Consistently predicting the next disruptive company is the holy grail if you are interested in start-ups. We’ve been testing out some Deep Learning techniques on our data to help make such predictions, and have had some interesting early results we thought we’d share.
Word2vec is a Deep Learning technique first described by Tomas Mikolov and his team at Google in 2013, and in basic terms it allows a model to be built for a particular dataset (or corpus) in which words are represented as vectors. One of the most interesting outcomes of this approach is that we can gain insights about text by analysing word vectors arithmetically.
In a classic example of the power of Word2Vec (trained on a large dataset), the vector of Queen is found to be almost equal to King + Woman – Man. In other words, if you remove Man from the King and add Woman to it, logically you get Queen, but the Word2vec model is actually able to represent this arithmetically. This is quite amazing considering the model has only been fed raw text, with no specific training about the concept of Man and Woman, King or Queen.
Doc2vec is an extension of Word2vec that treats documents (in our case, company descriptions) as vectors. Using Word2vec and Doc2vec techniques trained on VentureRadar data we decided to use this concept of adding and subtracting vectors to see if we could predict likely companies to be the next big disruptors.
It’s something of a cliche in start-up land to describe a new company as the ‘Uber for X’ or the ‘AirBnB for Y’, however this can be useful shorthand to communicate where a specific business model is being applied in a new way. In our test we used this way of thinking about new disruptors.
We asked our model to find:
You can see the full results below in the grey boxes.
Overall the results are generally returning back the types of companies we’d expect for these queries. There were a couple of tests we tried where the results were much less accurate, and in these cases the issue seems to be related to lower numbers of companies (so less training data).
In the next steps we’ll try to automatically predict which markets are likely to have new business models applied to them (and be disrupted next), so just knowing for example that Uber already is applying a disruptive business model would allow us to predict where else their model could be applied, and identify the best emerging companies in each of those areas.
(1) Finding an Automatic (connected car company) for the Home
[Using the calculation: Automatic – cars + homes]
Expected Companies: Smart Home companies / Top 5 Results:
iControl Networks: “iControl Networks, Inc. is a software and services company providing interactive solutions for broadband home management. The iControl OpenHome™ Software Platform has made the “connected home” a reality, and enables home security companies, broadband service providers, and utilities to offer their customers the next generation of home management, security, and connectivity.“
AlertMe: “AlertMe provides a single platform for data, insight, control and intelligent automation in the home. Based on a multi protocol home gateway and cloud-based applications, AlertMe integrates many devices and services in the home through a single ‘super app’ to create a system that connects everything, and has the ability to intelligently automate devices so they work together seamlessly.”
Green IQ: “Our vision is to bring smart-home technologies to the next level, making home automation an easy and fun experience, at an affordable price. We, at GreenIQ, believe that a smart home is also a green home. Our first product, the GreenIQ garden computer is an affordable, easy to use garden computer that controls your garden’s irrigation and lighting, and saves you money on your water and electricity bills.“
TreeGreen: “The energyEGG is the flagship product from TreeGreen. The energyEGG uses a patented occupancy sensor to detect when a room is empty, and then it wirelessly signals plug sockets and lights to ‘switch off’ the power. It also has ‘one touch’ manual control and a user set timer to delay when the ‘switch off’ command occurs.“
GreenWave Reality: “GreenWave Reality is a global innovator in the Smart Home Services market which includes Energy Management, Connected Lighting, and Home Monitoring services. The GreenWave Reality solution delivers a breakthrough consumer experience making it easy to conserve energy, save money and enhance lifestyles.“
Review of results: The model is identifying smart home companies, as we’d expect.
(2) Finding a Ginger.io for Cyber Security
[Using the calculation: Ginger.io – health + cybersecurity]
Expected Companies: Cyber Security monitoring companies / Top 5 Results
Apvera: “By applying machine learning Apvera leverages predictive analytics to understand the user behavior to anticipate and prevent future security breaches by identifying irregularities in usage behavior. Apvera helps companies understand enterprise security by focusing on user interaction levels and how relationships with services and applications may be deemed a threat; ensuring compliance with IT policies. “
DB Networks: “DB Networks is a cybersecurity innovator of Intelligent Continuous Monitoring. Our customers include the world’s largest financial institutions, manufacturers, healthcare, and governments. Intelligent Continuous Monitoring offers deep visibility into your core networks. Our customers enjoy new insights and a situational awareness of their IT infrastructure they have never had before including.“
Lancope: “Lancope, Inc. is a leading provider of network visibility and security intelligence to defend enterprises against today’s top threats. By collecting and analyzing NetFlow, IPFIX and other types of flow data, Lancope’s StealthWatch® System helps organizations quickly detect a wide range of attacks from APTs and DDoS to zero-day malware and insider threats.“
Malcovery Security: “Malcovery Security is the leading provider of actionable cyber security intelligence and forensic analysis, delivered through software and services that target cyber criminals and their activities. The company’s patented and patent-pending technology provides the ability to identify the ‘root sources’ of cybercrime attacks (servers, perpetrators, locations, etc.), delivering rich intelligence information about cross-brand attacks and targeted attacks, as well as advanced notification of emerging e-mail-based threats.“
Mentat Innovations: “We are builders of Machine Intelligence In Motion at enterprise scale, and beyond. Imagine a cybersecurity solution that is able to flag malicious activity against a background of shifting legitimate behaviour within the enterprise firewall. Imagine moving from diagnostics to prognostics to preempting faults in manufacturing plants, large server farms, or vehicles in motion. Imagine a machine intelligence offering early warnings of cardiac arrests on the basis of live data from wearable sensors.“
Review of results: The model is identifying cyber security monitoring companies, often those using artificial intelligence, as we would expect.
(3) Finding an Uber for Deliveries
[Using the calculation: Uber – passengers + parcels]
Expected Companies : P2P delivery companies / Top 5 results:
Cleveron: Cleveron is an Estonia-based postal technology company that specializes in the design and manufacture of customer-focused automated parcel terminals and software. With a broad experience in logistics, Cleveron has a unique understanding of the challenges posed by the rapidly growing parcel delivery market. The mission of Cleveron is to make sending and reception of parcels as effective, fast, convenient and favorably priced as possible by offering modern technology-based solutions.
Bellhop: “Ship your packages to us and we’ll deliver them to you at your most convenient time. Bellhop solves the problem of failed parcel deliveries by providing on-demand package delivery. It was created to fulfill last mile logistics and prevent roadblocks to fulfillment. With Bellhop you can buy an item online and ship your package(s) to our hub. Once the package(s) arrive at the hub, customers use the Bellhop app to schedule delivery in 30-minute to 1-hour windows, 7 days a week.”
PackLink: “PackLink is an online parcel delivery network which allows users to compare and purchase parcel and courier services at discounted rates. This door-to-door delivery service allows users to save up to 70% on parcel delivery in 3 simple steps: enter the package details, compare the prices and confirm the delivery.”
CarPal : “Your personal same-day delivery courier. CarPal is transforming the way local goods move around in cities by enabling anyone to get any product delivered on the day of purchase. CarPal’s revolutionary urban logistics & on-demand delivery platform connects customers with local couriers, who deliver goods from any office, restaurant or store in a city. CarPal’s mission is to become the on-demand delivery infrastructure for every major city in the world.“
Kolipo: “kolipo provides tools, a network of local stores (parcel shops), same-day, and late-night delivery solutions for businesses and ecommerce companies to avoid failed parcel deliveries.”
Review of Results: The model is picking out new parcel delivery concepts, although in this case a smaller number of the results are exactly applying the Uber model to parcel delivery. So this result wasn’t quite as accurate as the first two.