You may also be interested in:

All Articles

OOPS!

We apologize, but it looks like the article you're trying to find is no longer available.

Below is a list of our most recent, trending articles that cover a variety of treasury and finance topics. If you need help finding what you're looking for, let us know and we'll be happy to help. 


Cutting Through the Hype: Pearson's AI-based Cash Forecasting Marvel

  • By AFP Staff
  • Published: 3/22/2022
Pearson's AI-based Cash Forecasting Marvel

Cash forecasting is top of every financial professional’s mind, particularly as we continue to climb out of the coronavirus pandemic worldwide. In a recent webinar, “Cutting Through the Hype: Pearson’s AI-based Cash Forecasting Marvel,” Nicolas Christiaen, CEO and co-founder of Cashforce  joined James Kelly, senior vice president and group treasurer of Pearson  as Kelly walked us through their journey to an AI-based cash forecasting system. Below is an excerpt from the webinar. 


Pearson’s cash forecasting journey  

One of the key things is that Pearson tried to use software as a service (SaaS), and to keep it as simple and effective as possible. “When I started at Pearson five years ago, I wanted to focus on the area where I felt we could add the most value, which was providing capital structure advice,” said Kelly. This included dividend share buybacks, debt raising as an in-house risk manager and an in-house cash expert.  

The methods they had for analyzing the cash flows weren't good enough for the treasury team to be able to provide a level of expertise to really understand what the key drivers of their cash flows were, where the pain points were, and how to improve working capital; they didn’t have the right tools for that. So, Kelly started the company “on a bit of a journey.” 

One of the treasury team’s original two challenges came in the form of a pretty complex setup. They have 600 bank accounts using 60 different currencies, and 30 businesses all running their own treasury centers and their own cash management operations with limited reporting to the center. “And with that setup, there's a lot of data and lots of different formats and lots of different currencies coming in lots of different ways,” said Kelly. Trying to make sense of it was challenging to say the least.  
When Kelly started, he wanted to centralize a lot of the activity and information while leaving a fair degree of decentralized decision-making. And so, one of his early objectives was to stop asking the operating companies to produce their own forecasts. 

Typically, what they saw was someone, usually an accountant, who would spend a couple of hours a week trying to work out what their cashflow looked like in their particular business. They weren't necessarily responsible for all of the movements in and movements out, so they'd have to talk to lots of other people, try and get some information, and then spend so long working on the process that, whether the data was right or not, they didn't have time to think about it.  

“I wanted to get to a place where we could produce forecasts from the center that would act as a baseline for discussion,” said Kelly. “We could then go back to the operating companies and say, okay, we built a basic forecast for you, now tell us why it's wrong.” 

The automation allows his team to manage their cash balances, produce medium-term forecasts, and understand what's going on in working capital. They make all of the intercompany payments, so they don't have to worry about money moving from one business to another, plus it’s a much better tool for managing FX risk and making investments.  

Kelly wanted to be able to clearly communicate to his team, and to the broader business, what they were hoping to get to, and that boils down to a few basics: visualizing all the key data, having straight-through and automated transactions, and making sure payments are delivered with minimum transaction failure. Those were the fundamental principles Pearson built its transformation on. And in order to get there, they put together a tech stack to support it.  

Pearson has connections to its banks via SWIFT. Those statements and payment files are transferred to Oracle and City Financials, their treasury system. They bring in information from the bank statements, all their trades, etc., and use Cashforce for aggregating their cash management data. “Our treasury management system is really good for point-in-time transactions; it's not very good for time-series analysis,” said Kelly. “What Cashforce gives us is really good, granular time-series analysis and the ability to interrogate that.” 

Pearson also uses the Bloomberg platform for its FX dealing and interest rate swaps, and values them via Mars in Bloomberg. A lot of their end-state group accounting ends up in Oracle, so they export from the treasury management system into Oracle for the accounting side, and extract data out of Oracle and share that with Cashforce to help inform their forecasts. 

How does the process work? Bank statements come in via SWIFT to City Financials. Pearson does its bank reconciliations there, and then imports the statements into CashForce, where they have approximately 150 rules, which effectively slice and dice the statement entries into various categories so they can analyze the types of transactions that look similar to each other. Pearson uses City Financials as a translation engine; they get files in different formats from the banks and convert that to a common data hierarchy using the TMS. Then they import it into Cashforce, which means it only has to deal with one set of rules.  

For some of their non-operating cash flows, things like dividends and mergers and acquisitions, because they're large, unusual and slightly unpredictable, Pearson uses an Excel import that they apply for those few transactions to make sure they’re feeding in the right balances. Cashforce is then used to produce the forecasts and to retrospectively test them using the actuals versus the forecast analysis. Because Pearson has a very clear idea of how a forecast was built, it allows them to get granular as to why the version of the truth they were expecting didn't quite happen. 

“We can get quite granular despite the fact that we're looking at large amounts of data, millions of records, and we can actually get down to invoice level quickly to identify what's gone differently,” said Kelly.  

