Three things to avoid doing with AI in a recession, and what to do instead

This is a guest post by Shaun McGirr, EMEA RVP of AI Strategy, Dataiku

Some companies see AI as a tool to help them do what they already do at lower cost or faster, while others see AI as something deeply strategic that will help them do what no one else in their industry is doing. As economic volatility makes both types of companies reassess the value of their AI initiatives, and the enterprise platforms supporting them, it is critical to avoid “magical thinking” and focus on building realistic, lasting, and scalable impact.

Believing in AI as a silver bullet is dangerous at the best of times, but when procurement scrutiny on business cases increases in recessionary times, catching, keeping up with, and then surpassing the competition is as much about what not to do. What are these pitfalls, and what to do instead?

Don’t get caught pursuing the wrong set of AI use cases

Being more successful at anything requires you either maintain your current success ratio, but simply do more, or do the same amount in a way that improves the success ratio. Work harder or work smarter — while the former may be sexy, attention-grabbing and easier to sustain in good times, success under economic constraints will require working smarter (i.e., doing more with the same resources).

And the easiest way to do that is to choose not to do certain things. AI-driven survival and success may come down to ensuring you don’t select the wrong use cases, as doing so might break stakeholder support for AI in your organisation.

When first applying AI, businesses often seek the mystical low-hanging fruit: as-yet unoptimised processes, with ample data and plenty of upside from simple use cases. In good economic times, there will be plenty of time to find these, and so simply doing more work can bring success.

But when the call goes out from the boardroom to cut costs and find efficiencies, then improving the success ratio of AI must become the priority, and this should change your calculus.  Sometimes business processes that seem to present a massive opportunity for AI are already highly optimised and there’s no further gain to be had from implementing AI. In many insurance companies, for example, debt collection and claims processing have already been optimised over decades of incremental process change.

In these instances, using AI to slightly improve just one part of a process can be too expensive for the value it generates. Companies in this position should instead think creatively about applying AI to make a big change to less obvious, less optimised processes. This is analogous to any investing: take a portfolio approach that does not rely on a single asset class (low-hanging fruit) generating all the value, instead diversify AI investments across business processes with different risk/reward profiles.

Don’t forget the value of AI goes beyond the use cases

It’s essential to calculate the total value of your AI initiatives, and this goes beyond just the use cases delivered. You need the bottom line of that ratio as well: what did it cost you and was that money well spent? If you’ve got, for example, 40 data scientists, and each AI use case only delivers half a million dollars in benefit, and you only deliver a handful a year, the numbers simply won’t add up!

And in recessionary times, just finding more valuable use cases is unlikely to improve ROI fast enough. Adopting the mentality of small teams, already accustomed to delivering more with less, may be useful. Retailer MandM Direct is instructive: even during the boom times for online retail generated by Covid, its small data team stayed roughly the same size (about five people), yet grew the value of what it delivered. How? Like all enduring growth, the answer was increased productivity through improved ways of working. Doing less of the manual busy work of AI, reusing prior assets, and delivering smaller projects, but faster.

This example shows that beyond just delivering the right use cases, delivering them differently can further recession-proof your AI initiatives and generate lasting business benefits. Whether you adopt a small-team mentality to increase agility, or enlist additional skill sets to ensure your AI experts remain focused on just the toughest problems, remember you are building a capability to deliver use cases, which is valuable in itself.

Don’t let reality escape you: controlling spending is essential

Finally, companies must be realistic about the costs of developing and applying AI in a recession because the “innovation funding” available only a few years ago for purely experimental projects has dried up. Today, you need to prove you can turn innovation into money, scale it, and apply it in different contexts, or even different markets. Every dollar spent must generate value several times over, and this requires imaginative sharing and reuse of your AI investments.

Concretely, this might mean developing one approach to forecasting product-level demand in retail stores and applying it to your entire global business. This will generate more value faster than building separate, tailored models for every market. In turn, this makes the reuse of the same AI assets for supply chain purposes easier to scale, which in turn can help your finance team control costs. Adopting such a reuse-first mentality will reduce initial development costs by helping you avoid over-ambition upfront, and lets you capitalise on those savings many times over.

Reuse can manifest in another way: avoiding the costs of the most expensive kinds of AI models. Cloud computing has made many things very cheap, but the appetite for data of the most sophisticated AI models is increasing faster than cloud costs are decreasing. This means the most impressive AI models available today can only be built by a handful of companies. So if you are not a Silicon Valley giant, think seriously about whether a custom-built image recognition AI is actually your core business, or whether you can borrow someone else’s as a starting point, save 99% of the cost, and deliver value next month.

Data Center
Data Management