Complete Intelligence

Categories
News Articles

AI for Supply Chain Forecasting and Proactive Planning

This article originally published at https://www.linkedin.com/pulse/ai-supply-chain-forecasting-cas-milner/ on January 27, 2021. It talks about one of the CFO pain points, which is planning.

 

 

How much confidence do you have in traditional price forecasts for the components of your supply chain? Your answer is probably “not much”, if you have been in business for over a decade — or even just during 2020! But AI can do better — much better — at price forecasting than the standard statistical technique of linear regression most of us learned in college.

 

Complete Intelligence has built a comprehensive platform for making very accurate supply chain ingredient forecasts. The forecasting Saas have done the hard work of aggregating (and cleaning!) billions of data points from many high-quality sources, including import/export trade data, all feeding the AI algorithm engines to produce amazingly accurate predictions. You should follow the postings of Tony Nash , for his economic commentary based on many forecasts for exchange rates, basic commodities, and supply chain components important for world economies and local business operations.

 

Many companies have antiquated, inaccurate processes for forecasting costs in their supply chain. Their standard statistical forecasting is usually done with linear regression – a straight-line projection of historical costs, into the future. But the price behavior of most commodities is not linear, it is non-linear. Artificial intelligence algorithms are especially suited to making accurate forecasts using non-linear data, which is why they are increasingly applied to dynamic financial forecasting.

 

Many industries are especially sensitive to supply costs:

 

  • Manufacturing (electronics, energy equipment, automotive, health supplies, pharmaceuticals, metals, plastics, papers)
  • Extraction operations (oil and gas, forestry, mining)
  • Services (transportation, shipping, hospitality, food and beverage)

 

Supply chain cost planning is a core process, and AI tools are destined to become key ingredients, deeply embedded in operations.  They enable automation of proactive planning and monitoring to digitally transform the organization. The licensing cost for these financial forecasting tools or financial projection software is a small fraction of the operations cost – and potential savings. It is also worth noting that having reliable forecasts of future price trends can create a rational basis for supplier negotiations. Simplify financial planning with AI and machine learning.

 

I’m excited about the AI-driven digital transformation of micro-economic forecasting, and would eagerly discuss the benefits with you.

 

#SupplyChain #AI #EconomicForecasting

Categories
Podcasts

Stories from the Cloud: The Forecast Calls For…

Tony Nash joins veteran journalists Michael Hickins and Barbara Darrow at the Stories from the Cloud podcast to talk about the forecast calls for businesses, and how AI and machine learning can help in predicting the futures in budget forecasting. How does his company Complete Intelligence dramatically improve forecast accuracy of companies suffering from a huge 30% error rate. He also explained the AI technology behind the CI solutions and strategic toolkit, and how this practically applies to global companies. How can they benefit from this new technology to better be prepared in their budget planning and reducing risks in costs?

 

Stories from the Cloud description:

It’s not easy to predict the future. But when it comes to business and cash or financial forecasting tool or software, the right data and the right models are better than any crystal ball.

 

Tony Nash, CEO and founder of Complete Intelligence, explains how AI and the cloud are giving companies better cash forecasting software tools to see into their financial futures.

 

About Stories from the Cloud: Enterprises worldwide are turning to the cloud to help them thrive in an ever-more-competitive environment. In this podcast, veteran journalists Michael Hickins and Barbara Darrow chat with the people behind this massive digital transformation and the effects it has on their work and lives.

 

Show Notes

 

SFC: Hey, everybody, welcome back to Stories from the Cloud sponsored by Oracle. This week, I am here, as always, with Michael Hickins, formerly of The Wall Street Journal. I am Barbara Darrow. And our special guest today is Tony Nash. He’s the founder and CEO of Complete Intelligence. And this is a very interesting company. Tony, thanks for joining us. And can you just tell us a little bit about what the problem is that Complete Intelligence is attacking and who are your typical customers?

 

Tony: Sure. The problem we’re attacking is just really bad forecasting, really bad budget setting, really bad expectation setting within an enterprise environment. Companies have packed away data for the last 15, 20 years, but they’re not really using it effectively. We help people get very precise, very accurate views on costs and revenues over the next 12 to twenty four months so they can plan more precisely and tactically.

 

SFC: It sounds like a big part of the mission here is to clean up… Everybody talks about how great data is and how valuable it is. But I mean, it sounds like there’s a big problem with a lot of people’s data. And I’m wondering if you could give us an example of a company, let’s just say a car maker and what you can help them do in terms of tracking their past costs and forecasting their future costs.

 

Tony: So a lot of the problem that we see, let’s say, with the big auto manufacturer, is they have long-term supply relationships where prices are set, or they’ve had the same vendor for X number of years and they really don’t know if they’re getting a market cost, or they don’t have visibility into what are those upstream costs from that vendor. And so, we take data directly from their ERP system or their supply chain system or e-procurement system and we come up with very specific cost outlooks.

