in order to get paid on time, this strikes me as a very useful place to start utilizing A.I. within your organization.

(2) Hiring

We received several examples related to hiring, including the following three:

“We no longer conduct phone screens for hourly high turnover positions anymore. AI services handle 300-500 phone screens per month, allowing our talent acquisition team to stay lean and focus solely on in-person 2nd round interviews”

“When we post a job description on a public platform like LinkedIn or Indeed, we are often overwhelmed with the volume of incoming applications. Often, in a batch of 300 applications, maybe 4-5 tend to be actionable. Prior to A.I., we would have to manually sift through all of this volume, which took days. After having outsourced this initial screening to A.I., we pare down our list of actionable candidates in seconds to minutes”

“(Presumably builds off of recorded transcripts of job interviews) A.I. compares qualitative assessments of potential new hires against our established job scorecard criteria. AI is very effective at surfacing insights regularly & easily disregarded by an interviewer.”

(3) Quoting

The quoting use case was summarized so well in a single submission that I’ll just present it to you more or less verbatim:

“The problem: Previously, we quoted every job in QuickBooks Desktop. Pricing lived in spreadsheets and in the heads of a few veteran sales reps. Building a quote for a complex project (featuring multiple products and line items) meant hunting down rates, calculating labor and freight by hand, and cross-checking against past jobs. It was slow, inconsistent, and error-prone, and a new hire needed years of experience before they could quote confidently.

Using Claude Code as our core development tool, we built a tool called Sales Hub, a custom sales application on React, Microsoft Dataverse, and Azure. At its center is a Quote Assistant that has our entire pricing logic built in: configurations, flooring, lighting, labor, freight, site conditions, and the dependencies between them. The sales rep describes the job in plain English, and the system builds an accurate, complete quote. This tool integrates with HubSpot (CRM), QuickBooks Online (accounting & finance), and our broader software stack, so a quote flows into the rest of the business automatically and without re-entry.

As a result:

• Speed: You can build a quote in 15 minutes that would have taken half a day in the old system
• Accuracy: Pricing logic is applied consistently on every quote, eliminating missed line items and manual math errors.
• Polished output: Clean, professional quotes and invoices generated instantly, with every revision tracked so we know exactly what changed and when.
• Knowledge transfer: Twenty years of quoting expertise is now embedded in the tool. A brand-new sales associate can build a quote with the accuracy of a veteran rep, which shortens ramp time dramatically.
• Sales velocity: This is now limited by how fast we can talk to customers, not how fast we can price jobs.
• Reduction of key person risk: Institutional knowledge is no longer concentrated in a few people, which makes the business more scalable as we grow toward our ultimate targets”

(4) Billing

Although billing is closely related to the quoting use case that we just discussed, it is indeed a separate use case, as demonstrated by the following submission:

“I run a small commercial HVAC company and our ERP occasionally let billing items fall through the cracks. My old COO used to catch them by hand. Now a scheduled job scans the ERP twice a week and builds a short digest of anything that looks unbilled, underbilled, or stalled. I review the digest myself in about five minutes and send the follow ups to staff by hand. It has caught real dollars we would have missed, and keeping me as the reviewer avoids the noise problem: my team never sees AI output directly, they only see a short note from me. If you have an ERP with an exportable ledger, you can stand this up in an afternoon.”

