Machine learning – should CFOs ‘build’ or ‘buy’?

14th March 2019

In a world of robotic cleaners, digital personal assistants, chatbots and augmented reality, machine learning is coming of age. That doesn’t mean it’s well defined though. Most people think they know what it is, everyone knows they want it, not everyone has it and not everyone can have it.

Well actually that’s not true. Everyone can integrate machine learning into their business, but it does require a reasonable level of digitisation and great data. And the further along the digital pathway you are, the more options you have.




What can you aspire to?

Machine learning is the process of giving computers the ability to learn without being explicitly programmed. It requires enough iterative data for the machine to recognize patterns and learn from them. So, what you can achieve depends on where your business is on its digital journey. If you already have several years’ worth of well-structured and robust (dependable) data, then it is feasible to build your own machine learning into the system.

But the key question is whether it is better for the finance function to develop its own machine learning capability or hang on to the coat-tails of the software vendors who are working urgently to bring solutions to the market.

Build or buy?

The advantages of building your own machine learning capability is that it can be tailored to your organisation’s requirements, but the downside is the steep learning curve, specialist skills required, development and maintenance cost. For data masters, shooting for the moon can be an industry defining decision, but it comes with risks, and costs, that may be prohibitive to some.

Instead it may be better to leverage the capabilities that software vendors can offer. And some of it may already be in your systems. FSN’s 2018 “Innovation Showcase” highlighted the progress of several leading financial software vendors.

For example, BlackLine’s reconciliation software, aimed a high-volume environments, leverages machine learning to reduce the amount of manual intervention in the reconciliation process. In many instances, up to 95% of matches can be reconciled automatically, leaving around 5% to be checked manually. But BlackLine has also collected more than ten years of records of how these exceptions were fixed, and so can teach their machine learning algorithms to discover patterns within the exceptions, raising the automatic matching rate closer to the 100% mark.

BOARD International’s “BEAM”, has been designed with finance and business users in mind, effectively incorporating all the power of predictive analytics into daily business operations and the decision-making process. It automatically analyses any historical series and defines the best algorithm to predict it (and identifies seasonality and outliers) based on what has happened in the past. The machine (BOARD) periodically re-instructs itself based on the results achieved by its predictions, continuously improving the outcome. All of this happens in what BOARD describes as a “Grey” box. This conveys the idea that even though the results are automated it also fully engages human beings.

Anaplan says it too expects to be able to improve the planning process straight out of the box using machine learning from the point of implementation. It is working with a number of ML techniques, including TensorFlow, an open source library of software used for machine learning computation. The company is currently working to produce pilot studies and proof of concepts (POC) for several Fortune 50 customers by understanding the true application of ML in the planning use cases. These POCs are designed to improve planning for specific areas like revenue forecasting, sales predictions, workforce optimization and demand planning.

CCH Tagetik is preparing for the era of machine learning by not only ensuring its Finance Transformation Platform™ has the capacity to manage and support the quantity of data needed to enable machines to learn, but also ensuring the quality of that data by codifying, normalizing and verifying both financial and non-financial information. 

OneStream has found another innovative use of machine learning in the finance function, by enabling user organizations the flexibility to ‘scale with the burst’ or draw more capacity during times of greater processing demand. This is especially important because financial consolidation, budgeting and forecasting are particularly prone to peaks and troughs of demand at month-end, quarter-ends and year-ends. Now, these demand changes are managed as they occur, but machine learning techniques are being introduced by OneStream which, based on usage data, will anticipate these peaks and troughs, and automatically adjust processing capacity to suit each organization's needs for truly autonomous scalability.

The question of whether to ‘build’ or ‘buy’ machine learning into finance systems is moot is many circumstances, because its already baked into some of the modern applications used by businesses today. FSN’s research identifies that around 14% of businesses with more than 10,000 employees are experimenting with machine learning (and artificial intelligence), but whether you ‘build’ or ‘buy’, all organisations should start by ensuring they have centralised, trusted data for the machines to learn from.

By Gary Simon, BSc, FCA, FBCS, CITP

Chief Executive of FSN & Leader of the Modern Finance Forum on LinkedIn with more than 52,000 members