10 microtrends for Applied AI start-ups
A ‘micro’ trend is more than a ‘fad’ but less influential than a ‘macro’ trend. Mark Penn, who coined the term, referred to microtrends as “the small forces behind tomorrow’s big changes”. While artificial intelligence, machine learning and blockchain are all macro trends that will persist beyond 10-20 years, microtrends typically last 3 to 5 years before being replaced, or becoming implicit as a result of widespread adoption. An example could be the application of a specific AI technology to a common use case, one which eventually disrupts an entire market horizontal or industry vertical.
Based on 490 interactions with ‘Applied AI’ startups (22% of all FP leads in 2017) and examples from our own portfolio, we use these insights and observations to predict the top AI microtrends for the next 5 years.
AI chatbots become implicit in customer service interactions
Commoditization of underlying NLP technologies has made bots somewhat ‘faddish’. A host of APIs, frameworks and off-the-shelf (click and build) solutions means it is easier than ever to build AI chatbot assistants (so easy we built one ourselves!) The growing number of ‘bot’ startups suggests this is indeed a “microtrend”, and it is reasonable to predict that up to 80% of all ‘top of the funnel’ customer interactions will be automated via virtual assistants in the next 3-5 years, regardless of whether they take place on voice or messenger platforms.
An increasing number of startups are basing their entire proposition on chatbot applications. This includes everything from robo-advisors capturing investment goals, to wellness bots keeping tabs on your mood or eating habits. More sophisticated solutions adopt a combination of automation to speed up the interaction layer, and human intervention to execute the last mile e.g. Wealthsimple, which opens up professional investing to high street investors, or Babylon, providing instant access to GPs in the NHS 24/7.
While the more sophisticated solutions could yield venture scale opportunities, applications focused solely on the interaction layer will become implicit across service interactions.
Automation makes SaaS even better, faster and cheaper
SaaS has had a good decade, becoming the dominant design for delivering software and value to customers. It has made it faster, easier and cheaper to spin up a new provider for almost any task an organisation might wish to achieve.
Investors in this space are now increasingly asking “what comes after Saas?”, making comparisons between the “No software” movement from cloud computing with a growing perception that SaaS is dying. Gil Dibner summarises it well: “The easy availability and mass adoption of cloud-based (SaaS) technology makes advanced software systems so much easier/cheaper/faster to build that “value” is rapidly bleeding out of the software stack”.
Remember “No Software”? (Image credit Computerwoche)
Beyond the web of connectivity services that help to glue different cloud platforms together, (using API connectivity apps such as Zapier), machine learning and AI automation could offer a significant differentiator to the next wave of SaaS startups.
Ironically perhaps, a great example here is in software quality assurance testing. For instance, using a combination of computer vision to automatically detect site changes and model-based test automation, it is possible to significantly reduce testing time compared to traditional script-based methods.
Blockchain-based applications meet the enterprise
Weathergate Capital provide one of the best descriptions on the potential for blockchain-based enterprise applications, defined as: “[The] underlying technology that allows computers that neither know nor trust each other to do business together in a transparent way”. They predict current Saas and client server applications will be replaced by the decentralized web (see graphic), where globally distributed resources “can cooperate to achieve a common goal even though they don’t necessarily know or trust each other”.
Blockchain-based architectures replacing SaaS and client servers (Source: Weathergate)
This type of decentralised architecture could provide new opportunities for startups developing AI applications, whether through ‘decentralised intelligence (e.g. TraneAI, which trains machine learning models in a decentralized way), or open up the AI black box by helping to explain itself better. On the flipside, machine learning algorithms have the potential to make blockchain networks more efficient and hence more scalable.
Next generation ecommerce
As opposed to AI being the fundamental economic value behind a startup’s product or service, we are increasingly seeing AI as an enabler, particularly in next generation ecommerce. There are a variety of startups out there pitching new products and business models capable of disrupting existing markets by leveraging off-the-shelf technologies.
One use case is in driving operational efficiencies in a startups operational model. For instance, some direct to consumer startups have championed the “zero stock, zero lead time” model to build a cost advantage versus incumbents.