(5) Implementing an Internal Operating System

If you’ve read this blog for any meaningful amount of time, you’re likely familiar with the fact that I am a big fan of internal “operating systems” for small businesses, with EOS being the most widely used of the bunch, though not the only one (for those who are unaware, EOS is based on the book “Traction” by Gino Wickman. Links to previous episodes/blog posts are provided here and here)

When a CEO implements such an operating system within her business, she is faced with the choice of implementing the system herself, or hiring an external “implementer”. When I added EOS to my own business, I chose to utilize the services of an external implementer, and generally recommend doing the same for others. However, these can be expensive projects, and footing the bill can be difficult, especially for smaller companies. This is why I found the following response to be so interesting:

“We created an EOS implementer agent. Many small businesses find implementing EOS to be very impactful, but statistics show that implementing without help typically doesn’t lead to success, and hiring an implementer runs $40k+ a year, which is a big ask for smaller businesses. I created an agent using all of the EOS/Traction materials that operates as my implementer. I have individual skills for each meeting type (L-10 meetings, quarterly off-site meetings, etc). This agent works as my implementer and so far has been very effective. While this may not be as effective as hiring a professional implementer, if the choice is between no implementer and an AI implementer, I would definitely recommend the AI implementer.”

(6) Repaying “Technical Debt”

Anybody who has ever run a software company can tell you (likely with some residue of exhaustion) about “technical debt”, a concept that essentially describes the accumulated impact (measured in time, cost, and required rework) of product and engineering decisions that prioritize speed over quality and scalability.

Sometimes technical debt is created when engineers are asked to “build it now” instead of “build it right”, and at other times it’s created when code is built on tools that are old, outdated, or are no longer supported. Based on the following response, it seems as if A.I. is rapidly shrinking the amount of time & cost associated with paying back this debt, which is a potentially massive source of uplift for existing software CEOs. The submission read:

“We upgraded our code base from older versions of angular and java to the latest versions. If done manually, this would have taken us 9-12 months but with the help of AI, we’ve been able to do it only in ~3 months”

(7) Build Versus Buy Decision for Internal Software Tools:

If the “Repaying Technical Debt” use case that we just discussed is a possible tailwind for software CEOs, then the next comment is a possible headwind, as it discusses how a CEO has built custom software using A.I. to handle highly specific internal requirements that just wouldn’t or couldn’t be offered by an off-the-shelf solution. This CEO said:

“Rapid development of high-quality custom software tools (internally and for our clients) that precisely match the company’s requirements (unique workflows, data model, integrations; handling nuances and edge cases that used to require time-consuming workarounds, etc) that no SaaS tool can provide. AI-powered development has accelerated development times from months to days, completely changing the “build vs. buy” math in many situations.”

(8) Proposal Writing and RFP Responses

“Much of our new work is sourced through bidding on RFPs. These documents can be incredibly long, cumbersome, and time consuming to fill out. While certain sections of certain RFPs are indeed unique, a large percentage of sections across all RFPs are either very similar or totally identical. By adding every RFP response we’ve ever submitted into a single database, we can now upload new RFPs into that same system, and often ~80% of the document gets pre-populated for us in a matter of seconds. This frees our team up to focus their time on the ~20% of the RFP that is unique, complicated, or particularly important”

Investing in Small Businesses

The name that I’ve ascribed to the recurring theme as reported by investors is the “A.I. Investment Committee”. Investors haven’t outsourced their decision making to A.I., but they are using A.I. as a tool to improve their decision making by challenging assumptions, identifying blind spots, asking thoughtful questions, comparing current deals to past ones,  and scoring the extent to which any given investment opportunity meets the investment parameters that they have deemed to be important. In effect, A.I. is playing the role that an Investment Committee might play in a typical Private Equity firm.

The following response is a word-for-word prompt that one investor uses within a Custom Project that they have created within Claude (ChatGPT offers similar functionality by way of a custom GPT):

“I am an investor in Search Funds. Going forward, when I have a new investment decision that I need to make, I plan to upload relevant deal documents into Claude (CIM, quality of earnings, a Q+A transcript between myself and the entrepreneur in question, etc.), and I want Claude to help me think through whether or not I should make the investment (and why) based on the data that it has been trained on.

