SAN FRANCISCO - If you work for an enterprise business, the odds are that every strategy deck you’ve seen in the last 18 months has featured the words ‘transformation’, ‘digitization’, and ‘AI’ in it – probably all on the same slide.

It’s great that global businesses are starting to realize the once-in-a-generation opportunity that AI presents to cut costs, deepen customer relationships, and generate new revenue. The problem, however, is that most companies today think of AI as the rocket fuel for 'moonshots'; very grand, sexy, and ambitious new business initiatives that can take a great deal of influence to green-light and many millions of dollars to launch.

Leaders tend to hatch big ideas, which are often far outside of the company’s core expertise, that overshadow the more immediately impactful, smarter and – let’s face it – boring applications of AI within existing operations that could actually set a company up for life.

Here is how to ensure that you avoid failed moonshots and use AI wisely instead.

Why are AI moonshots foolish?

Let’s define a moonshot with a real launch attempt by a real company.

A large bank noticed that Wealthfront and Betterment were beginning to eat its lunch. The bank decided to compete by creating its own robo-advisor - an intelligent software that makes investment recommendations for consumers.

The bank’s vision was to increase revenues and scale its advisory business using a front-end service made of chatbots, NLP, and significant volumes of research. For the robo-advisor’s technical foundation, the bank selected a very well-known cloud-based AI provider that famously won a game show some years back. (You know the one.) This provider is equally famous for its cost and the number of consultants from its parent company required to get it out of bed.

The project ultimately failed. Why? Because this third-party cloud-based AI provider couldn’t handle graphical information. There was too much variation in the research formats, and the pilot simply ended up not being as good as human advice. The cost of this failed moonshot was not publicly disclosed, but it was likely in the tens of millions of dollars.

The technical failure itself is a teachable moment, but the strategic failure is actually even more important. There is no reason why big banks shouldn’t create robo-advisors, but this bank should have begun its foray into AI with a very clear understanding of the confines of its back office.

Why launch something totally new before cutting half a billion dollars of expense out of high-volume, unstructured processes like customer onboarding, fraud detection, and loan processing? Why not build political capital, enthusiasm, and budgets by going for certain victories? There is a time for taking a big swing with a new technology to drive revenues, and it comes only after demonstrating AI’s impact on the bottom line.

Where should AI be aimed instead?

The smartest companies are aiming their AI-driven transformation efforts at their operations.

According to the US Bureau of Labor Statistics, the financial services industry alone spends $42 billion on manually intensive data work, while the US collectively spends $130 billion on data entry. Gartner sizes the business process outsourcing industry at $250 billion, and PWC estimates the cost of global data processing to $3 trillion. That’s three Apples sunk into pure, avoidable cost.

Does it make any sense that the world is pouring so much money into such boring, repetitive work? Of course not, especially when, according to McKinsey, 51% of the activities performed in the US economy can be automated with currently available technology. Imagine if all the boring tasks that keep people from doing meaningful, value-generating work were done by machines? How much more fulfilling would jobs be? How much happier would customers become? How much faster would companies be able to respond to change?

Take, for example, the insurance industry. Claims handling is the drivetrain of every insurer. According to PwC, up to 80% of insurance premiums are spent on claims payments and handling costs. The claims process begins with what’s called notice of first loss (aka, “I got into an accident!”), percolates with extensive data gathering before payment decisions and concludes with appeals.

During each leg of the claims process, there is unstructured data: call center notes, claims adjuster comments, emails, open-ended customer responses in proprietary messaging platforms and even social media communication. While robotic process automation has been highly effective at moving data into applications once it’s been structured by people, it leaves the high-volume judgment work to agents and adjusters. This is why the smartest insurers use machine learning (ML) - the practical, proven branch of AI - to automate the processing of unstructured data.

The result is significantly lower operational costs, better customer outcomes and an improved work experience for their employees. Is it a sexy application of AI? Not especially, but this boring, 'Everyday AI' is rapidly widening very thin margins in the insurance industry and in other data-intensive industries like banking, healthcare and pharma.

Related: Why Large Enterprises Should Consider AI

How can a company put “Everyday AI” to work?

The business world has been conditioned to believe that implementing artificial intelligence is only possible through the considerable and expensive efforts of data scientists and engineers. This is one of the five common myths of AI.

The core of born-digital companies like Amazon, Facebook and Google are data scientists and engineers, and they’ve built and grown their businesses on an AI foundation. For years, AI has been limited to technology companies who use AI not as an accelerator or a change agent but as a business model.

AI has actually been too complex and expensive for any born-analog company to apply to legacy operations because of the complex refinement required of what had always been a raw resource. Data scientists were essential for cleansing data, selecting machine learning algorithms, applying the data to these algorithms and determining which trained model would perform at the optimal levels of accuracy and completeness.

This is no longer the case. Two relatively new capabilities make AI a refined, everyday resource for born-analog companies.

The first is AutoML. Like the name suggests, AutoML automates the long, complex process of turning ML from an empty algorithm into hardened, reliable automation of virtually any high-volume business process that has unstructured data.

Excellent examples include anti-money laundering (AML) and customer onboarding in banking, patient intake and revenue cycle management in healthcare and pharmacovigilance in pharma, not to mention ubiquitous shared services processes like order-to-cash, accounts payable and employee onboarding.

Rather than making data scientists suffer through the thankless slog of data cleansing and feature selection, AutoML performs this work by identifying patterns in the tasks that people perform everyday, like extracting information from trade settlements or routing customer service emails. You can watch a short video about how AutoML works here.

The second is AutoQC. AutoQC automates the quality control of AI-driven automation by making bots the “makers” and people and other bots “checkers.” One of the problems of ML is knowing when models reach minimally acceptable levels of quality and detecting when a model is incapable of delivering an accurate output.

By statistically checking and validating the output of models, AutoQC not only ensures accuracy but also incrementally improves both the statistical confidence of models and also the emotional confidence of process owners. One of the biggest banks in the world uses the combination of AutoML and AutoQC in its most sensitive operational processes and wanted 100% human validation of automation output for the first month of production. They concluded the “helicopter customer” period with total confidence in AutoQC’s capability to catch and correct errors by using the best of automation and the best of their people.

AutoML and AutoQC combined make AI self-service for business people, and the smartest companies in the world are using AI today to automate thousands of business processes in every industry. By building this foundation, these same companies will be in a much better position to actually pull off their future moonshots. Not to mention that they will also be able to catch and successfully ride the many coming technology waves that AI will create.

Adam Devine is Chief Evangelist and SVP Marketing for WorkFusion, the leading provider of AI-driven process automation. After launching and leading the company’s marketing capability for 6 years, he is now focused on driving market awareness and understanding about intelligent automation by speaking at conferences, leading content creation and press and analyst relations, and working closely with customers and partners to educate their teams and promote their success. Catch WorkFusion’s presentation at The AI Summit San Francisco September 14^th