With AI, baby steps make more sense than moonshots

Competitive pressures are driving companies to adopt AI in greater numbers.  However, they differ wildly on the best rate and mode of adoption.

Many executives worry that if they don’t move fast, current competitors or some new insurgent rival more adept at AI will come along to eat their lunch, according to a recent Accenture study noted in Fortune.  Face it, no one in any industry wants to be 'Ubered'.

Research indicates that the faster businesses adopt large-scale “moonshot” AI projects, the greater their chance of success. A recent survey of about 2,500 executives by MIT Sloan Management Review and Boston Consulting Group found that companies that view AI merely as a new tech tool, analogous to what the spreadsheet was in the 1980s, are more likely to fail than those who take a more dramatic, revolutionary stance and integrate AI throughout their tech and business strategies.

Don’t let the perfect be the enemy of the good

These researchers claim that a bigger, splashier approach requires a significant overhaul of business processes and strategies to boost success rates. Accordingly, they recommend that companies take on large AI projects that focus on boosting revenue versus cutting costs, and that tie the technology into their broader business transformation efforts. But, oh by the way, they also note that the risk associated with such aggressive efforts is higher than with smaller projects.

Other pundits push a more incremental, step-by-step approach to AI. Last year, for example, researchers Thomas Davenport and Rajeev Ronanki looked at 152 corporate AI projects and found that companies that launched “highly ambitious moon shots” were less likely to succeed than those pushing more modest AI projects, according to their Harvard Business Review report.

These smaller “low hanging fruit” efforts are those that enhance, rather than completely rework existing processes. As an example, Davenport and Ronanki cite the case of a research hospital’s “moonshot” attempt to use supposedly revolutionary AI technology to diagnose and treat cancer – a project that failed. But AI, applied to a narrower purpose, allowed the hospital’s IT staff to recommend or book lodging for patients’ families, and to determine which patients needed help paying their bills.

Sure, that’s less dramatic than curing cancer, but it also provided measurable results and did not cost the hospital millions of dollars, which the cancer project did.

The basic argument is: why reinvent the wheel when you can improve it with less risk, angst, and downtime?

A step-by-step approach can bring benefits—say reducing errors and repetitive work in expense reporting because the system flags mistakes or omissions before the report is submitted. Anyone who has sent in a report that bounced back for lack of a receipt or a colleague’s name can appreciate this perk. There is also less need for retraining because workers continue using familiar software that is augmented on occasion by smart “chatbots” when they encounter a problem.

Play small ball

To borrow a baseball analogy, a “small ball” strategy that relies on singles or doubles versus swinging for the fences every time leads to scoring in smaller bunches, and typically with fewer strikeouts.

In another example: Instead of ripping and replacing an existing accounting system with an AI overhaul, a company (or its SaaS provider) can implement modular improvements, like enabling existing software to “recognize” invoices that are due for payment soon and prioritizing those over those that are less time-sensitive. That process ensures the company earns discounts based on payment terms and/or avoids late payment penalties.

Remember - garbage in, garbage out

A caveat - even if a company opts to go slower  or smaller with AI it needs to do some hard advanced prep work.  To apply AI to projects big or small, companies first must make sure that their foundational datasets are cleaned up. In many cases businesses will also need to supplement their internal operational data sets with high-quality outside data—say on demographics, weather, or information about customers and/or partners.

For all the talk about the need for reams of data to feed machine learning systems, it is more important to focus on data quality first. If corporate information is not relevant, accurate, or up to date, the best AI in the world isn’t going to generate terrific results.

But once that critical business data is in tip-top shape, it is better to get started with AI in a more graduated way. It might behoove those who promote moonshot AI to remember that Neil Armstrong’s dramatic moon landing in 1969 was the culmination of a series of 27 NASA missions spanning the Mercury, Gemini, and Apollo programs that took place over more than a decade.

In short, even moonshots require participants to take smaller—albeit very important—incremental steps.  That’s a helpful lesson to remember in the AI era.