McKinsey reports in recent research that 56 percent of respondents have adopted AI in at least one function, but there are also examples showing how hard it is to deliver business impact from AI. For example, Gartner reported that 85 percent of ML fail to deliver, and other reports share why deploying ML to production is so hard.

Is failing to bring ML to production a failure, or is it part of the experimentation and learning required for organizations?

I think the latter, that AI is learning and experimental, and here’s why: For many organizations, it will require integrating multiple technologies to deliver long-lasting business outcomes. 

Emerging Tech Requires Integrated Platforms

We only have to look back over the last two decades of web technologies for examples of why business outcomes require the integration of multiple technologies before emerging ones can deliver business impacts – especially for SMBs and many enterprises. Examples:

• Developing websites started in the late 1990s, but it wasn’t until retail businesses leveraged e-commerce, content management systems, and digital marketing practices that online retailing became a growing business channel.
• Enterprises have been storing unused dark data in their data centers for decades, but when analytics (including ML, data visualization, and other data science programs), DataOps (and specifically, the ability to automate third-party data integration), cloud databases (including data lakes and NoSQL stores), and APIs became mainstream, it enabled more businesses to launch and grow commercial data services. 
• While most businesses created mobile-friendly websites and deployed mobile apps, it required a combination of social sign-on capabilities, easy payment integrations, location services, and other phone + cloud services to enable rapid/viral user adoption. 

You can see that it’s not just one technology that drove widescale business adoption and successful business outcomes – especially for mainstream SMBs and for less technical enterprises.

Business Outcomes: AI, IoT, AR/VR, Hyperautomation

So, the success criteria for ML may not be its deployment into production. Success might be better defined as showing today’s ML as a step in the journey of using a business’s proprietary data, analytics, ML, or AI in an integrated, customer-enabling set of products and services. 

This will be true for other emerging technologies. AR/VR needs businesses to develop content, evolve experiences, and create unique value propositions for the emerging tech to evolve to mainstream business capability. Large-scale IoT needed in smart cities and smart buildings need more than just sensors and require architectures that likely include edge computing, 5G, data streaming, and real-time ML.

In my recent episode of 5 minutes with @NYIke, I share three emerging tech integrations that are transitioning out of the emerging categories. SMBs and enterprises should experiment and look for opportunities because, in a few years, laggards may face an uphill battle to catch up.