AI Deployments: Expect (and Embrace) the Unexpected

By Juan Martinez, Senior Writer
October 29, 2018 • 3 minute read

Testing and adapting a moderately successful or unsuccessful AI deployment today puts you in an excellent position to dominate your field when AI use cases become mainstream.

No two Artificial Intelligence (AI) deployments are the same. Each project takes a modest amount of training and tweaking, and sometimes, a project can require a complete overhaul. This process is what allows AI to function as humans do, especially in a workplace scenario. Think about how much training and real-life experience a human requires to master a role. AI is no different (much faster, but not much different). The assumptions your team makes about how AI will perform, the use cases for which you intend to deploy, and the ROI you expect to generate will likely be different compared to what customers and employees see in two years. Don’t let this hold you back from making an AI investment.

When you think of AI today, think of the first-generation iPhone. When it launched in 2007, there were very few actual, practical business use cases. However, as companies invested time and money toward developing business cases, mobile commerce, mobile banking and other applications turned into massive revenue generators. The companies that began testing products and services from the start were the first to reap the rewards of new use cases, while the late responders played catch-up for several years.

In this post, we’ll examine some of the positive and negative events that can occur while testing AI system deployments. We’ll talk about why it’s necessary to remain flexible and open-minded while finalizing your deployment plan, and we’ll tell you what to realistically expect in terms of ROI and cost-savings. The key point to remember is: Testing and adapting a moderately successful or unsuccessful AI deployment today puts you in an excellent position to dominate your field when AI use cases become mainstream.

Be Open-Minded About AI Use Cases

Until you begin training your AI to interact with customers, you won’t be able to say with certainty that your intended use case will be viable, especially if the use case is more complex than password resets or FAQs. If you go into a project with the mindset that you must stick to a strict plan for how, when and why a digital colleague is deployed, you’ll absolutely encounter an unexpected event. In some cases, an unexpected occurrence can mean having to rethink the deployment, or switching to a less complex use case. In the vast majority of use cases, an unforeseen incident means making a few simple tweaks and then moving forward.

In many cases, these unexpected events are positive and revelatory. You’ll learn that a digital colleague is not only good at answering FAQs, but that it can also recommend products, or help finalize orders. However, until you begin to work with an AI system that’s connected to your IT infrastructure, company data and customer-facing solutions, you won’t know how effective an AI system can be.

AI Technology Enhancements Are Accelerating at a Rapid Pace

If you’re sitting on the sidelines waiting for a better AI system than what’s currently on the market, you’re already too far behind your more prepared competitors. Within the testing and development phase of any given use case, engineers may stumble upon a new feature or shortcut to dramatically improve how an AI system interacts with customers, or processes data, or presents information visually. While you’re waiting for this new feature to be presented in a press release, your competitor is testing and perfecting the new enhancements and they’re getting closer and closer to deploying it en masse.

AI Budgeting and Forecasting

Let’s go back to the 2007 iPhone comparison. If you sat back and waited until the ROI on building a mobile app had undoubtedly positive, you would have wasted precious quarters and fiscal years and lost out on the massive ROI spike that occurred when the mobile use cases became mainstream. It’s likely that the forward-moving banks and retailers that developed mobile applications for the 2008 iPhone lost money in 2009, and maybe even in 2010, but by 2011 the uptick in mobile commerce spending was massive. During the fourth quarter of 2010, mobile commerce accounted for just 2.4% of all digital commerce spending in the United States. By Q4 2011 that number increased almost 400% to 8.8%. Today, mobile commerce represents a quarter of all digital commerce spending, which amounts to a 1200% increase compared with 2011. Those companies that stumbled and failed in 2008 are undoubtedly the ones who were prepared for the 2011 spike and they haven’t looked back since.

For AI, 2007 is now. How far along on the maturity curve will your company be when the next 400% increase happens? Don’t let strict projections stop you from running an AI pilot or small production. You’ll be surprised. You may even be mildly frustrated. But in the future, you’ll be viewed as an innovator.

Previous Next

A Beginner’s Guide to Conversational AI

Crossing the bridge between digital assistants/chatbots and real Conversational AI requires a fuller understanding of how the technology works and its potential business value.

In our latest white paper, A Beginner's Guide to Conversational AI, we explore these subjects for companies pursuing a near- or long-term technology strategy that includes Conversational AI solutions and Digital Employees.

Download our white paper to learn how to generate business value with Conversational AI.

Download