Our 100th episode, one host down but Chain of Thought Reasoning top of mind

🚀 The Digital Restaurant Podcast - 100th Episode Special! 🎉

Welcome to a milestone episode of The Digital Restaurant! This is our 100th episode, and while it's a moment of celebration, it also comes with a bittersweet note—Meredith is off to Seattle to take on an exciting EVP role at Starbucks. We’ll miss her insights, but the show must go on!

In this episode, Carl Orsbourn takes the mic solo to explore the next evolution of restaurant and retail tech: moving beyond SaaS into holistic reasoning and AI-driven orchestration.

Key Topics Discussed:

✅ The Downside of SaaS & Siloed Systems
SaaS unlocked incredible capabilities for restaurants and retailers, but it also created fragmented systems that hinder efficiency. APIs were meant to solve this, yet they often fail to integrate seamlessly.

✅ The Need for Data Unification
APIs, when implemented, are usually linear and siloed. Carl explains why data unification is essential—moving from fragmented systems to a connected web of insights.

✅ AI Agents & Chain of Thought Reasoning
Businesses can’t rely on isolated software. The future lies in AI-driven models that adapt and learn. Carl explores Chain of Thought (CoT) Reasoning, which threads insights across departments, bridging silos and improving decision-making.

✅ What Amazon’s 2002 API Mandate Teaches Us
Carl discusses Jeff Bezos’ API mandate, which forced Amazon to break down silos and create an interconnected platform. This move helped Amazon become an AI-driven powerhouse—a model businesses should consider.

✅ Real-World Use Cases
AI-driven orchestration can transform demand forecasting, merchandising, supply chains, and personalization. Imagine:

• Localized marketing & staffing auto-adjusted in real-time.
• Retail shelving resets triggered by AI-detected sales patterns.
• Hotels personalizing guest experiences dynamically.

✅ The Urgency of Fixing Data Quality First
Carl shares a key quote from Invisible’s CEO:
"When great AI meets bad data, the data always wins."
With only 8% of AI models making it to production, clean data, model training, and continuous learning are essential.

🔮 The Future: From SaaS to Model-Driven Innovation
Companies waiting for SaaS integrations will be left behind. The winners will be those shifting toward AI agents that dynamically align decisions with business goals.

Transcript

Hi everyone, welcome to a slightly different version of the digital restaurant. First of all, you'll notice my partner in crime is nowhere to be found and I am thrilled. Why? Well, if you haven't heard the news Meredith has got a new job. She is heading up to Seattle to take on an EVP role. And basically is leading all of Starbucks development activity for the chain.

I am absolutely thrilled for Meredith. Obviously being part of a public organization, taking time into things like this are typically not done. And it's obviously a shame for the listener base here on the Digital Restaurant that you're not going to be able to get Meredith's insights anymore.

But that doesn't necessarily mean the show can't go on. So first request. I am on the hunt for guest [00:01:00] hosts. That's right. If you are in the restaurant environment yourself, if you're in the tech space supporting restaurants, how about you coming on and joining me on future episodes of The Digital Restaurant, where we talk about the top headlines affecting the world of restaurants, retailers, consumer orientated industries, and the way in which technology and perhaps in my context, AI is affecting them.

So if you are interested in that, please reach out to me. I'd love to hear from you. Today, though, I wanted to talk about something I've just written an article about recently, which is the journey of how we're moving on from SaaS and software and how that's leading us towards this world of what I'm calling holistic reasoning.

So the reality is, is today's industry challenges, regardless of whether that for you is transaction decline, competitive density, whether it's about the rising costs that continue to add more and more complexity. The sassification of our industry has been [00:02:00] part of what's helped us address and tackle these issues, right?

It has indeed, over these last few years, unlocked incredible capabilities and it's promising, I think it's fair to say, operational transformation. Yet, it's also created some big, big challenges. And the challenges are perhaps best exemplified by Introducing them almost like a silos. What do I mean by that?

Well, APIs, these abilities for different pieces of technology, different pieces of software or SaaS to talk to each other, have not necessarily always worked. In fact, I think they have failed to align systems holistically across their departments or their goals. Now, it isn't just the systems themselves, it's their inability to integrate with one another seamlessly.

And in the article, I talk about like a data record in one system, mirroring the fields of another, but on the same end, they're perhaps not talking [00:03:00] about the same thing. So what does that, what do I mean by that? It could be a a name of a restaurant, site one, two, three, four, five is represented in one particular system, but in another system, it's the name of the location, right?