Another obvious use cases for this is in personalisation, where recommendation engines built on collecting user preferences helps companies such as Thread to help people “dress well without trying”. Based on the 80:20 rule (in the context of machine learning), it is possible to reach 80% accuracy by using aggregate data i.e. “people who bought this also bought that”, with the remaining 20% inferred from individual-level choice sets and behaviors. These models become better with each new customer or customer interaction and will no doubt become more ubiquitous over time.
Thread.com helps you find clothes you’ll love
Data-driven marketplaces backed by impeccable execution
We often hear there are few marketplace opportunities left to invest in, and those that are left are hardest to execute. This is absolutely true. Successful execution of online marketplaces will be driven by a combination of AI / data-driven platforms and impeccable execution, one being as important as the other.
A good example of differentiated, “data-driven” marketplaces versus the more generic “digital” solutions of the last decade is a company in our portfolio called Lexoo – a legal marketplace that matches qualified lawyers to appropriate legal contracts using a combination of relevant variables. Founded by a former corporate lawyer, to execute on this, it requires a high degree of domain expertise and execution to make it work.
Compare this to MyTamarin a marketplace in (arguably) a much tougher market, such as childcare. Proposing an “AI first” solution to what is a complex and enduring social problem requires much more than a data-driven matchmaking platform based on personality profiling: “Empowering parents through personal(ity) touch. Whether you’re looking for a long-term maternity nurse or occasional night nanny, we’ve got you covered!” Success in this market would require a complex operating model to overcome adoption barriers.
Online marketplaces of the future will depend equally on AI-driven (matching) technology as they will impeccable execution.
Micro trends in industry verticals
It is not just market horizontals that are demonstrating some interesting microtrends, as there is increasing deal activity across several industry verticals (below).
AI Heatmap (Source: CBI Insights)
Optimisation of financial services
Automated trading almost certainly accelerated the recent sell-off in equities, which saw the Dow Jones Industrial Average crashing 800 points in ten minutes. A large part of financial service transactions are driven by algorithms & computer trading, which remain highly inefficient (particularly on the basis that nothing fundamental has changed that might influence asset prices). AI has the capacity to bring enhancements to contexts as varied as enterprise payments to consumer finance.
AI and Patient-centric care
Looking through the list of 106 startups in CBI’s AI and health overview, new business models are everywhere, from mental health and drug discovery to lifestyle management, virtual assistants, hospital management, and medical imaging and diagnostics. Some of the most exciting opportunities are in AI and patient-centric care. For instance, the Berlin-based healthtech startup Ada Health, which transformed their backend doctor diagnostic tool into a consumer service.
Food supply chain “eaten by AI”
This is one of the fields with a lower impact from software, but could be reaching a turning point as a result of emergent AI technologies. Those lucky enough to have had a glimpse into Ocado’s distribution centre can see the impact that automation can have for retailers at the consumer end of the value chain. Working backwards from the production to agriculture, stories of D.I.Y. Artificial Intelligence used in a Japanese family farms to tomato sorting machines inspire countless possibilities for AI to eat up opportunities in agriculture.
The long road to autonomous transportation
Whilst a whole host of companies are claiming technology readiness within the next 3-5 years, autonomous vehicles will be held back by various system-level barriers. Commentators predict two paths to autonomous transportation: The “evolutionary” path, where today’s cars get self-driving features bit by bit (e.g. Tesla’s autopilot feature), and the “revolutionary” path, where totally self-driving cars start as test vehicles and become more mainstream as they can drive in more places (Google cars). Regardless it’s a long road to becoming ubiquitous.
Heatmap: Applied AI microtrends
As a summary for the next five years, we have created our own heatmap of Applied AI microtrends, which sets out each applications as it evolves from ‘emergent’ to ‘ubiquitous’ and eventually ‘implicit’ across a variety of use cases.
Accelerating the learning curve:
Edtech is a large and growing market. We are seeing great leaps beyond the rapid adoption of digital solutions (such as MooCs or online learning platforms) to a new wave of edtech startups. AI is helping to optimise the learning experience, making it more efficient, more effective and (yes, if some can believe it) more enjoyable. Multiple data inputs (not just text but also audiovisual) can be analysed using deep learning techniques. Everything from speaking a foreign launguage with real time feedback, speeding up revision with NLP summaries or zipfian distributions, leveraging pattern recognition to learn musical instruments and even style transfer applied to the art of sketching.