Finance professionals apply their technical knowledge to help businesses and stakeholders make better decisions and new technologies support this by providing better and cheaper data to support decision making; generating new insights from data; and freeing up time to focus on more valuable tasks.
When considering artificial intelligence, the technologies that are of most relevance to finance functions are:
Machine learning technologies
Machine learning was defined by Arthur Samuel, who coined the term in 1959, as: “a discipline that uses statistical techniques to allow computers to act without explicit programming”.
Despite portrayals in science-fiction, this doesn’t mean that computers can act of their own accord. It means that algorithms when given certain data are able to train themselves to see patters in that data and predict outcomes based on those patterns.
Machine learning techniques there lend themselves to substantial improvements across all areas of finance. The technology can enable teams to do more with data, as well as to improve existing models and tools. For example:
- Models can be developed to automatically correct data quality or fill data gaps.
- New insights can be provided by systems identifying patterns within large data sets which finance teams previously couldn’t detect.
- Machine learning can be embedded in existing finance models or tools enabling them to self-learn from previous experience and improve their accuracy.
There are lots of different ways that a machine can “learn”:
Cognitive language technologies
These technologies use statistical techniques to analyse, understand and mimic human language, both written and spoken. These technologies help to improve how humans and machines interact by converting human (or “natural”) languages into machine languages and vice-versa.
Cognitive language technologies, include: natural language processing and generation; semantic computing; speech recognition and speech synthesis.
While such technologies can be used individually, they are often combined together to deliver a desired outcome. For example, speech recognition, natural language processing and generation are all crucial for effective digital assistants, such as Apple’s Siri and Google’s Alexa.
In business the most commonly used technologies are natural language processing and natural language generation:
| Natural language processing | Natural language generation |
|
| Data input |
Unstructured data - data that is not organised in a way that a computer can read it. Examples: emails, conversational speech or social media posts. |
Structured data - data that can be read by a computer because it is stored in a consistent format. Examples: spreadsheet or database entries. |
| Analysis driven by | Machine learning |
Rules and goals |
| Cognitive capacity |
Ability to recognise variations in human speech |
Ability to imitate natural variability in human speech |
| How it works |
Using statistical techniques and machine learning, the programme converts natural language into structured data that computers are able to ‘read’. For example: each word in a sentence is tagged to a category of words with similar grammatical properties (noun, verb) based on its relation to the rest of the text. |
Natural language generation is rule based, so humans are required to set goals around what data the computer should look for. One such goal could be: show changes in revenue. The goals have to directly relate to the structured data held in the spreadsheet. The structured data can then be used to identify data that is relevant to pre-set communication goals. |
| Outcome |
Computers can then perform tasks using the restructured data, such as extracting specific information. |
The system is able to identify relevant information and convert it into an unstructured format, such as text that will read as if it were written by a person. |
Computer vision
Computer vision is similar to natural language processing in that it converts unstructured data that a human can understand into structure data that a computer can interpret, but in this case in terms of images.
For finance, the most common application of computer vision is optical character recognition (OCR). This is used to convert unstructured text (that a computer can't otherwise read), such as handwriting, into structured data that a computer can recognise.
In combination with other technologies, such as robotic process automation, finance will be able to explore new opportunities to automate processes where previously they had been limited due to unstructured inputs.
As the wider business begins to adopt computer vision technologies, finance will have access to more data and information to support operation decisions.
Find out more about AI and cognitive computing
Learn more about artificial intelligence and cognitive computing by completing ICAEW and Deloitte's eLearning module. This resource includes more details on these cognitive technologies and their implementation, including examples of how they are being used by finance teams and businesses.
AI and cognitive computing
This eLearning module outlines what cognitive computing technologies are most relevant to finance, alongside explaining common misconceptions about the technology and AI. It also covers the benefits, how to get started and includes case studies.
Get started