Like Miller's End or something, which is Those two things don't correlate with each other. And that therefore means, then, APIs Who are there, of course, to try and solve those issues, they often require customization and building or customizing those APIs can create long engineering queues and the biggest technology companies typically have the most prominent biggest clients, the larger chains, and usually those larger chains take longer to make decisions or to invest in the necessary infrastructure to enable them.

APIs to work. What that ends up happening is you get slower progress across the board. And because those custom APIs are just happening at the bigger chain level, what ends up happening is the overall interface between different forms of technology [00:04:00] becomes slower as well. But even when they are implemented, they're inherently limited.

They typically are like an A to B type of interface, right? They allow one system to talk to another. But the communication is linear and it's siloed. It's like being at a networking event where you can only speak to one person at a time. You know, not really a particularly good way to collaborate or communicate.

You know, the analogy I use in the report that I wrote is, imagine you're in the company's mail room. You know, you're responsible for tagging and forwarding all the different letters to different departments. And of course, the mail room is definitely essential in any organization. But it's not connected to the phone calls coming in, the emails coming in, the chatbots where employees and customers and vendors are talking daily.

And it's really in a similar way of thinking about, like, the mail room. SAS has certainly made it easier for us to be able to have a mail room where things are scanned, or call centers where calls are coming in and redirected, or CRMs and chatbots which are operating in a way to allow people [00:05:00] to be able to communicate effectively in different interfaces and sometimes with automation.

But if those tools are operating in silos What can happen is you lead to inefficient outcomes and certainly disjointed workflows where the organization isn't working together in a coherent fashion. So look, how do we solve for this? The way I, I think it gets solved for is by not thinking of technology as isolated systems, but as a web, a connected ecosystem that seamlessly strings data together, regardless of the system that we're talking about.

The line here that I I appreciate Michael from, from Oh my goodness, I forgot Michael Luganov's company. Anyway, AI company he's working in is all around this idea of data unification being the bedrock of progress. Data unification is the bedrock of progress. So unless we can find ways to unify our data, it's going to be really challenging for us to progress our technological capabilities.

Now, some businesses are approaching this by saying, you know what, [00:06:00] we just need a single holistic platform that just serves as a one stop solution. It helps me with our procurement challenges. And I've read what they've said. They said they can do all these different things and therefore that's what I need.

But for many, this approach isn't flexible enough to accommodate their unique needs. You know, the Microsoft CEO recently said that he thinks SAS It's going to evolve into something completely different. He, he described SAS applications as CRUD databases with a business logic layer. And I guess really what I'm reading into that is, if we built models or agents instead of relying on systems, what I mean by that is, if we built what we're actually after to happen, as opposed to taking a software off the shelf per se, where you actually have to adjust your business around.

The limitations and functionality in the ways in which that particular software package works, then perhaps we could get to a place where our data can be more unified. Now, these models that can learn and adapt [00:07:00] as data is being entered in could really enable you to have a game changer, of course. It can be proprietary then to your own business.

By training these models, businesses could replace outdated processes and make smarter decisions that could improve efficiency. And certainly when you think about what Nadella said, you know, he says if you can logically rapidly migrate to AI agents that are capable of reasoning and adapting and orchestrating real time operations, then why wouldn't you?

Now these AI agents, I'm sure all of you are hearing lots of things about, can move businesses really from this kind of rigid way of thinking to more dynamic, interconnected ways of thinking. And the term that's often used here is called chain of thought, or sometimes for short, C. O. T., reasoning. And of course this is all connected into the wider world of AI and how it's threading insights across multiple departments.

Almost think of data inputs from different areas of the business, which once connected, can [00:08:00] actually lead to better decisions. So when Chain of Thought Reasoning connects workflows then to objectives, you actually get to what are the most core important things. For the organization. What do I mean by that?

Well, we all know that when one department has one set of objectives, and another department led by a different leader has a slightly different set of objectives, even though they're part of the same organization, you get into things like departmental politics, or certainly just delaying tactics. If you've got chain of thought reasoning in place, then actually you can move past that because it's actually taken into account all the various different factors.

And it's trying to say, well, actually, if you want the entire business to move forward to the goals that you set out, this is the right path forward. And that therefore is a really exciting future because that gives you an opportunity to have AI transforming these silos into networks of actionable insights.

Businesses no longer operate as a collections of disconnected systems in this world. Well, in this way, they [00:09:00] become ecosystems that dynamically align every decision across the overarching goals of the business. But look, none of that works. Actually, none of that works unless there's a conversation happening about data accessibility, data cleaning, and data tagging.