 

We do the same on the sales revenue side. But say for an automaker, a very specific cost outlook for the components and the elements that make up specific products. So we’ll do a bill of material level forecast for people so that they can understand where the cost for that specific product is going.

 

Before I started Complete Intelligence, I ran research for a company called The Economist and I ran Asia consulting for a company called IHS Markit. And my clients would come to me and say, there are two issues at both companies. Two issues. First is the business and financial forecasting tool or even strategic toolkit that people buy off the shelf has a high error rate. The second issue is the forecasts don’t have the level of context and specificity needed for people to actually make decisions. So what do you get? You get very generic data with imprecise forecasts coming in and then you get people building spreadsheets and exclusive models or specific models within even different departments and teams and everything within a company.

 

So there are very inconsistent ways of looking at the world. And so we provide people with a very consistent way and a very low error way of looking at the future trajectory of those costs and of those revenues.

 

SFC: So I’m curious, what is the what is the psychology of better business forecasting software? So on your customers and I’m thinking, if I’m a consumer, so this is maybe not a good analogy, but if I’m a consumer and I look at the actual costs are of a phone that I may have in my pocket, I may think, jeez, why making a thousand dollars for this? But then part of me says, such things mark up and well, I guess so. There are uncertainties in financial projections. So on me, I mean, I don’t need a financial projection software to tell me that the components of the pocket computer have don’t add up to what I paid for them. But I kind of understand that there needs to be money made along the way. I just I want it. Right. How does that translate on a B2B perspective? What are the people’s attitude about price and how do they react to the data that, as you said, I mean, heretofore, it’s kind of been unreliable.And all of a sudden, I think you say a lot of procurement projections have been around 30 percent, which is huge. Right. So how does that happen and how do people react to something that seems more trustworthy?

 

Tony: Well, I think that expectations depend on the level within a manufacturing company that you’re talking to. I think the more senior level somebody is, of course, they want predictability and quality within their supply chain, but they’re also responsible to investors and clients for both quality and cost. And so at a senior level, they would love to be able to take a very data driven approach to what’s going on. The lower you get within a manufacturing organization, this is where some of the softer factors start to come in. It’s also where a lot of the questionable models are put in as well.

 

Very few companies that we talk to actually monitor their internal error rates for their cost and revenue outlooks. So they’ll have a cost business forecasting software model or a revenue forecasting model that they rely on because they’ve used it for a long period of time, but they rarely, if ever, go back and look at the error rates that that model puts out. Because what’s happening is they’re manually adjusting data along the way. They’re not really looking at the model output except for that one time of the year that they’re doing their budget.

 

So there really isn’t accountability for the fairly rudimentary models that manufacturing companies are using today. What we do is we tell on ourselves. We give our clients our error rates every month because we know that no no business forecasting software model is perfect. So we want our clients to know what the error rate is so that they can understand within their decision making processes.

 

SFC: And it’s kind like a margin of error in a political poll?

 

Tony: Yeah, we use what’s called MAPE – mean absolute percent error. Most error calculations. You can game the pluses and minuses. So let’s say you were 10 percent off, 10 percent over last month and 12 percent under this month. OK. If you average those out, that’s one percent error. But if you look at that on an absolute percent error basis, that’s 11 percent error. So we gauge our error on an absolute percent error basis because it doesn’t matter if you’re over under, it’s still error.

 

SFC: Still wrong, right.

 

Tony: Yeah. So we tell on ourselves, to our clients because we’re accountable. We need to model the behavior that we see that those senior executives have with their investors and with their customers, right? An investment banking analyst doesn’t really care that it was a plus and a minus. They just care that it was wrong. And they’re going to hold those shares, that company accountable and they’re going to punish them in public markets.

 

So we want to give those executives much better data to make decisions, more precise decisions with lower error rates so they can get their budgeting right, so they can have the right cash set aside to do their transactions through the year, so they can work with demand plans and put our costs against their say volume, demand plans, those sorts of things.

 

SFC: I have to just ask I mean, Michael alluded to this earlier, but I want to dive into a little more. You had said somewhere else that most companies procurement projections are off by 30 percent. That’s a lot. I mean, I know people aren’t… I mean, how is that even possible?

 

Tony: It’s not a number that we’ve come up with. So first, I need to be clear that that’s not a number that we’ve come up with and that’s not a number that’s published anywhere. That’s a number that we consistently get as feedback from clients and from companies that we’re pitching. So that 30 percent is not our number. It’s a number that we’re told on a regular basis.

 

SFC: When you start pitching a client, obviously there’s a there’s a period where they’re just sort of doing a proof of concept. How long does that typically last before they go? You know what? This is really accurate. This can really help. Let’s go ahead and put this into production.

