The Apply AI podcast explores how Artificial Intelligence and Machine Learning can revolutionise businesses. Solutions that save time and money, increase revenue and future proof their business.

Improving supply chains operations with an AI Co-Pilot: Eamonn Barrett & Remi AI

Remi AI is helping e-commerce retailers, traditional retailers and manufacturers to bring Machine Learning & AI into their supply chain decisions. In this episode, we explore Remi’s path to market, as well as their long term vision for companies of all sizes to utilise an operational AI brain.

Listen below or on your favorite listening service: YouTube | Spotify | Apple

The following summary was generated by Sonnant‘s SEO ready content & keyword transcription. Sign up for a free trial.

Free Sign Up

What is Remi AI and what applications are you currently using it for?

Remi fundamentally helps e-commerce retailers, traditional retailers and manufacturers to bring AI and Machine Learning into the supply chain decisions. Our long term vision is to have an AI brian and every company that’s looking for opportunities to help you make your operation more efficient. Supply chains are the first cab off the rank. The reason being is that it’s very complex, very data-heavy. Lots of decisions involved and a lot of people make those decisions with Excel at moment. So the opportunity is enormous. That is the opportunity to generate value for these businesses. We help businesses with that demand forecasting. We suck in all the data that we possibly can. Be it events, weather, currency fluctuations as well as of course, their internal data. It gives them a strong understanding of what’s coming in future. Once that’s in place, we can start to help with other decisions such as inventory management. What should I purchase? Where should it go in my store network?

So you go from the warehouse to the all the way to the front of house?

And that is, that’s a task that’s as old as retail itself. It’s a fine balance between having enough stock to service the demand versus holding way too much and wasting a whole lot of cash. So it’s an AI playground in terms of the use case. The final area we work in is price optimisation. So that’s helping e-commerce businesses to try to transition away from a static pricing approach. Where once you might raise prices yearly because that’s what inflation is doing it for you’ll be able to take advantage of more short term margin and revenue opportunities.

“Can it be done better than a human” is something we talk about often in AI. In data-heavy verticles – Remi would be far better than a human already?

Yes, taking about forecasting specifically, we sort of cross two dimensions. One of them is certainly we’re aiming for far more accurate forecasts. So the platform’s able to consider a multitude of data streams that – too many data points for one human to consider. But the second is in the automation as well. So if you’re used to doing things in Excel and you’re considering using more data, but it might take you hours to update your Excel sheet because a SQL query might be just massive. So there’s the automation piece as well.

So how good is it now and how much better do you see it becoming in the next couple of years?

The forecasting engine is the most developed part of our platform. It is as close to an industry-standard and market-leading that it can be. The optimisation side of the platform is the younger side of the platform – the inventory management and the price optimisation. So obviously the technology is excellent now, but more data is going to improve it.

We’ve got big plans for where we want to go. The ability to consider more data is certainly one. We have plenty of ideas around how to optimise the performance of the platform. So that it doesn’t take a lengthy amount of time to train the models, to test and deploy. Then it’s about further optimisations on the engine looking for other use cases. So right now we’ve got price optimisation, inventory management and they both operate in a similar way. That’s the forecast and it’s a decision. There are many of these cases in a business where even that very same approach could apply. So manufacturing, for example. It’s a forecast of demand and then we need to make this many people on roster. So in those two simple examples, there is the same approach.

Could you tell us where the rubber hits the road?

The optimisation side of the two key successes that we’ve had was born out of a pretty lengthy history. We also consult on technology & AI. We’ve delivered a lot of optimisation pieces around predictive maintenance. But the inventory management, in particular, is just such a juicy use case here. You’ve got a complex problem, where you’ve got conflicting constraints – that is not wanting to have too much inventory without running out. You can optimise, say, one product isolation. However, it’s when you start to consider multiple locations and products at once. Having them share information between the locations, products and more. That’s when you start to really think – this is pretty cool.

What does your sweet spot look like in terms of customers? Does it look like having multiple locations or a requirement of size?

We primarily work with,  mid-cap businesses. That is, 50 million to 1-2 billion in revenue. We found that it’s that size business where we can typically bring a large step change in operations. They’re probably pretty fast-growing and may have outgrown their current systems. We can come in and just really take charge as far as their forecasting and inventory management. You can buy any one service on its own, then all of their data is acquired in the system and training those models begins.

What human capital investment does a company need to make to get this up and running?

It is enterprise software. So it’s not quite a point and shoot and it depends on a great deal of things. One of them is the services required. Forecasting, for example, is far easier to implement than inventory management. It’s just that it’s a whole lot simpler. On the client-side, where the data is coming from, can speed up or slow down the process as well. If they have one of the well-known systems that we’ve integrated, they’re far easy to work with. So on average, one in three months. And in the grand scheme of enterprise software, that’s really quick.

Is there an element of Machine Learning pertinent to per client that improves their optimisation and an aggregated level that you can pull in? Insights from cohorts across the industry?

When a customer comes on board, we build up an entire version of their relevant models using as much data that we can add in – such that there’s never any customer data shared between models. However, where we can further customise them is where we can start to add in external data streams that might be relevant. If you’re an eCommerce business in Germany, for example, Oktoberfest is probably a big event for you. Having adopted a relevant data stream that for events such as Oktoberfest, we can adjust for seasonal factors. The new Machine Learning way – we always knew that was a spike there, but we didn’t quite know why.

And as far as the customers you’ve got, how quickly do they generally see a return on investment?

It depends on the size of the business. For example, if you’re a large business that is turning over five or 10 billion dollars in revenue – you’re probably losing 10 to 20 million per month of opportunities. At the smaller end or a business that hasn’t seen much growth since they started, it may take more time. The sooner they have access to all these data points, however, and start recording it over a longer period of time, the bigger that benefit is going to be. We can be up and running for a few thousand dollars a month. It’s almost a no brainer, isn’t it?