AI: Where to Invest?
Consumer, Cloud, Enterprise?
This week I was a guest on Brent Leary and Paul Greenberg’s CRM Playaz.
The topic was ‘AI's impact on Enterprise Apps Angel and Seed Investing.’
You should click through to hear the discussion but here is a snippet from X.
This week’s essays focus a lot on the same questions. Dario Amodei writes an essay called ‘Machines of Loving Grace’, and there are responses. Rex Woodbury makes a strong case for consumer applications. Sequoia Capital shows a graphic with a big blank space - leaving the question unanswered: Sequoia Capital’s ‘Generative AI’s Act o1
But Sequoia does state:
We are seeing a new cohort of these agentic applications emerge across all sectors of the knowledge economy. Here are some examples.
By bringing the marginal cost of delivering these services down—in line with the plummeting cost of inference—these agentic applications are expanding and creating new markets.
In this context, having started as being sure AI would not destroy cloud companies it stated:
That being said, we are no longer so sure. See above re: cognitive architectures. There’s an enormous amount of engineering required to turn the raw capabilities of a model into a compelling, reliable, end-to-end business solution. What if we’re just dramatically underestimating what it means to be “AI native”?
My own view is that AI is not a layered architecture at scale. That does not mean there will not be applications or services for both consumers and work. But it does mean that the default is that AI use cases at the customer or consumer layer are likely to be features not businesses.
A feature can survive for quite a while, and even get traction like NotebookLM from Google (see this week’s Video of the Week). But it is most likely going to be incorporated into the AI platforms eventually, and probably quickly.
If I am right then the current VC inclination to avoid investing (see Aileen Lee’s piece from last week) is prudent. And it may be the case that there is no compelling case to change that behavior.
Tomasz Tunguz this week talks about the next AI S Curve, but is rightly focused on the performance of ever larger models.
I am not saying that this landscape will not change. A lot depends on how aggressive OpenAI, Anthropic and others go after the use cases. But The Information’s article - 78 Artificial Intelligence Startups That Could Be for Sale This Year - suggests that the mood away from investing in late entrants or features is in full swing.
Recommended by LinkedIn
One use case I feel sure about is using AI to evaluate venture backed companies. This does not use LLMs, but rather combines heristics that are created by and score humans with Machine Learning pipelines.
Rob Hodgkinson wrote about how SignalRank’s platform combines scoring human attributes with data driven decision making. You would have received it as a cross-post on Thursday.
He explained how SignalRank captures human attributes and transforms them into predictive signals by quantitatively analyzing investor behaviors and decisions rather than focusing on company-level data or financial fundamentals. Here's how this process works:
1. Codifying Investor Behavior: SignalRank ranks investors by round stage (such as seed, Series A, Series B) and captures attributes like the number of successful unicorns they’ve backed, their efficiency in identifying unicorns (unicorns/investments ratio), and their Multiple on Invested Capital (MOIC). This ranking allows SignalRank to evaluate investors' perceived quality rather than startups' quality directly.
2. Aggregating Social Sentiment: Instead of purely relying on company fundamentals, SignalRank aggregates investor patterns and decisions across multiple rounds. It considers which investors are leading specific rounds and assigns value based on these decisions, essentially converting social sentiment (investors endorsing a company by investing) into quantifiable data. For instance, a Series B led by a highly ranked investor may imply greater potential value for the company.
3. Round Scores and Patterns: By examining investments over a rolling five-year period, SignalRank identifies patterns across multiple funding rounds. Each round is converted into a "round score" and then further into a "company score," benchmarking them against similar rounds in the same funding vintage. This creates percentile scores for rounds and companies, giving an objective signal from subjective investor activities.
4. Dynamic Thresholds and Market Heat Analysis: SignalRank also uses a dynamic thresholding system to adjust how it interprets investor signals in different market conditions. For example, during a market bubble, investor behavior may reflect herd mentality rather than independent decision-making. SignalRank’s system aims to recognize such situations by using recent market data (90-day averages) to maintain rigorous investment standards.
5. Emphasis on Investor Quality Over Company Data: SignalRank believes that exceptional investor behavior reveals more about future success than current company-level data, which can be noisy or unreliable. By focusing on investors, who are playing "multiple hands across multiple tables," rather than entrepreneurs playing a "single hand," the model aims to aggregate consistent human judgment over time.
6. Combining Human and Machine Insights: SignalRank captures qualitative human attributes like investor reputation, their history of selecting winning companies, and the implicit social hierarchy of investors. These subjective factors are turned into structured data, which the SignalRank algorithm uses to create objective signals. This approach allows the model to harness top investors' intuition, experience, and networks, making it a combination of human insight and machine learning precision.
In essence, SignalRank’s method turns the socially constructed elements of venture capital—such as investor endorsement, market perception, and the quality of backers—into quantifiable metrics that can be analyzed, scored, and used for predictive outcomes. This strategy helps identify companies with a higher probability of delivering significant returns based on the historical patterns of investor involvement.
We made our 20th investment into the SignalRank Index this week. Supporting early stage investors by underwriting their pro-rata shares in Series B rounds. Qualified and Accredited investors can buy the index for $25.54 a share if they invest a minimum of $500,000.
This is the first index of private market assets available to qualified and accredited buyers. Once listed it will be available to retail buyers also where the minimum purchase will be a single share.
As money flows into the index it is used to support our partners pro-rata in the next set of companies. The goal is to have an index of around 200 companies at scale, constantly refreshed through exits and new investments.
Click through below to see how data intelligence can select top performing private companies.
Hat Tip to this week’s creators: @DarioAmodei, @2020science, @rex_woodbury, @ttunguz, @steph_palazzolo, @rocketalignment,Matthew B. Crawford, @alex, @jasonlk, @michael_bodley, @AndreRetterath, @mwseibel, @cademetz, @mikeisaac, @eringriffith, @mgsiegler, @DanMilmo
Contents