There are many ways to be successful. Often the path you choose in business is so incredibly contextual—to your market, to your people, and to things as seemingly benign as your geographic location. No one article, or person, can define a path to success. However, what I can do is offer some insight into known failure patterns on the path to success.
AI and ML can only exist on top of a foundation built on solid data and analytics
Succeeding in AI Requires Being Good at Data Engineering & Analytics
First and foremost: if you aren’t good at data engineering, and business analytics, you won’t be good at ML/AI. In my experience, management teams often assume they can leapfrog best practices for basic data analytics by directly adopting advanced technologies such as ML/AI.
Heed these words: Companies who rush into artificial intelligence before mastering the fundamentals will end up paralyzed. Not only do you need to be good at data engineering and business analytics, you also need to embrace advanced automation. Lack of automation means people manually use the ‘bug-entry device’ (a.k.a. the keyboard) to process pipelines, do ETL, move data, and create downstream assets which does not scale. All of these activities can and should be automated to avoid the very real outcome epitomized by the infamous AI phrase “garbage in, garbage out”.
If you do not have technology that creates and maintains data pipelines and basic data engineering, your ability to solve problems with AI will become elusive. I think the human body is a good analogy here - if you had to spend all your cognitive cycles control the basic elements of breathing, digesting, pumping blood ,you’d have no time left over to the fun stuff. AI requires a highly sophisticated application of data to draw correlations and operate properly. In other words, give me a couple of years and I can make you into an overnight success in AI.
Do Not Undervalue Data
The best algorithms in the world cannot overcome bad data. Given a choice, I would much rather have more and better data than more sophisticated algorithms.
Data. Identify the real data that can impact your business. I know this sounds obvious but it is not always clear and there are definitely red herrings. A great example of identifying and using the right data is in horse racing. For years, horses for racing were chosen on lineage, which turns out to not be very useful. The key data was left ventricle size, and spleen size.
Lesson? Get the right data, focus and understand how to use it. This could be the difference between becoming the next Google or going out of business like Altavista. Remember, you don’t necessarily need to know WHY that data is important right now. Have you got data that is proprietary or impossible for your competitors to gather? Comb over that and figure out how it relates.
Choose Frameworks For Thought
AI done right will fundamentally change your business, and whatever you start off believing is the correct path most certainly will not be. So, in these highly dynamic situations, there are two frameworks I merge when thinking about something as transformational as AI: the Golden Triangle of People, Process and Tools, and the OODA Loop.
The OODA loop is Observe, Orient, Decide and Act. The goal of the OODA loop is to compress cycle time to maximize feedback and learning. Never has the OODA loop been more important than in the constantly shifting field of ML and AI.
The Golden Triangle: People, Process & Tools
Over my many years of running startup engineering teams, I’ve relied on the golden framework to help get me oriented in a new unknown situation.
- People. Always start with people, identify your key role players in the AI space and then understand what they bring to the table. After you take stock of what you have, you’ll have a map of what you need. Augmenting with outsourced help will bridge you as you grow your competence, however, AI/ML is important enough that you will want to have a core competency in this space.
- Process. Process is defined as the actions or steps you take to achieve a specific goal. Process is where the Golden Triangle overlaps with the OODA Loop. Ensure you’ve identified desired outcomes and reviewed the process and goals with the People.
- Technology. Consult your team and review the available AI technology. Chances are you will be deploying on one of the big cloud platforms: Amazon (AWS), Microsoft (Azure) or Google Cloud Platform (GCP). Considerations of the specific capabilities of your people come into play here, for example, if you are very new to AI and ML, perhaps DataRobot’s auto ML product is a good place to start, while if you have a higher level of expertise, Google’s Tensorflow may be a more powerful technology. Fit the Technology to your people and should be chosen after the problem is clearly understood and the solution requirements have been defined.
How to Manifest the OODA Loop
Start small. Execute pilot projects to test your digital maturity and stress your data pipelines to both find opportunities for improvement as well as gain momentum.
Observe & Orient
- Define the nature of the challenge
- Contextualize and chart a map of the territory
- Identify obstacles
- Recognize opportunities
You can both drive AI outcomes and figure out how to improve your human-driven analytical capability at the same time. Make sure your people know they are building a lab AND a factory. Use this opportunity to improve all the elements of the golden triangle:
- Remove the people-based skepticism that exists when making a huge shift in how you operate. AI will scare people and there has been more than enough hyperbole to make people wonder what’s real.
- Improve the overall process of collecting, cleansing and serving data to consumers of all types. It may be that AI is the desired “goal”, however, if you can improve your Business Intelligence at the same time, that has an immediate impact.
- Tools & tech are more important now than ever. Test before you Invest.
Decide & Act
- Actions to carry out
- Guiding policy
- Resource allocation
- Coordination to create impact
- Harvesting opportunities
Act Now: Do Not Wait
I believe the Cloud and AI battles will converge, and the winner will create multi-trillion dollar companies. Why? Consolidation of vendors, and the huge cost of infrastructure and infrastructure expertise has created a market force where the winner takes all, much more so now than even a decade ago.
If you were late to the party on digital commerce, imagine making that mistake again but an order of magnitude worse.
Starting now may result in some failures, however in failure is the learning that is what AI is all about. Flashy applications of AI could mean flashy failures on a large stage, so remember that AI will permeate largely invisible tasks—you don’t need to hit a home run, you just need to figure out the game.
