The general wisdom is that as an early to mid-stage startup your priority should always be product or business first and enabling technologies later. Once you get to the product-market fit, you can always invest in more tech.
In most cases for a early to mid stage startup, ML/AI is an enabling technology that can help do things better — better decisions, lower costs, faster processes.
However, ML/AI is no longer in the realm of a “fancy expensive tech that we will implement once we have lots of funding”. In many cases, implementing some sort of AI might be the magic ingredient to your product-market fit. Chances are you probably already know this but have been staying away from digging deeper due to a feeling of high technology or cost barrier.
This post is meant to help you, a founder or a CXO on a relatively less budget, make an informed decision on using AI in your product or business from the ground up.
Note: I am using AI as an umbrella term to encapsulate all Machine Learning and Deep learning software and tools that have matured and are available today for use off the shelf.
Hiring for AI-readiness
Everywhere I speak about AI, there is one question that I am asked — how do you get the right resources for AI ? And my answer almost always is, well you don’t hire resources for AI. You hire smart people who are good a in a few things and then you give them a mandate to solve problems using AI.
Do I need to hire a data scientist ?
Yes — if you basing your business on AI.
No — if it is a one time system that will run in the background.
As a startup, unless your are funded from day 0, you probably do not have budgets to hire PhDs. It doesn't even make sense to hire PhDs or data scientists unless your product or business is directly in AI. For all other folks building a business where you are looking to leverage AI or be ready for AI, below is my Rx list —
- Hire a good Python programmer — the beauty of Python is that it is powerful, simple and works well across different use cases. So if you have a great Python guy, you will be able to leverage him across different problems and he can become your go-to person for AI. As a startup, being efficient with resources is crucial. Plus, it’s important not to have a technology person boxed in and only capable of doing one specific thing, isolated from the larger business or product challenges. Good python programmers are also reasonably good with math and that is a plus.
- Look for great communicators and team-players — AI is a rapidly emerging and changing field. Things are evolving as you read this; new tech is replacing old tech and new solutions are getting found. You want to have someone who can speak up, within the team and also outside to others, who is humble enough to say he doesn’t know and ask for help outside and smart enough to share back things he has found out. Only a two-way communication approach will work, since the community expertise is crucial for getting to a solution fast.
- Hire someone with natural curiosity and domain understanding — To apply AI, the first task is to understand what one has to apply and for whom — without the domain understanding and a natural propensity to ask why or understand why not, a good programmer is of no use. Don’t hire people who are not naturally curious and constantly tinkering to find new things, even if they are good programmers. For an AI role, such a hire will not work even with all other resources.
Choosing the right tools of trade
AI solutions are one of the most actively developed areas today, and there is a host of services available that your engineering team can use — ranging from free libraries and frameworks to paid services.
Rapid prototyping for AI
As a startup you want to rapidly test a set of data, build models and come to some quick conclusions or build new process or features that leverage AI, rapidly. Following are a few options, not specifically in any order, that you can choose from. Consider this a startup CXO’s toolkit or ready-reckoner for AI solutions.
One key thing to remember is that staying on the same programming stack and building a set of capabilities for data modelling, analysis and running deep learning models is a huge plus for a small engineering team.
Keras is a high-level neural networks API, written in Python and allows for very quick prototyping — from ideas to results, in a very short span of time. As a early stage business this is very valuable, because you need the agility to know that a hypothesis works or doesn't work and then decide whether to focus on it or move to something else.
Pandas is an open source, BSD-licensed library providing high-performance, easy-to-use data structures and data analysis tools for the Python programming language. You can ask your engineers to use combination of these two tools along with some other Python frameworks such as scikit-learn.
Google Cloud Platform / Amazon AI toolkit have a wide range of services available for use. In case you need some heavy-lifting — image recognition, voice recognition, natural language processing — consider GCP or Amazon AI toolkit. Both of these platforms have python bindings making it easy to build solutions that use Keras or Pandas with them.
Note: At Kontikilabs we are coming up with a detailed white-paper comparing AI services of all major platforms. Reach out to us on hi[at]kontikilabs[dot]com for this free whitepaper. We will be happy to help!
Calling in experts Vs building in-house
This is a tricky question to generalise on. As a startup you want to make sure you do things fast, and at a lower cost but at the same time protect your IP, in-case you stumble upon something interesting. The general thumb-rule that I prescribe to folks is:
Build if you plan to make this part of your core IP or product flows and consider this as a edge over the competition. In other words, you are an AI-first business.
Outsource if you plan to leverage this as a tech to lower costs, make better decisions via analytics or make a small improvement in your current feature-set — that is, you are an AI-ready business.
The challenge is that many times you may not realise in the beginning whether you are likely to become AI-first or AI-ready because you are only testing out a hypotheses. In such cases, it is always safer to keep things in-house or work with a very trusted partner, in case one is accessible.
AI-Readiness is key to startup success
AI is rapidly becoming an infrastructure item for businesses, and therefore every startup must examine their business from an AI lens. If you have data, you have scope for leveraging it in some way. I hope this post will help you in some way to become more AI-ready in 2018.
