Artificial Intelligence (AI) is perhaps one of the most exciting technological challenges of the 21st century, and many tech geniuses are focusing their sights on solving its major problems. If you're thinking of launching an Applied AI startup, it might help to consider some primary Dos and Don'ts before you commit your resources.
Don’t Reinvent the Wheel
So much AI infrastructure has already been determined by leading engineers employed by industry giants such as Microsoft and Google, that starting from the ground up is a waste of everybody's time. Just as Amazon and IBM have already dominated the market for cloud services and platforms, there are also huge numbers of researchers developing open source solutions. So don't try to reinvent the wheel, but see what you can build on top of existing infrastructure.
Don’t be the Answer to an Unasked Question
If you don't have an application for your AI tech, there's little point in spending your valuable time and money on developing it. If there's a market for it, then there'll be a need for a tech solution, and although some concepts may be dazzlingly tempting, they're not going to be profitable without a definite use case. With the rapid expansion of the industry as a whole, it's likely that the next generation of startups will be use-case driven and industry-specific.
Do Use What's Already Out There
Again, don't reinvent the wheel, but make use of what's already been laid down in terms of infrastructure and tools. There are plenty of affordable SaaS options for machine learning to get you started, as well as open source libraries where you can learn more.
An open source software library with flexible architecture allowing easy deployment of computation apps across a huge range of different platforms. Originally developed by Google Brain, it offers strong support for deep and machine learning. TensorFlow's™ flexible numerical computation core can be used in many other scientific domains.
- Scikit-learn is a commercially usable, open source library offering advanced analytics in Python. It can carry out clustering, classification and regression tasks, as well as preprocessing, model selection and dimensionality reduction.
- The Apache Community is a huge collection of open source software products developed and made available for the public good. The OpenNLP project offers a toolkit for natural language processing, which includes speech recognition for sentiment analysis, sentence segmentation and parsing. Apache Spark™ is a unified analytics engine which you can use for large-scale data processing, combining existing workflows in SQL or Python.
- Microsoft Azure is a comprehensive cloud service, providing the infrastructure and tools to build your own AI-powered apps and models. Using the impressive toolkit, you can build simply and intuitively on top of their existing libraries, to produce new and innovative experiences.
- IBM Watson allows you to create machine learning models, and monitor and improve different models with automated experiments. It gives back insight on the data you provide.
Do Focus on Meaningful Datasets
Data's all around us (to paraphrase an old song), and it may be that you are a data architect, considering how to build a supporting structure for potential AI applications. What may facilitate a quicker entry into the market, though, is to think about partnering with some other innovative startup who can provide you with access to privileged or proprietary data. You could also use APIs to build datasets into open source or paid databases. Huge proprietary datasets and back-end build are a waste of your time.
Do Address AI Beneficial Use Cases and Pain Points
Back to reinventing the wheel, you want to be looking for areas of development where AI can tackle previously insoluble problems, or work ten times harder on current solutions. Grant Cardone's 10x Rule has become a work ethic as well as a production goal for many startups, but in AI it can be difficult to live up to, as gains tend to be incremental rather than massive. In order to address this problem, try looking for applications where AI tech's attributes or core benefits might offer a significant category enhancement, or will greatly improve the application's existing solution.
For startups, identifying pain points and setting your AI application to solve them is to make yourself indispensable in that particular market niche. The four main categories of pain point are:
- Procedural: current processes are approaching the problem incorrectly
- Productive: current solutions are inefficient and time-wasting
- Financial: current solutions are too costly
- Support: current solutions offer inadequate consumer feedback and support
Many human pain points arise from direct interaction between the business and the customer, so a beneficial AI use case would be to design an application for enhancing customer service experience. These can be as diverse as retail, educational, fintech and social applications – you just need to find the point where applying an AI solution is going to resolve painful experiences.
Everyone wants to combine their own subject matter expertise with AI to solve industry-specific problems, but you mustn't forget that you're a startup looking for a scalable business. You need to achieve a balance between technical know-how and efficient execution, to make your Applied AI offering a marketable product.
Do Apply AI Capabilities to Industry Opportunities
Many industries are already adopting AI solutions, especially those which are highly digitized and natural early-adopters. Consider what niche opportunities are afforded for AI applications by, for example, fintech, medtech, media and communications, and where they would make most impact.
Don’t Go Chasing Rainbows
Large tech outfits already have the lead in commoditizing AI and AI infrastructure, so you need to look for smaller loopholes and venture scale opportunities. Some of these might be in the B2B applications, where huge datasets are already available. Try thinking about Applied AI in a range of different consumer scenarios, cross-matching potential market applications with different AI techniques and core attributes.
The take-home from all this? Applied AI is experiencing a fundamental shift, from ideas that push technology forward to those which are pulled by market use cases. Contact us for one of our experts to get in touch.