 

Tony: Well, I think typically, when we when we hit the right person who’s involved in, let’s say, category management or they actually own a PNL or they’re senior on the FPNA side or they’re digital transformation, those guys tend to get it pretty quickly, actually. And they realize there’s really not stuff out there similar to what we’re doing. But for people who observe it, it probably takes three months. So our pilots typically last three months.

 

And after three months, people see side by side how we’re performing and they’re usually convinced, partly because of the specificity of projection data that we can bring to to the table. Whereas maybe within companies they’re doing a say, a higher level look at things. We’re doing a very much a bottom up assessment of where costs will go from a very technical perspective, the types of databases we’re using, they’re structured in a way that those costs add up.

 

And we forecast at the outermost leaf node of, say, a bill material. So uncertainties in financial projections are solved. A bill Of material may have five or 10 or 50 levels. ut we go out to the outermost kind of item within that material level, and then we add those up as the components and the items stack up within that material. Let’s say it’s a mobile phone, you’ll have a screen, you’ll have internal components. You’ll have the case on the outside. All of this stuff, all of those things are subcomponents of a bill of material for that mobile phone.

 

SFC: So I am assuming that there is a big role here in what you’re doing with artificial intelligence, machine learning. But before we ask what that role is, can you talk about what you mean by those terms? Because we get a lot of different definitions and also differentiations between the two. So maybe talk to the normals here.

 

Tony: OK, so I hear a number of people talk about A.I. and they assume that it’s this thinking machine that does everything on its own and doesn’t need any human interaction. That stuff doesn’t exist. That’s called artificial general intelligence. That does not exist today.

 

It was explained to me a few years ago, and this is probably a bit broader than most people are used to, but artificial intelligence from a very broad technical perspective includes everything from a basic mathematical function on upward. When we get into the machine learning aspect of it, that is automated calculations, let’s say, OK. So automated calculations that a machine recognizes patterns over time and builds awareness based on those previous patterns and implies them on future activities, current or future activity.

 

So when we talk about A.I., we’re talking about learning from previous behavior and we’re talking about zero, and this is a key thing to understand, we have zero human intervention in our process. OK, of course, people are involved in the initial programming, that sort of thing. OK, but let’s say we have a platinum forecast that goes into some component that we’re forecasting out for somebody. We’re we’re not looking at the output of that forecast and go, “Hmmm. That doesn’t really look right to me. So I need to fiddle with it a little bit to make sure that it that it kind of looks right to me.” We don’t do that.

 

We don’t have a room of people sitting in somewhere in the Midwest or South Asia or whatever who manually manipulate stuff at all — from the time we download data, validate data, look for anomalies, process, forecast, all that stuff, and then upload — that entire process for us is automated.

 

When I started the company, what I told the team was, I don’t want people changing the forecast output because if we do that, then when we sit and talk to a client and say, hey, we have a forecast model, but then we go in and change it manually, we’re effectively lying to our customers. We’re saying we have a model, but then we’re just changing it on our own.

 

We want true kind of fidelity to what we’re doing. If we tell people we have an automated process, if we tell people we have a model, we really want the output to be model output without people getting involved.

 

So we’ve had a number of unconventional calls that went pretty far against consensus that the machines brought out that we wouldn’t have necessarily put on our own. And to be very honest, some of them were a little bit embarrassing when we put them out, but they ended up being right.

 

In 2019, the US dollar, if you look at, say, January 2019, the US dollar was supposed to continue to depreciate through the rest of the year. This was the consensus view of every currency forecaster out there. And I was speaking on one of the global finance TV stations telling them about our dollar outlook.

 

And I said, “look, you know, our view is that the dollar will stabilize in April, appreciate in May and accelerate in June.” And a global currency strategist literally laughed at me during that interview and said there’s no way that’s going to happen. In fact, that’s exactly what happened. Just sticking with currencies, and for people in manufacturing, we said that the Chinese Yuan, the CNY, the Renminbi would break seven. And I’m sure your listeners don’t necessarily pay attention to currency markets, but would break seven in July of 19. And actually it did in early August. So that was a very big call, non consensus call that we got months and months ahead of time and it would consistently would bear out within our forecast iterations after that. So we do the same in say metals with things like copper or soy or on the ag side.

 

On a monthly basis, on our base platform, we’re forecasting about 800 different items so people can subscribe just to our data subscription. And if they want to look at ag, commodities, metals, precious metals, whatever it is, equities, currencies, we have that as a baseline package subscription we can look at, people can look at. And that’s where we gauge a lot of our error so that we can tell on ourselves and tell clients where we got things right and where we got things wrong.