And these are fund foundational elements of building systems that can basically reason effectively and to adapt. So it doesn't matter if you have. The best AI team, the best AI models built, these challenges are going to remain unless you can solve for data. You know, our new CEO at Invisible was on a town hall with us recently.

And he had this line which I love. He said, When great AI meets bad data, the data always wins. When great AI meets bad data, the data always wins. And that's so true. You know, only 8 percent of AI models That I've basically been produced to try things out. Ever make it actually into production. As in, actually utilize for their initial purpose.[00:10:00] 

Which really reflects how crucial it is to ensure that data and the focus on getting the data right up front is critical before going too deep into developing of models yourselves. That is what we're doing at Invisible, by the way. It's where we are supporting certainly at least 80 percent of the foundational models with evals and supporting them.

To be able to make sure that any types of deviations that are affecting the quality output of that model can be easily and quickly identified and rectified so that the model gets a better result. So anyway, the point around all of this is if you can demonstrate bottom line impacts of these types of focuses, whether that be into the data cleaning or the data tagging or certainly into the AI and the generative AI model development.

You've got an opportunity to do that, but you've got to commit to a data strategy that involves committing to model development versus continued investment [00:11:00] into SaaS off the shelf solutions. And I think most personally, it's about recognizing the need, you know, organizations and their leaders need to recognize the need to improve data quality and expected outputs generated from things like intelligent orchestration.

No model out there is going to be able to perform how you expect it without attention. on what I think are three key factors. Clean tag data, model training and evaluation, and then continuous learning, which is the idea of AI agents evolving with changing inputs and retraining on that new data to adapt to the internal shifts that are happening in the organization, but also the market trends out there as well.

Anyway, let's move on. One of the things that drew my attention, I'm reading this great book right now I'll have to put in the comments. But it was about the memo that Jeff Bezos of Amazon sent out in 2002. It was called the API, the famous API mandate memo. In that, he wrote, All teams will henceforth expose their data and functionality through service interfaces.[00:12:00] 

Teams must communicate with each other through these interfaces. There will be no other form of inter process communication allowed. No direct linking, no direct reads of another team's data store, no shared memory model, no backdoors whatsoever. The only communication allowed is via service interface calls over the network.

It doesn't matter what technology they use. All service interfaces, without exception, must be designed from the ground up to be externalizable. That is to say, the team must plan and design to be able to expose the interface to developers in the outside world. No exceptions. Anyone who doesn't do this will be fined.

Thank you. Have a nice day, Jeff. Wow, right? Now, Amazon has transitioned into this kind of AI driven powerhouse back in 2002. Offers a blueprint, I think, for what organizations need to think about today. You know, before that infamous memo, and all the transformation that's happened that's turned Amazon into the Goliath that it is today.

They face the same challenges as many other firms. They have silos in its [00:13:00] organization, silos in data and silos in technology. Bezos recognized the need for radical re architecting of its technology and organization and his vision was built around this idea of having a centralized, software driven platform that integrated options, aggregated data, and enabled advanced AI applications.

And of course, it wasn't simple. This took many, many years. He hired a, a big kernel leader from Microsoft called Brian Valentine to lead the redesign. And they ended up having this platform called Santana, which was a modular, standardized architecture that was supported by agile teams. And that created this standard foundation, which really kind of.

So we were able to attack the silos that existed within the organization and ensure that data was accessible across every application used within the within the company. And that allowed things like Alexa to come about and also a number of other things in relation to Amazon's own recommendation engine that I'm sure each of us use on a daily basis today.

Now look, [00:14:00] this idea orchestration really is very transformative, right? True transformation lies in the interdependability of systems and decisions. So if you have these AI driven agents employing chain of thought reasoning, which of course bridging these departmental silos, that can enable businesses to reason across data sets, goals, and workflows.

And that interconnected intelligent. allows businesses to align marketing campaigns with operational capabilities, ensuring better seamless execution. Let's talk about some very clear ways as to how this could happen. Imagine tying the customer feedback that typically comes in either through a marketing team or an operations team and how that is collated to help inform Product design, staffing, and maybe even supply chain decisions around distribution.

Imagine how that data could then be used to predict and adapt demand fluctuations, which of course will improve efficiency as well. [00:15:00] You know, when it comes to restaurants or any retail orientated company or CPG, this really plays into three key areas. Collaborative demand forecasting, you know, no longer will sales be sales data be analyzed in isolation.

