Learning What Resonates with the Market

Source: Deep Learning on Medium

Learning What Resonates with the Market

Human Collective Intelligence and AI

Tom Kehler, CEO, CrowdSmart

Resonance in physics is defined in terms of synchronous vibration. My Ph.D. thesis was in fact on understanding magnetic resonance in thin films. In sound we experience resonance as a deep reverberation that is generally pleasing. In the same way when we speak, we want something to resonate with our audience. In essence we want to be in synch with our audience.

Producing a product that resonates with a market need leads to explosive growth. In the same way, when the investor market gets in synch with a company’s mission, the value of that company often skyrockets.

So how do you predict what will resonate with a market? This issue has been my passion since the onset of the internet. In the mid nineties I had the privilege of being CEO of Connect, one of the first e-commerce platform companies. Connect was originally Apple Computer’s Development Community and was pivoted into an e-commerce platform that went public. There, for example, we provided the platform that launched the first on-line sales of digital images: PhotoDisc (later to become Getty Images). While it was obvious that the internet provided explosive efficiency in delivering product messages and marketing programs, it also had the hidden promise of being a powerful listening platform.

While at Connect, an old friend came by with a process that he claimed would detect alignment of a group of people around a prediction or decision. We tried it initially with index cards which worked for small groups, but the process was cumbersome. My friend formed a company and I invested. I also connected him to another friend and colleague who had a Ph.D. in computer science from MIT and had been the CTO of my first company, IntelliCorp. My heart was now in this new emerging venture. Shortly after the Connect IPO, I exited and went to the new company.

Over the subsequent months that led into a couple of years we figured out how to scale this to a process that would work asynchronously with large audiences. With the new process we could ask an open ended question like: “What did you think of the SNL host?” and learn with statistical accuracy the statements that most resonated with or represented the audience feedback. We patented the algorithm and engaged leading companies with consumer facing websites including NBC, General Motors, LEGO, and Procter & Gamble. Through that technology, we learned that LEGO’s audience resonated strongly with an extensive engaging building experience: the Imperial Star Destroyer was born. The Imperial Star Destroyer was the largest and most expensive LEGO product and its one year planned inventory sold out in 7 weeks. It resonated with the audience.

The road to commercialization of that technology was not easy. We started as Conclusive and could not get external funding. I took a job as CEO of Adaptivity (whose original plan was to be an object-oriented development platform) under the condition that the investors would buy Conclusive. They did and that technology later became the core business. We renamed the company to Recipio.

Recipio had a great early start signing on NBC, GM, LEGO, P&G and many others. As we were scaling, we ran into the funding freeze following 9/11. In December of 2001, we could no longer make payroll. A breakfast at Buck’s in Woodside in late December 2001 led to Recipio getting acquired by Informative. We managed to keep the Recipio team of 20 intact and move them into Informative by late January of 2002. Once again we were off to the races and the Recipio technology and customer base took over Informative’s revenue line.

At Informative, we began to focus the technology on “how to improve your net promoter score”. There we were mining for what would shift the energy of a customer base to strong promotion and recommendation. It worked well at that and we ended up in two case studies (Intuit and LEGO) in Reichheld’s book “The Ultimate Question”. We got acquired by Satmetrix in 2007.

After 12 years over four companies I decided to move on but not away from the idea. I returned to my roots and began to dig into the mathematics of why the algorithm converged and worked so well. We knew it was related to a Bayesian concept and that convergence was related to a Markov process.

In 2013, we recreated the technology but now on a firm mathematical foundation recognizing that the underlying support for convergence was in fact a close cousin of the PageRank algorithm based on the same mathematical theorem. I now was focused on taking this to the investment world with the question: How can we detect the future market in the early stage investor community for a startup? We formed CrowdWisdomTech (CWT).

Enter CrowdSmart. In 2015 I co-founded CrowdSmart with Fred Campbell, Markus Guehrs, and Kim Polese. CrowdSmart licensed the CWT core technology and by 2016, we were in business with a small fund testing the application of human collective intelligence and AI to predict startup funding momentum.

Based on best practices in early stage investing, we developed a process that combined quantitative scoring of startups’ market opportunity, the team fit for executing the strategy and their access to networks. The qualitative discussions and brainstorming were all captured by our learning algorithms and NLP technology. We then trained machine learning models against ground truth data based on follow-on funding momentum. Funding momentum, the ability of a startup team and story to extend their runway and increase perceived value to investors, is the highest correlate to a profitable return on investment.

The CrowdSmart process brings together approximately 25 accredited investors and experts. The expert team has access to a deal room (with the normal set of complete investment materials for a startup seeking seed funding of $1 to $5 million). They are then led through a process of interaction that is like a brainstorming session with the startup and other investors. The process is single blind (startups don’t know the investors identity but they see the data and investors see each others’ data but they don’t see identities).

It occurred to me that we were on to a method that linked to early work I had done on knowledge acquisition and knowledge representation. That work was published in 1985 under the title “The Role of Frame-based Representation in Reasoning” with Richard Fikes. With contemporary NLP technology we were able to link reasons for investment to underlying concepts.

The CrowdSmart platform builds a knowledge model that links opinions and scores gathered through this interaction. Rather than model knowledge as a frame-based semantic network, we modeled the knowledge model as a Bayesian Belief Network (BBN). The BBN for each startup is used to generate data for a machine learning model (black box) trained against ground truth data. The result is a score that reflects the likelihood of a company’s success along with an explanation of why.

How does this tie back to the original idea of learning what resonates? The figure below is but one extract from our system but it indicates the power.

Example of score and resonant themes with investors

Each time an expert scores (e.g. let’s say they give the team a 9 on a scale of 1 to 10), they are asked to share their reasons for their score. Following their score and reasons, they are asked to engage with other investors not around their scores but around their reasons for scores. The system then learns the following:

  1. What reasons are most relevant to the group of evaluators?
  2. What themes or topics best describe collections of reasons?
  3. What is the correlation between highly relevant topics and overall scoring?

The resonance phenomena is fairly straight forward. In the given example, the expert audience resonated strongly with potential customer demand for their product. They also resonated with the technical team’s ability to deliver and finally they aligned around the high potential for an exit. The peaks in popularity of these themes correlated with high quantitative scores.

The story above evolved over two years. The resonance curves above were collected before anyone knew whether or not the company would raise their seed round. Nor did they know that 100 days after the seed round the company would get a series A term sheet from a top institutional VC followed by a Series B 8 months later that included another top tier VC.

In early 2019, CrowdSmart pivoted from proving the technology with a small fund to providing the technology to investors, corporations and accelerators in a SaaS system. In the early stage market, the system is testing at a >80% accuracy in selecting those companies at seed stage that are likely to go on to growth rounds.

We can now help investors and corporations find what is likely to resonate with the market. We believe this is the future of investing in transformative innovation. We are at the beginning.

I measured my first resonance curve of a magnetic thin film 50 years ago! Understanding why systems align and resonate has been a life-long journey. The ability to detect what resonates in the market is powerful and promises to deliver great value. If you are interested, we would love to have you partner with us to see if we can find what resonates with your future.