Ordering from Domino’s AI-enabled DRU Assist may still feel too clunky for anyone but the nerdiest of junk food-lovers, but Domino’s is all in. The pizza giant is banking on AI to increase its operational efficiency and revolutionise customer interactions.

They’re not alone in their AI ambitions. In “Reshaping Business with Artificial Intelligence”—a 2017 global report of 3,000 executives, managers, and analysts—MIT found that businesses have tremendous expectations for the operational efficiency AI will generate, even though their own relationship with the technology hasn’t quite lived up to the hype so far.

The report found that around one in five companies surveyed had introduced AI into some of their processes, and one in 20 had extensively built AI into their operations. A little under 40 percent of all companies had an AI strategy in place, which rises to 50 percent with large companies included.

Here are some options to consider for those chasing AI-facilitated operational efficiency.

Experiment enthusiastically—on a small scale

Granted, your organisation’s biggest gains from AI will come from deploying the technology at large scale, but new research from Capgemini details the merits of piloting services that tackle small problems first. Once you establish the benefits at this level, you can then scale up.

Motorcycle dealership Harley-Davidson NYC benefited from this approach in 2016, when it launched a promotional campaign to clear excess stock called, “48 Bikes in 48 Hours.” It trialled the autonomous marketing platform Albert to test how effective it was at getting to know its audience and meeting customer conversion goals. Albert can recognise top-performing ad concepts, uncover behavioural patterns that trigger specific actions, predict optimal pricing, and optimise keywords. The result for Harley-Davidson was phenomenal, doubling the company’s all-time sales records in one weekend. Unsurprisingly, the bike retailer has bumped up its commitment to the platform tenfold and also established a call centre to handle the deluge of generated leads.

The Capgemini report also highlights that a significant number of businesses are stepping over low-hanging fruit to engage in overly ambitious “moonshooting.” Low-hanging fruit, in the context of the report, refers to AI that’s relatively low in complexity and cost—but high in benefits. Some common examples include the adoption of virtual assistants or chatbots, consumer behaviour analysis, and product and service recommendations.

Aim to solve specific problems

AI is the closest thing businesses have experienced to an operational efficiency silver bullet. As any techno-utopian who corners you at a party will explain in detail, you can utilise this tech to increase sales, improve customer satisfaction, and produce company insights. Its functions include machine learning, image and video analysis, speech recognition, natural language processing, and even automation in the form of self-healing—like secure printers that have the ability to detect an attack proactively, shut down immediately, and restart to a previous, uninfected version of the BIOS.

But IT specialists need to ask not what AI can do, but what AI should do for their organisation. Before getting carried away with the shiny machine learning toys on offer, think about what pain point your organisation is trying to address for its customers. Only then, you should proceed to considering if and how any of AI’s present manifestations can improve your organisation.

For example, if you’re a pizza company, you should harness AI to make the experience of ordering and receiving pizzas as easy as possible—rather than spending millions working out how to have a 3D hologram of Tupac Shakur appear and start singing “California Love” whenever customers open their pizza box.

The PwC report “Sizing the Prize” recommends you first question the value of different AI options in achieving business goals and, second, examine your organisation’s capacity for change well before spending big on AI. AI hype has become so overblown that even forward-thinking outfits have felt the need to point out that AI is not a hammer capable of driving home every business-problem nail.

After conceding that AI will be ubiquitous in software products by 2020, Jim Hare, research vice president at Gartner, made the following observation, “AI offers exciting possibilities, but unfortunately, most [software] vendors are focused on the goal of simply building and marketing an AI-based product rather than first identifying needs, potential uses, and the business value to customers.”

Don’t make the same mistake. As obvious as the advice sounds, learn to swim in the shallow end of the pool first, and only after that should you start indulging your most glorious tech fantasies.