Prior to the onslaught of COVID-19, Pearson got into a position where they were running things smoothly based on historic information and the artificial intelligence models they use for regular cash flows, things like payroll and AP. Those methods continued to work well through the early stages of the coronavirus pandemic, then it got more challenging on the receipt side. “We suddenly had much larger volumes of refunds. People who had booked their professional assessments for, say the equivalent of the AFP, would say I’m not going to be able to take my test in a month's time, and I’d like my money back for now,” said Kelly. 

What they witnessed at the height of the pandemic were parts of their business that had never closed before had to close for health and safety reasons; they saw refunds, particularly in their professional testing business; some of their customers paid more slowly in order to preserve their own cash; and the treasury team moved from having a single forecast to forecasting ranges — high, medium and low. 

The team used Cashforce to help them quickly identify which of those ranges looked like it was the most likely, to try and narrow things down as fast as possible between the different scenarios, identify where they were, and then use it to go back into their longer-term liquidity forecasting. Because things weren't as bad as they thought they were going to be, Pearson was seeing more outflows in certain areas, which allowed them to rebuild the forecast and test it real time. They could then go back the following day and say, we said it was going to be X, how's that hypothesis playing out? They can now run multiple models using the AI, and machine learning allows them to focus on the areas that need it.  

Where things continue to be relatively predictable, they can use machine learning, and the history and information in their ERP is a good guide as to what's going to happen in the future. “When we get into the unpredictable — such as what if a couple of customers suddenly start to pay us really late, what does that tell us about the broader macro environment? should we be assuming that is a one-off or does that apply to others? — the fact that we can drill down and isolate what's working and what's not working in our models has been really valuable,” said Kelly.

 

Data quality 

A lot of people would say they have data in the company, but the data quality isn't perfect. How good does the data need to be to start? And if you don't have perfect quality data, can you ever start? 

“This was a problem we encountered when we were getting going,” said Kelly. Now they’re in a position where they’re feeling more confident sharing larger amounts of data. Kelly advised that the big piece is about keeping it manageable and being able to keep control. For example, Pearson has around 6 million customer records for one of its U.S. schools. That's a huge amount to control, map and manage. You've got to have a clear idea of how your data is going to fit together, as well as the benefit you're going to get from it.  

Kelly’s team started out working from bank statements, coding them and saying, let's look at our top 30 customers and noting when they're paying us. Let's have a look at some basic areas that are fairly simple to control, things like payroll. We should be able to identify our payroll payments and be able to flow those through.  

Some of this depends on how precise you want to be. The big question is: what is your forecast for? If it's around fundraising, then you are working in the hundreds of millions or the tens of millions. If it's around short-term liquidity, then how big of a buffer are you prepared to work from? How aggressive do you want to be in terms of depositing your future receipts? You need to understand what your endpoint is, and whether that extra data gives you the extra specificity that you want, because there is a cost for hosting data, and there's a cost for analyzing data. Be really clear about what you want and what you hope to get out of it, and work on small data that you can control at the beginning, then scale up and learn from what you do.  

When Kelly’s team set this up, one of the big things they wanted to be able to do is to quickly drill up and down. Pearson now has a five-level hierarchy of cash flows, which they’ve aligned to their group accounting setup for cash flow analysis. This allows them to quickly sort the data into the pictures they want to see, for example, all of California’s cash flows and transfers.  

“You are either importing data that's readily available, or you are getting the system to quickly generate data, or you are pulling in data from other sources you are connected to,” said Kelly. Although it might seem complex in practice, it works well and allows his team to build different models.

Getting buy-in

Ultimately you want everyone on the team to be able to use and benefit from your AI-based cash management system. So the more involved and engaged other members of the team are in the process, the more likely you are to succeed. It also gives you the opportunity to build good relationships. What Kelly and his team found is that the more they asked questions, the more people started to offer “a little report that will give you this [information].”  

You need to keep the team engaged, and you need to keep the broader business engaged. Do it bit by bit, and make sure you connect with the business to make the interaction something that impacts the business overall.  

“If you try and build a business case around ‘this is going to save us interest with low rates,’ that can be difficult,” said Kelly. Framing it in terms of savings in hours provided a much more granular understanding of what's going on for the Pearson team. For example, the way through a large and complex dataset would be very difficult if you had two people working full-time on data mining; they wouldn't be able to do it as well as the Pearson team’s exercise does.  

“I think our CFO would say that they're very happy with the amount that we spent, and we’d do it again in a heartbeat. That said, it’s not an overnight fix; it takes time to get it right,” said Kelly. 


PARTNER CONTENT:

 

Register by June 3 to Save $725
Energy Abounds -- New Ideas and Connections Are Made
There is power in being together. Join your peers at AFP 2022 for connection and impactful learning you can't get through a screen.
Register now.

Copyright © 2022 Association for Financial Professionals, Inc.
All rights reserved.