By Yurii Buchchenko

Restaurants often fail to understand the infrastructural challenges rooted in adopting sound AI practices and effectively integrating the technology that it encompasses. Choosing the right infrastructure for your restaurant chain is critical to successfully integrate AI technologies.

To begin understanding what you need to do, you need to think in terms of how you can get AI technologies properly functioning. Consider these three parts of the problem:

1. Assess your infrastructure: What do you have, and what do you need? A thorough infrastructure assessment needs to gather all relevant information to ensuring proper AI functionality across your locations.

2. AI technology: What technologies or approaches are you integrating? Voice ordering, predictive analytics, or supply-related algorithms/technologies will demand different things from your infrastructure. These needs dictate you will need to upgrade your infrastructure or reconsider what you want to implement.

3. Ensuring functionality: When you have assessed your infrastructure and whether it accords with the AI technology you want to adopt, you can begin to figure out whether the current infrastructure can support the operation of the specific AI technology.

Many restaurants often start implementing AI technologies without understanding their infrastructure or the specific requirements of the AI technology. It’s entirely possible that the current infrastructure meets the requirements, and there is no need to change anything—no new servers to buy, no configurations, deployments, etc. This would be a cost-effective approach, but it’s rare.

Adapting your infrastructure vs. purchasing new resources

What do you do if your current infrastructure doesn’t support your requirements to adopt your desired AI technologies? Then you must consider purchasing and installing new resources, or adapting the AI technology to the current resources. Purchasing and installing new resources, believe it or not, is generally cheaper and simpler. Adapting AI technology to the current resources might not even be possible, depending on the chosen technology and how much control you have over it.

If the AI technology is proprietary, you’ll need a special contract with the provider, and they will handle the adaptation for you. If the AI technology is being developed in-house, then you already likely understand that AI development is not simple. Adapting it to specific hardware is even more complex—you’ll need specialized AI skills and experience, and a regular Data Science engineer won’t be enough. You’ll need a specialized machine learning (ML) engineer with hardware expertise.

What to consider when purchasing and implementing new infrastructure for AI adoption

In essence, it’s better to move toward purchasing and installing hardware. But not all restaurants and chains are the same. Here’s what you should consider when purchasing and implementing new infrastructure to support AI technologies.

Choosing the right cloud solutions

To stay flexible, it’s best to prioritize network connectivity. That way, you can ensure good AI performance regardless of the size, number, or configuration of restaurants. From there, all complexity can be shifted to the cloud.

When investing in cloud computing, it’s critical to understand the main risk. That is, if your network connection fails, your business stops at that single point of failure. While this doesn’t happen often, recovery is very quick, which is why restaurants often prefer this approach. The beauty of cloud solutions is that you can use extremely powerful hardware, with maintenance delegated to the cloud provider. AI technologies tend to require this kind of hardware.

Choosing a cloud provider is exceedingly complex, with many nuances that seasoned development teams can help restaurant companies assess to determine the best choice

The main criteria that companies should look for in cloud solutions as it relates to AI technologies are accuracy and precision, followed by performance and speed. Why accuracy and precision? Because the key difference between traditional software and AI is that AI always operates within a certain level of accuracy. Traditional software will always say “2+2=4,” whereas AI will say, “there’s a 99.999% chance it’s 4.” A simple example: AI voice ordering might recognize native American English accents with 99% accuracy, but might only recognize 10% of native Spanish speakers’ English.

Choosing local hardware instead of cloud solutions

To determine if local hardware is the right approach, you have to consider whether some of your locations will have limited access to stable internet connections. In the case of, say, remote areas in villages or deserts, only good local hardware will do. AI brings unique hardware requirements to restaurant servers, specifically the need for a good GPU chip. This is rare in traditional restaurant configurations. That means you will likely need to upgrade all your servers.

Maintaining and repairing network connections is cheaper and simpler. But replacing a server is not just a physical act. It requires reconfiguring the operating system, installing all software components that connect to other devices (cash registers, monitors, kiosks, etc.), reconfiguring the network, and setting up storage.

To add to this complexity, many activities vital to network function won’t work immediately. They will require multiple iterations of testing and tuning. It’s also likely that the server will contain components from different suppliers, meaning one vendor won’t know all the nuances and will have to consult the technical team of another vendor, which may not be readily available. And if that’s not enough, each restaurant location may have different hardware configurations, so you won’t be able to reconfigure all locations using the same template.

With all that said, maintaining local hardware requires specific attention and expertise. And you must have patience and preparation for long delays before migration can be successfully completed. 

Taking a hybrid approach with local hardware and cloud solutions

If you want the highest level of reliability, investing in both local and cloud solutions makes the most sense. For example, your restaurant may primarily rely on an internet connection for cloud connectivity, but if something goes wrong, you have business continuity by falling back on local equipment.

Key stakeholders for infrastructure needs in AI adoption

When assessing what makes the most sense for your business, you will want to involve key stakeholders like the restaurant’s system administrator and the AI technology provider’s technical director. The system administrator can answer questions about the current infrastructure, and the technical director can provide the requirements for the new technology.

Taking a phased delivery approach

Taking a phased approach to selecting and implementing new infrastructure is almost always justified, especially if it’s critical for the business to continue operating uninterrupted. Again, this may seem obvious, but few consider the accompanying complexities. At a minimum, the phased approach implies creating a special orchestrator, both as an additional and temporary software component and as a process. Your software must be able to operate in two modes: without AI technology and with AI technology. It should have the capability to switch seamlessly between both. At most, you need to implement a specialized monitoring and control process, staffed by people with additional responsibilities.

While the phased approach is more expensive and complicated, it is far less risky. Because ultimately your business will need to operate without interruption.

Prioritize the right solutions for your AI goals

Here are some key considerations for upgrading and implementing infrastructure upgrades for AI adoption:

1. Prioritize investing in good internet connectivity and cloud solutions.

2. If ensuring a stable internet connection is difficult, be prepared for significant challenges in upgrading local hardware or consider abandoning AI deployment in those locations altogether.

3. Always use the phased delivery approach and be prepared that delivery won’t happen quickly at all locations.

4. Prioritize systems and approaches that support precision and accuracy as well as reliable performance and speed as it applies to the specific AI technologies you plan to introduce.