 

SFC: You know, if I were a client, I would I would ask, like, OK, is that because you were right and everyone else is wrong? Is that because you had more data sources than anyone else, or is it because of your algorithm or is it maybe because of both?

 

Tony: Yes, that would be my answer. We have over 15 billion items in our core platform. We’re running hundreds of millions of calculations whenever we rerun our forecasts. We can rerun a forecast of the entire global economy, which is every economy, every global trade lane, 200 currency pairs, 120 commodities and so on and so forth. We can do that in about forty seven minutes.

 

If somebody comes to us and says, we want to run a simulation to understand what’s going to happen in the global economy, we can introduce that in and we do these hundreds of millions of calculations very, very quickly. And that is important for us, because if one of our manufacturing clients, let’s say, last September, I don’t know if you remember, there was an attack on a Saudi oil refinery, one of the largest refineries in the world, and crude prices spiked by 18 percent in one day.

 

There were a number of companies who wanted to understand the impact of that crude spike on their cost base. They could come into our platform. They could click, they could tell us that they wanted to rerun their cost basis. And within an hour or two, depending on the size of their catalog, we could rerun their entire cost base for their business.

 

SFC: By the way, how dare you imply that our listeners are not forex experts attuned to every slight movement, especially there’s no baseball season. What else are we supposed to do? I wanted to ask you: to what extent is the performance of the cloud that you use, you know, important to the speed with which you can provide people with answers?

 

Tony: It’s very important, actually. Not every cloud provider allows every kind of software to work on their cloud. When we look at Oracle Cloud, for example, having the ability to run Kubernetes is a big deal, having the ability to run different types of database software, these sorts of things are a big deal. And so not all of these tools have been available on all of these clouds all the time. So the performance of the cloud, but also the tools that are allowed on these clouds are very, very important for us as we select cloud providers, but also as we deploy on client cloud. We can deploy our, let’s say, our CostFlow solution or our RevenueFlow solution on client clouds for security reasons or whatever. So we can just spin up an instance there as needed. It’s very important that those cloud providers allow the financial forecasting tools that we need to spin up an instant so that those enterprise clients can have the functionality they need.

 

SFC: So now I’m the one who’s going to insult our readers or listeners rather. For those of us who are not fully conversant on why it’s important to allow Kubernetes. Could you elaborate a little bit about that?

 

Tony: Well, for us, it has a lot to do with the scale of data that’s necessary and the intensity of computation that we need. It’s a specific type of strategic toolkit that we need to just get our work done. And it’s widely accepted and it’s one of the tools that we’ve chosen to use. So, for example, if Oracle didn’t allow that software, which actually it is something that Oracle has worked very hard to get online and allow that software to work there. But it is it is just one of the many tools that we use. But it’s a critical tool for us.

 

SFC: With your specialization being around cost, what have you looked at… Is cost relevant to your business and so on cloud? How so?

 

Tony: Yeah, of course it is. For us, it’s the entry cost, but it’s also the running cost for a cloud solution. And so that’s critically important for us. And not all cloud providers are created equally. So so we have to be very, very mindful of that as we deploy on a cloud for our own internal reasons, but also deploy on a client’s cloud because we want to make sure that they’re getting the most cost effective service and the best performance. Obviously, cost is not the only factor. So we need to help them understand that cost performance tradeoff if we’re going to deploy on their cloud.

 

SFC: Do you see this happening across all industries or just ones where, you know, the sort of national security concerns or food concerns, things that are clearly important in the case of some kind of emergency?

 

Tony: I see it happening maybe not across all industries, but across a lot of industries. So the electronics supply chain, for example, there’s been a lot of movement toward Mexico. You know, in 2018, the US imported more televisions from Mexico than from China for the first time in 20 some years. So those electronics supply chains and the increasing sophistication of those supply chains are moving. So that’s not necessarily sensitive electronics for, say, the Pentagon. That’s just a TV. Right. So we’re seeing things like office equipment, other things. You know, if you look at the top ten goods that the US receives from China, four of them are things like furniture and chairs and these sorts of things which can actually be made in other cheaper locations like Bangladesh or Vietnam and so on. Six of them are directly competitive with Mexico. So PCs, telecom equipment, all these other things.

 

So, you know, I actually think that much of what the US imports will be regionalized. Not all of it, of course, and not immediately. But I think there’s a real drive to reduce supply chain risk coming from boards and Coming from executive teams. And so I think we’ll really start to see that gain momentum really kind of toward the end of 2020 and into early 2021.

 

SFC: That is super interesting. Thank you for joining us. We’re kind of up against time, but I want to thank Tony for being on. I want to do a special shout out to Oracle for startups that works with cool companies like Complete Intelligence. Thanks for joining us. Please try to find Stories from the Cloud at on iTunes or wherever you get your podcasts and tune in again. Thanks, everybody.