Succeed with Do-It-Yourself AI technology (DIY-AIT)
Reading time: 6 minutes
DIY-AIT can be one of several valuable tactics for delivering superior operating performance in your mainline business.
Apply it carefully, not universally.
Use DIY for tactical, opportunistic projects, not major strategic initiatives.
It’s also very useful for experimentation and learning.
Do not try to emulate Amazon or Google
Most big AI-technology leaders such as Amazon and Google “did it themselves” with AIT to achieve superior operating performance in their mainline businesses. They chose to make AIT available to others, both in open source and similar kits and as fee-for-services cloud-based DIY-AIT services. These DIY services are evidence of a “platform strategy.”
A platform strategy is an approach to entering a market which revolves around the task of allowing platform participants to benefit from the presence of others. (Platform strategy explained here.)
Apple, Amazon, Google, eBay, and Airbnb exemplify firms exploiting platform strategies. So do Nvidia, Microsoft, Intel and thousands of others. Firms with platform strategies aren’t rare. However, it’s hard to make platform strategies work, and it’s harder still for entities building their own platforms on top of the platforms of others to make their own platform a commercial platform success.
Know your business
Under the direction of John Geyer, Metlife’s Chief Innovation Officer and CEO of Metlife Digital Ventures:
Metlife interviews 100-plus leaders across MetLife each year and ask them, “What capabilities would give you strategic advantage?” (Source.)
That strategic advantage should tie to one or more of the specific metrics on which the board is measuring your CEO or divisional head.
Take on AIT risk for the sake of your business, not someone else’s.
The DIY-AIT platform providers are playing a numbers game — as do other platform players like UBER. IBM and Microsoft were probably the first to demonstrate this convincingly in the opening days of the PC era. They attracted huge numbers of startups building applications on top of their OS platform, PC-DOS and MS-DOS. The more, the better, and not all would succeed.
Today’s cloud AIT services leaders are doing the same thing, encouraging everyone to build on their platforms in the hope that some become successful in their own right. With the right cloud AIT services focus, the leaders do not need to bet on creating the best super-star applications of the future (as if they know what those applications are.) All they have to do is attract the right critical mass of others to experiment, pilot, develop, deliver, support, and evolve their best bets to get the commercial flywheel to accelerate, in much the same way the iPhone, iPad, and App Store successes accelerated dramatically between 2007 and 2017.
Are you going to create the best super-star application of the future?
No, not if you’re in a typical enterprise outside of the business of building and selling AIT related services.
DIY-AIT can be a great business, a career, a stopgap, distraction, or, in the worst case, a hobby. Which one will it be for you?
Did you or someone else you know
- Build your own email system in the 1980s or 1990s?
- Modify the kernel of your network operating system in the early days of networking?
- Get so enamored of Codd’s 12 rules of relational databases that you modified your homegrown DBMS to incorporate at least some of the 12 rules in your production IT systems?
If you answered no to all three, you probably avoided disaster by being born too young.
Did you laugh at these questions because you know some people who did one or more of these things?
Robotic Process Automation (RPA)
RPA works. Technologies like RPA typically use:
- Screen scraping to automatically grab input (without required human intervention)
- Low-level natural language processing to parse the scraped content to determine what action to take
- A scripting language to take actions that integrate together two or more applications.
This is a perfect example of a “stopgap” DIY approach. It’s typically far less expensive than reengineering existing applications. At moderate scale, it’s manageable, but, too many scripts piled on top of one another can create an inflexible, brittle and difficult-to-maintain mess.
DIY is Beguiling to some
Benn Hamm, the engineer in the story, could have just brought in an exterminator to get rid of his rats. That would have been too easy. Benn spent several months, collected, annotated, and trained his deep neural network-based system on 23,000 images. If you have a deep inner geek to satisfy, this sounds beguiling! Remember, too, that Benn had more than a dozen coworkers, experts at Amazon, helping him.
No one has commercialized it yet
Assume there was no commercial, DNN based, image processing system to automatically control a cat door and block access to the cat if the cat were carrying a dead or dying small animal in its mouth.
If there were such systems, then Benn could have gone to Consumer Reports, Popular Science, Capterra or Wired to get an analysis of the different products in the market and gone out and bought the one that best met his needs.
Real researchers, engineers or wanna-be’s might be tempted to DIY.
There’s a lot of personal value people can achieve by launching into a DIY project. They may have the skills and want to demonstrate them or they may be seeking an opportunity to acquire them.
Why not DIY?
The larger the organization, the higher level you achieve, the less valuable MacGyver-style skills become.
What business are you in?
In an article on Hackernoon.com Cassie Kozyrkov writes about why businesses fail at machine learning. The most important lesson in her metaphorical story about building a bakery and not an oven is her message “stick to the focus of your business.” The last time I looked, most businesses were not in the business of selling AI technologies or services. Many of the entities encouraging you to DIY with AIT are the business of selling AIT.
AI experimentation and research should not be central missions for most enterprises.
Data scientists perform research in the process of helping business people make informed decisions. Enterprises benefit from this work but don’t confuse that with creation and life cycle management of production-ready, industrial-strength products.
Establish an AI Center of Technical Expertise (AI-CTE), reporting to either the CTO or the Chief Data Scientist
Every enterprise needs a small kernel of experts, conversant in the technologies and applications available in the market: a minimum of 2, more likely 4 to 8 for large enterprises. See “Action Plan” AI-CTE in Is AI needed to cure this existential business crisis? (Part 2) The AI-CTE’s job is not to build systems. It is to evaluate, recommend, and educate.
Explore the many ways AI technologies (AIT) can enter and benefit your enterprise
For most enterprises across the next five years, most AIT will enter embedded inside other goods and services.
Embrace commonplace AIT.
Some of these technologies – such as speech-to-text and language translation – will become so commonplace that we will no longer marvel at them or call them Artificial Intelligence.
Discover new AIT-based capabilities your enterprise systems vendors offer.
Every major enterprise systems vendor is primarily focused on delivering new capabilities based on AIT. Examples include invoice reconciliation and resume classification. Secondarily, the enterprise systems vendors also support DIY efforts with extensible AIT services: they allow or encourage you to build custom functions and applications using their AI services such as Salesforce’s Einstein and SAP’s Leonardo.
In aggregate, enterprises will get higher business returns at lower cost and risk exploiting new, vendor-provided AIT-based capabilities than if they go off and try to build it themselves from scratch.
Push vendors to offer products that meet your needs instead of significant DIY investments.
Step outside your comfort zone of known vendors.
Small innovative firms represent more risk but take on more of the risk of major business breakthroughs
Seek out innovation, particularly among young firms, startups building on technologies provided by cloud AIT providers.
Bound DIY projects with a clear exit strategy
In the early DIY email, networking and RDBMS examples, the early DIY winners were typically losers in the longer run because they stuck with their DIY solutions long after there were far better commercial alternatives.
With AI technologies in 2019, DIY projects either fail to see the light-of-day or they create their own momentum, which blinds the organization to emerging commercial alternatives.
- Cognitive biases play strongly
- Technical debt grows
- The rapid rate of change makes it harder for DIY programs to track commercial progress
AI technologies exist at several levels of abstraction. Pick the most appropriate level of abstraction on which to focus your investments.
Disclosure: The views and opinions in this analysis are my own and do not represent positions or opinions of The Analyst Syndicate. Read more on the Disclosure Policy.