A practical guide to KBQ-led enterprise AI strategy

When “big data” reached the apex of the technological hype cycle several years ago, one would have been hard-pressed to find an enterprise that wasn’t scrambling to accumulate massive unstructured datasets — whether they needed them or not. The haphazardness of this bandwagoning was arguably a large part of why, in 2015, around 60 percent of big data projects were failing to advance beyond the piloting and experimentation phase.

While big data analytics has subsequently become a fixture of the enterprise landscape, its rocky path to prominence is symptomatic of the ill-conceived approach large companies often take to the adoption of emerging technology. Instead of carefully considering how the tech solution du jour will help solve a distinct business problem, overeager enterprises tend to focus on adopting the solution — in whatever form — as quickly as possible.

As 2018 draws to a close, enterprises are facing growing pressure to engage with a new cutting-edge technology: artificial intelligence (AI). Recent research shows that 10 percent of companies currently use AI, 10 percent plan to start using it within the next 12 months, and another 10 percent plan to start using it within 12 to 24 months. Among companies employing 5,000 or more people, however, these splits fall at 31 percent, 18 percent, and 16 percent, respectively.

In other words, for enterprise-scale organizations, AI is on the brink of arriving in full force. If we are in fact just two years away from nearly two-thirds of enterprises adopting AI in earnest, it’s incumbent upon every enterprise decision-maker to start crafting a strategic plan for AI implementation.

To that end, high-level stakeholders would be well-advised to keep their organizations’ key business questions in mind as they go about drafting their AI implementation roadmaps. Not unlike big data, AI will be most effective not when used for its own sake, but when used to pursue actionable answers to clearly articulated questions.

Putting actionability first

The success of enterprises’ preliminary AI efforts hinges on the kinds of questions they ask. “In which areas do we want to leverage AI to stimulate growth?” “Which of our processes do we want to make more efficient through the use of AI?” “Which key business questions do we hope AI will help us answer?” These — not, “How can we deploy AI faster and/or more extensively than our competitors?” — are the hallmarks of a sophisticated AI implementation strategy.

And while, ordinarily, the viability of a new project depends on both its actionability and its potential business impact, when it comes to AI, it’s the former that’s of the utmost importance. Granted, the higher the business impact, the better, but incremental improvements driven by nascent AI initiatives tend to compound remarkably quickly, and thus should never be discounted.

That said, assessing an AI-powered tech solution’s actionability — even one that has been deliberately chosen in view of a set of clear business questions — requires a comprehensive review of the environment into which it’s being deployed.

Laying the groundwork for successful AI implementation

Most immediately, an AI solution without a steady stream of relevant data is like a luxury vehicle without gasoline: impressive to look at, but not getting anyone anywhere fast. AI’s power rests with its ability to process — and elicit insights from — sprawling structured and unstructured datasets at superhuman speeds, but enterprises that are new to AI and opt to adopt an off-the-shelf tech solution often unwittingly end up with a cutting-edge algorithm and nothing to “feed” it.

As such, wherever possible, enterprise decision-makers should deploy their AI solutions into business contexts that feature rich datasets detailing both the enterprise’s historical and contemporary activity. The former are critical to acclimating a generic solution to a specific enterprise’s needs, whereas the latter are a necessary condition of the incisive predictive analyses for which the most powerful AI solutions are renowned. If these datasets are inconsistent or excessively fragmented, no AI solution — no matter how sophisticated — will be able to produce actionable insights on a consistent basis.

In addition to a deployment vector guided by a precise business question and a steady stream of relevant data, an AI solution needs a pre-existing technological framework in which to exist. The underlying mechanics of a solution — the ways in which it receives data, processes it, and returns some sort of output — are meaningless without an appropriate context in which they can trigger tangible actions.

Consider, for instance, a chatbot. It receives input data (a typed query), runs the data through a natural language processing algorithm, and delivers a sapient response. As impressive and efficiency-driving as this may be, such a solution would amount to nothing in a vacuum. It needs a website on which to “live” and, ideally, an attractive, intuitive user interface to encourage site visitors to actually utilize it.

Though not always so neatly delineated, the need for proper technological contextualization exists everywhere AI solutions are deployed. Whether it takes the form of a natural language processing algorithm, a deep learning algorithm, or a neural network, AI as such must be embedded within a usable tech solution, which itself must be strategically situated with a company’s broader technological landscape.

In short, if the tech solution is a luxury vehicle and the steady data stream is the gasoline, the AI itself is the engine. No matter how much horsepower it boasts, it’s useless without wheels and a chassis, fuel, and a road on which to drive.

The financial services industry leads the way

Thus far, the financial services industry has been running in pole position in the AI adoption race. Over a fifth (21 percent) of financial services companies are currently using AI, and another 29 percent plan to be doing so within two years. Other than healthcare (14 percent) and IT (11 percent), no other industry has a current adoption rate above 8 percent.

In light of the requirements for successful AI deployment outlined above, financial services’ advanced position is ultimately unsurprising. Let us take AI-powered fraud detection as an example.

Banks and credit card companies have access to vast databases cataloging past instances of fraud. By feeding this historical data to machine learning algorithms, financial institutions have been able to level-up their AI solutions extremely quickly, resulting in ultraprecise fraud detection capabilities. What’s more, these institutions are able to gather millions of new data points every day — each transaction a customer completes serves to further refine fraud detection models.

Just as importantly, financial institutions have taken great pains to construct end-to-end technological frameworks that facilitate the jump from insight to intervention. As soon as an algorithm pinpoints an instance of potential fraud, the tech solution in which it’s embedded pushes an alert to the relevant account owner — usually via text message, email, and/or a mobile banking app. The account owner is then given the opportunity to confirm or dispel a suspect purchase’s fraudulence, resolving the issue one way or the other immediately.

Not only does this use case check all the boxes of successful AI deployment, it also happens to have a fairly significant business impact. First and foremost, it enables financial institutions to shut down fraudulent activity faster than was previously possible, which in turn minimizes the institutions’ direct fraud-related losses. Further, fairly or not, today’s consumers expect their bank or credit card company to spot fraud before they themselves do, meaning an effective AI-powered fraud detection solution has the ability to nudge consumers from one institution to another.

A roadmap for the future

For large organizations operating in industries other than financial services, AI-powered fraud detection — and other financially-oriented use cases like determining credit-worthiness — serves as an excellent example of how to deploy an AI solution effectively. For CTOs and other high-level technical stakeholders within such organizations, the challenge is convincing board members and CEOs to resist the cresting hype and instead take a measured, strategic approach to AI implementation.

As AI technology continues to mature, potential use cases will start to emerge in nearly every industry. That said, only those that are actually actionable — i.e. have access to a steady stream of data and an appropriate technological framework — will meaningfully differentiate their adopters from industry baselines.

Ultimately, artificial intelligence is not a self-executing panacea, but a tool. When wielded properly, however, this tool has the power to solve complex business problems that once seemed unsolvable. Activating this power takes a great deal of organizational strategizing and a healthy dose of patience, but these initial growing pains are a small price to pay for the benefits of doing business on the cutting edge.