More than Splashy Headlines: How AI Can Drive Real Value for Restaurants (Part 1 of 2)
Where’s the value in AI today?
Fiction writer William Gibson once famously wrote, “The future is already here, it’s just not evenly distributed.”
This is particularly true when it comes to artificial intelligence.
Over the last couple of months, the restaurant world has been abuzz with talk of AI, as ChatGPT has reframed the conversation about what is possible, and also what the potential issues of this technology could be.
But when it comes to the business implications of AI, I believe the discussion has been fairly off-target for two reasons. First of all, the type of AI that ChatGPT represents – generative AI leveraging natural language systems and open source data – is likely to be of relatively low value for most restaurants for quite some time, and second, AI has already been providing significant value for certain types of companies and business functions, and that will only increase as data platforms, processing power, and business strategies become more oriented towards using this technology.
In short, the future of AI is already here. It’s just been distributed unevenly.
So what is AI?
People often equate AI with some kind of future “singularity” where machines become sentient, (and in Hollywood, generally leads to a robot apocalypse). ChatGPT, because it uses the language of humans, and exhibits characteristics that we associate with human personalities, seems to represent this kind of AI, though its founders and most technologists liken it more to an advanced search engine than a future robot overlord.
But AI isn’t a specific technical term. It’s more of a conceptual framework generally defined as a computerized system that exhibits the thinking characteristics of humans. This tautology is problematic for businesses considering AI, particularly when we look at the optimal uses of the technology. Seeking human behavior in our computer systems actually ends up under-valuing many of the very characteristics of computer analytical processing that makes the use of this kind of powerful, scaled rational decisioning valuable to companies.
ChatGPT (and Bing’s subsequent use of the base technology) is deeply flawed as a business tool because of the subjective, relativistic, and inconsistent characteristics of its primary training source: the Internet. This massive trove of human discourse, conversational styles, and open source information, can produce an eerily human-like conversation, but it does little to improve business outcomes for companies that want more accurate insights on its own data, or more responsive action to those insights.
By insisting that artificial intelligence is only authentically achieved when it most accurately mimics human behavior and capabilities, we’re missing the very point of incorporating computers into our business processes: to make faster, better informed, and more intelligible and trackable decisions than humans can, or generally, do.
I would encourage a slightly different framing of the technology, which we prefer to refer to as “adaptive automated decisioning.”
The Two Parts of Adaptive Automated Decisioning
As the term suggests, there are two key parts of this process as we we see it:
Automated Decisioning
In automated decisioning, a system is set up with a task, or set of tasks, for which it is provided structured, contextualized data and an integrated functionality, which it is empowered to activate. So, for example, in one instance of an automated decisioning system that we’ve worked on for the optimal routing of orders, a system can be provided with a delivery guest’s location, driver proximity data, the count of clocked-in employees, and the current volume of orders. With this information, the system can make a decision to which location an order should be sent.
As a single function, the system can make very fast decisions, often in microseconds, on thousands of orders an hour. This can have a significant impact on sales performance, restaurant throughput, and overall guest satisfaction.
But the decision calculation is just the first step in an AI framework. While this baseline decisioning process can produce impressive results at scale, it’s not actually doing the kind of “thinking” that we rely on humans to do. Specifically, it’s not assessing the impact of its actions, and adjusting its parameters to improve the outcome.
That’s where the second part of an adaptive automated decisioning system comes into play.
Optimized Learning
For a system to operate continuously and autonomously over an extended period of time, there has to be a method for the system to evaluate the efficacy of past actions against a target outcome and adjust the parameters of its decisioning. In the above example of the order routing process, there may be parameters that adjust over time. For example, a team may become more efficient at making orders so it should be possible to increase the throttle at a location. In a static model, the system would continue to route orders as if this capacity assumption hadn’t changed.
Additionally, elements like time of day could change the relevance of driver distance because of traffic, or because guest patience varies in importance when it comes to waiting for a quick lunch compared to when they’re waiting for a Saturday evening dinner order.
The learning process is what makes adaptive automated decisioning technology so powerful. The parameters can keep changing with business conditions and the system can prevent decisions from becoming stale. A system that is built around learning can take into account a wide variety of factors in order to get to the best decision, while adjusting over time as changing conditions in those factors provide new outputs.
So where to go from here?
While the long-term value of generative AI systems like ChatGPT are still being identified. Restaurants have a significant opportunity to drive value from this more structured and seemingly mundane form of computer thinking.
In Part 2, my follow-up post, I'm outlining a process for restaurants to activate AI for their companies, and will also highlighted some things to watch out for as we all look at the massive potential of this technology.