It's going to be integrated with feedback from marketing campaigns, distributed performance and operational bottlenecks. And that will provide holistic forecast to optimize things like production schedules or to minimize waste. This will conflects into the way in which marketing. are thinking about the levels of social media spend and which geographies that spend should be applied to when a particular product is released.

That can affect the operations team in the way in which the labor schedules are influenced for cooks in the kitchen, and it can also provide real time updates to the suppliers of your organization to ensure that stockouts never happen again. For marketeers, localised merchandising could also be impacted here.

Imagine being able to take individual sales patterns based on the time of year, identifying the buying personas and the [00:16:00] patterns that produce individualised panograms by each location that update not on a fixed quarterly schedule, but based on forecasted margin uplift anticipated. against the cost of labor to undertake those resets in the retail stores.

You know schedules and task planners can be automatically cascaded to local store GMs and then prioritized against all the other tasks that they might be having to handle to be able to say if you tackle this particular task here this is going to be better for your bottom line. Upon completion of these tasks, you know, which could be done through yes, some IOT camera scans of newly merchandise shelves that could trigger then the initiation of marketing campaigns to the geo targeted areas that are supported by that very specific store.

Imagine that the reset has happened in a retail store. It's been identified through a camera AI detection system and that then is the thing that triggers the start of a marketing campaign. [00:17:00] Another example, personalized customer experiences. You know, I talk about this a lot in both of the books. But imagine in the hotel context, you've got a frequent business traveler that books a room.

And of course customers checking in with their loyalty app at the start of their experience will then only see the things that are most relevant to them. Marketing will use their preferences for quiet floors and particular amenities to offer curated packages for their trip next time. Vegan travelers won't see meat products on their digital tablet menu presented at the restaurant later.

And if they visit the spa at the hotel, maybe their known allergies are already taken into account with the massage oil that's used. You know, this chain of thought reasoning. As you can hear from the way I've been explaining it is tying together data from different areas and making sure it has its own role and purpose across every department within the organization.

Whether it be housekeeping, operations, the spa team, the restaurant, ensuring the room is prepared accordingly and guiding front desk staff and all the team that are helping [00:18:00] that guest have a great experience in the way that is completely tailored and in a way that supports that guest as best as possible.

And guess what? The guest may never need to know about everything that's happening in the background. Only that their preferences from some trip, maybe some months and years before, have influenced the way in which they think. So, a lot I've thrown at you today. A slightly different kind of episode, as you can tell.

The future of business technology isn't about patching together disconnected systems. It's about reimagining the entire framework. Agentic workflows offer a path forward, enabling systems that don't just respond, but proactively adapt, learn, and collaborate on real time data. That kind of shifts us away from SaaS into a model of an era, I should say, of model driven innovation, where AI works seamlessly across every business touchpoint.

You know, assuming SaaS platforms [00:19:00] are going to eventually talk to each other is a long wait that may never transpire. It's something I've just seen over these last five, six years that it just seems to be a way that isn't worth waiting on anymore. Even these holistic platforms that promise every feature under the sun.

Maybe these companies have acquired technology. But even then, when you really look underneath the cover a little bit, they don't necessarily work well with the current infrastructure. And therefore. That's happening because they're not built to work seamlessly together. So neither solutions like that or SaaS off the shelf platforms or even AI enabled capability is going to work unless focus is placed onto data hygiene and appropriate quality assurance.

The more we focus and our leaders focus on empowering this holistic intelligent orchestration and the decisions it is enabling their companies with, especially in an operating environment, The better we're all going to be in this kind of new paradigm. [00:20:00] Technology becomes less about tools. It becomes more about the partners designed not to add complexity, but to remove it and to address data disconnects in particular is going to be important too.

So you've got to be able to say, look, yes, we want to lean into generative AI capabilities. And I'm sure there's lots of leaders out there listening to this that are saying, yeah, we're doing this. We're thinking about doing this. But until you focus on the data disconnects first, you're always going to be pushing water uphill.

Businesses who embrace this holistic chain of thought reasoning and the data standards required can finally focus on delivering value from technology. And from that, they can then foster the creativity from the efficiencies that that's released upon their people and serve their customers with unparalleled efficiency.

The question isn't whether we'll embrace this future, but it's more about how quickly we can make it a reality. Thanks for listening to this special edition of The Digital Restaurant. I'd love to hear your views [00:21:00] and I'll see you next time. 

The Digital Restaurant Podcast is available for you to follow and subscribe wherever you listen to your podcasts. Watch us, rate us, and subscribe to The Digital Restaurant on YouTube, and follow along on all our social media digital restaurant channels. Thanks for listening.