Some of the items that I will upload include the following:

• All Stanford Search Fund studies going back 15+ years
• All relevant surveys and presentations that have evaluated which deals have succeeded and which ones have been less successful
• Initial investment memos for all of our existing portfolio companies, as well as quarterly updates from those same portfolio companies, so you can begin to tease out trends and themes on what tends to be predictive of success
• All quarterly updates that I have written for my own LPs
• Transcripts from relevant podcasts that I respect and enjoy
• My own personal evaluation checklist for new investment opportunities
• My own personal reflections on existing portfolio investments, including the primary drivers behind over- or under-performance

Of course, no investment opportunity is going to meet ALL criteria that I’ve listed as being ideal. That simply isn’t realistic, and every deal will have its risks.

It’s easy to fool oneself, so I want this project to ensure that the decisions that I make are consistent with what I have deemed as being important, my historical experience from prior deals, my lessons and experiences as an investor over the past X years, and the various frameworks, lessons, and reflections that I have articulated in the past.

When I upload documents and ask you to help me evaluate whether or not I should invest in any given transaction, I want you to start with the “so what” first, before you get into any of the supporting details: Should I make the investment or not? And (in 1-3 sentences or less), why or why not? I really value a simple and easy-to-articulate investment thesis, so if your answer is “yes, you should invest”, I want you to provide me with a simple, intuitive, easy-to-articulate investment thesis that a randomly selected 18 year-old could easily understand. From there, you can provide me with supporting details/explanations.

I have uploaded my full investment Evaluation Checklist, which contains my scored history of every deal that I’ve ever evaluated. When I ask you to evaluate a new deal, score it using this exact method and tell me where its weighted total ranks against the benchmark of the X deals that I’ve evaluated in the past. Remember that this scorecard is necessary but not sufficient: a high score is not an automatic yes. I have historically passed on top-ranked deals, and if a deal I clearly want scores in the bottom half, flag that plainly rather than adjusting scores toward the answer you think I’m hoping for.

Also note the following:

1) I appreciate concise responses and brevity (but without sacrificing meaning). Unnecessary fluff is not helpful or necessary

2) When I ask you to evaluate a deal, I will always upload a CIM. You should know that, sometimes, CIMs tend to emphasize the merits and undersell the risks of any given deal. This isn’t to say that the creator of the document has been disingenuous or underhanded, just that not everything should necessarily be taken at face value. As an investor, my job is to be skeptical, though not unduly so“

I think what I took from that response is not necessarily the specific substance of the prompt itself, but how detailed and structured a good prompt should be in order to illicit the most useful possible response. As mentioned by a CEO earlier in today’s episode, I also think it’s helpful to ask for the same type of output, following the same structure, for different questions of a similar nature, to ensure continuity, consistency, and comparability across your various requests.

Conclusion

So, what have we learned today?

Well, to begin with, I hope you learned about at least one use case that you might consider incorporating into your own daily workflow. But, beyond that, a few things stuck out to me:

• The first is just the massive variety of use cases that were reported, particularly among the Searchers and CEOs
• I also noticed that the most impactful use cases were careful not to outsource their thinking or their judgment to A.I., but instead to use A.I. as a thought partner to enhance their thinking and judgment through thoughtful questions, the identification of blind spots, or any cognitive biases that may be clouding their decision-making process
• Like so many things, A.I. in it’s “chat bot” form factor seems to be yet another example of “Garbage in, garbage out”, meaning that the quality of the output that you get is directly proportionate to the quality of your inputs (whether that be the prompt itself, the context, or the data & documentation that you provide)
• I also learned about some current limitations that still exist with A.I., at least in its current form: We talked about the dangers of outsourcing the production of your CIM to A.I., but several responses (that I chose not to highlight) also talked about the various things that they are not yet comfortable giving A.I. access to (notably email accounts, bank accounts, and purchasing systems), because they simply don’t trust it yet.

In any case, hopefully today we managed to go beyond the generic platitudes that we’ve become all too familiar with, and provided you with some specific, tactical, and practical use cases that you can consider implementing into your own daily lives.

---

Thanks to our Sponsors