Remember, You Are In The Prediction Business: Focus on Outcomes, Not the Path to Them
Avoid the natural human nature to understand all the details of why your AI or ML is making the conclusions and outcomes that it is. It is not at all hyperbolic to say AI changes everything, every facet of the business and capturing that value as soon as possible is critical in creating the momentum to enable an AI strategy. Done is the engine of more. Your team doesn’t necessarily need to know how a Convolutional Neural Network (CNN) works in great detail, while they do need to know what class of problems does it have appropriate application.
Take advantage of the G-MAFIA (Google, Microsoft, Amazon, Facebook, Intel, Apple) and leverage their pure research and basic science. as a lot of their work is either open sourced or documented for your consumption. Let them figure out the details. Build your in-house AI team and help them make an initial bet on which of the AI vendors makes the most sense for your business.
Avoid A Strategy Tax
What is a strategy tax? Sometimes products developed inside a company, typically a larger organization with more to lose, have to accept constraints that undermine their competitiveness, or might displease their user base, in order to further the cause of another product. AI can and should be employed to improve the customer experience of your service and offerings. Sometimes that is transparent - things are just better and smoother. Other times the application of AI changes the core interaction itself.
Be Specific Don’t just state that you want to be a leader in the AI space for your business problem. Design the roadmap that will get you there. Fuzzy strategic objects are everywhere right now. Understand the key metrics you want to improve and apply focus and resources there.
Coordinate Lack of coordination will almost always ensure that your AI/ML initiatives will fail. AI/ML benefits from coordination across a variety of disciplines. It truly takes a village to execute. Is the ability to choose and prioritize a problem? If it is, you are likely to suck at AI. Fix this.
Avoid The Tyranny Of Small Decisions
Avoid a situation where a series of small, individually rational decisions negatively change the context of subsequent choices, even to the point where desired alternatives are irreversibly destroyed. The devil may be in the details, however, moving a huge chunk of your business to be AI-driven requires being bold. Everyone will give you a reason why it won’t work. Remember that sometimes incredibly strategic decisions can be made quickly because some element is too overwhelming to get granular on. Anecdotally the decision for Google to build the TPU chip for Android was based on a ‘back of the envelope’ calculation of how much it would cost Google to store 3 minutes of speech for every Android device that existed. The pursuit of AI is in that class of ‘no brainer’.
Education Is Important
Ensure that your AI group has a mandate for education across all lines of business. Great ideas will emerge from within deep pockets and different groups. Remind people that AI is important and that you are looking for opportunities to boost employee engagement and drive innovation. Besides, who doesn’t want to learn about AI/MLthese days? It’s not just about internally sourced content, you should give your people the incentive to learn online and augment with your specific insight. Train at all levels, not just the math and computer people. Train your business management!
Don’t Forget AI Is Made Of And For People
Different from the people portion of the Golden Triangle, AI will change your customer base and your non-AI focused employees. Face it— there are going to be casualties in your organization as AI takes hold. AI can and will replace many jobs. People will need to evolve to other professions through training. Your customer base will evolve as well, and your analytical excellence will be required to understand how shifting human capital and customers change the nature of your business.
Here’s another truth bomb: you will also lose desirable people (highly regrettable attrition). Make sure to distribute leadership and knowledge so that there is no single point of failure when this happens. Success in AI will define some of your employees' careers and that will make them targets for the G-MAFIA.
AI is often held back by a culture that doesn’t recognize the need, a lack of data, and a shortage of talent. Keeping your team in the loop on the decision-making process, and enlisting them to help solve problems will help you get buy-in.
AI is a means, not an end. “Do you have an AI strategy?” is like asking “Do you have an app server strategy?” Create an environment for innovation, the perennial top concern for CEOs, is more important in the age of AI than ever before. Real innovation brings failure and requires strong willed commitment.
As for organizational structure, I’ll be honest, I don’t like the “center of excellence” model which is perhaps the most popular at the moment. Too often I find that it results in a “build it and they will come” attitude which often yields a lack of value creation. I do think that whatever structure you decide works for your organization, it should be a horizontal, cross-cutting team with the potential to work across all business lines.
Do Not Ignore The Realities Of Regulation And The Softer Side Of AI
Involve someone in the process who understands the regulatory/legal issues with your strategy. My Silicon Valley upbringing tells me that as long as it’s ethical, innovation trumps some regulation. Think about it—companies like Airbnb and Uber/Lyft would never exist if they strictly followed the rules.
There is also a trend in shaping AI & ML to be empathic and explainable. Can we trust it? What if it has a bias or makes weird decisions? Who in your organization is responsible? The morality of AI is a huge subject that is going to become reality in direct relation to how successful you become with the technology. The canonical example is self-driving cars. There will be situations where the AI will need to make a ‘lesser of two evils’ decision. A recent article “What if AI in health care is the next asbestos?” explores some very real ethical and moral dilemmas with regard to AI in healthcare: no doubt healthcare can benefit greatly from machine augmented intelligence, however when does it step over a line? AI can and does craft offers that can be very attractive to humans and we can drive behaviors of consumption that may not be appropriate.
And finally, and very importantly, AI is not a set it and forget it pursuit. AI is a continuous journey.