Building valuable AI startups..

Part I : Role of Dark Data

At pi Ventures, we are looking to funds early stage companies in Applied AI, ML & IoT. As a result, we do get quite a bit of deal flow, and given our backgrounds in similar space as entrepreneurs or operators, we end up meeting 3-4 startups every week. The experience has been a mixed bag - whereas we do meet entrepreneurs solving the pertinent problems with the relevant use of AI, we also end up meeting entrepreneurs who try to slap AI in their pitch or product to make them “hip” in some sense or other. For some of them, focus on AI seems superfluous or misplaced, or in some of them, it can even take them away from their core problem area that they might be focusing on, as a startup. Maybe it has to do with the increasing amounts of funding and M&As happening in the AI space.

So we thought, we should share our learnings through this blog, to help startups explore how they can use the transformative nature of AI to build “real AI” startups that have the ability to cause deeper disruptions and thus create higher business value. So we are writing our experience as part of a three blog series. In the first one, we will explore Dark Data and how startups can start building AI startups with business value in mind and looking to exploit it over a data-driven roadmap.

For that, we need to first dive into massively growing data universe and how it is fundamentally changing the view of the world for Machines.

The world that we are in, can be defined in the eyes of machines as a set of actors which communicate through data. Data is the medium, which communicates about the state of the world to machines. That is how machines react to the physical world that we are a part of. As the world gets richer in data that machines can sense, comprehend and act on, possibilities of their usage and effectiveness multiply. This data is growing massively, from 2009-2020, the growth is almost 50x. More notable this data is getting richer and richer, in describing the world around us to machines. Thus making them more capable in terms of their actions and abilities.

Pic Credit : pi Research, Data Source : IDC Research

So as we start to build out this data universe that defines our world for machines, let’s try and segregate, how this data universe could be better understood in three major segments :-

• Big Data or data generated by enterprise applications like supply chain systems, CRM, and order management. Adoption of the cloud has accelerated the flow of data across the enterprise and to a large extent, beyond it as well.
• Small Data is how we are defining data generated by personal devices like smartphones, smartwatches, fitness trackers revealing personal, contextual information. Small data is very compelling as it allows us to humanize data elements for behavior change or feedback. For example, the new Apple Watch 2 asks you to breath by allowing you to visualize your breath on the screen, the effect of which is almost hypnotic in some sense.

Pic Credit : Macworld

• Dark Data is how we are defining data that never came out before but now is available for analysis and action, most probably due to the Internet, IoT sensors or devices (e.g., real-time residential temperature data from “Nest” or the data coming out from the Ola Cabs ). This is the most critical of data sets that are emerging now, as they play a vital role in completing the information that can result in insights, that are comprehensive enough to drive action, either human or autonomous. Mobile is one of the first robust IoT devices and it is driving this early dark data out, as it reshapes both B2B (say hyperlocal logistic workflow) or B2C (mobile shopping or intercity taxi hailing) workflows. This data element is the most important for AI startups.

What we are seeing is that disruptive startups are combining, two or more data elements described above and creating insight loops through the use of AI/ML focused on a domain to break through. These insights loops generate new economic value in form of predictive business or extreme personalization, eventually building a business case for new startups. We are seeing some early evidence of this new economic value in a few verticals very clearly.

Dark Universe is a major discovery, which just completely reshaped our view of the universe. Taking a cue from that a term “Dark” data is being defined as the data, that has always existed, but so far, was beyond the realm of access for machines (as it was never available, or was not available for analysis or action in real time), thus reducing their view of our world. When we start to put sensors around in our world, we are starting to make parts of our world visible to machines through the data that these sensors produce. This “now visible” part of the world is essential for them to make decisions closer to how humans would do, only better in most cases. So whether it is the state of taxi (occupied or unoccupied) or position of vehicles in moving traffic in real time (Lidars Sensors in Google Car), this dark data coming out, will make their view comprehensive and create the whole story for machines to either make human actions more effective (in this case, telling taxi drivers to move to a part of the city, where there might be more demand coming through or going to come through) or enable complete autonomous actions (self-driving cars or enabling nest to auto adjust temperatures inside the room so as to lower energy bills).

Source : UBER Data Science Team

In the picture above, it is interesting to note how machines (in this case cars having UBER app) might view the map of New York for instance. Brighter spots showcase higher concentration of trips, as you move the cursor around. Very soon, inaccessible part of the worlds may not be the places that are difficult to reach, but places where dark data is not coming to light.

First thing is to look for solving a problem where you have a business value clearly visible to a business or to a consumer group. Quite a few teams, especially led by very competent technology founders, have gone and built an AI approach, which is revolutionary. However, we are increasingly seeing, if they are not able to narrow the application of this approach to a business value, they might run out of runway. So identifying a vertical or specific use case, where this approach creates a significant value is a critical first step for good teams. Post this, you need to see, if along with the business value, you are able to identify the critical Dark Data that can come out or is coming out as part of the economic use case, you need to put in efforts to extract maximum business value out of it. As such, what is critical for AI startup teams is a core data strategy with a central focus on either bringing the dark data out or if dark data is already coming out, exploiting it for new economic value. If they are bringing out dark data, they better have technology capabilities in understanding hardware, software and data as part of their core team. If they are leveraging dark data that is already out there, they better be really good in understanding that data and mapping it to the business value within a domain.

Now once you are successful in getting the Dark Data out, you need to build stronger moats through further application of AI, so that you can gain insights from data that have higher economic value and that can lead to actions that can either deliver extreme personalization for consumers or help in building predictive businesses or both and build further differentiation along the way. So, business value of data is a very critical metric, as without which, you might struggle to guide your AI efforts.

The best part is that given the ease and availability of AI frameworks and Libraries from Google, Microsoft, Amazon and others, some of the best teams can leverage most of the AI research effort that has gone in last 2-3 decades, almost “on tap”. As such, we believe that they can focus and should focus their efforts on building vertical solutions that are built on delivering business value in a domain. Based on the verticals and availability of dark data, startups need to fine tune their product strategy carefully. If they can do this almost “mindfully”, we believe that the vertical applications of AI are well within the execution capability of startups and can potentially lead to next wave of leaders in most verticals.

In the next blog, we will present further ideas on how to measure the business value of the Dark Data. Stay tuned.