Æxciting times: We are launching axite – Data-Driven Content Operations! Book your personal demo ➠
Æxciting times: We are launching axite – Data-Driven Content Operations! Book your personal demo ➠
Financial services offer unique challenges for software companies such as the
importance of security and the need for any advice given to be compliant.
This means that financial services companies need to ask software vendors
the right questions before deciding which vendor to choose.
This is where the importance of language comes in. As human beings, we communicate and explain data through language, and we use language to explain complex information and recommend actions. A small minority of the population are completely data literate, that’s why there is such a premium on data scientists. In fact, the US alone has a shortage of 150,000 data scientists according to a LinkedIn Study , so it’s clear 4 that companies can’t afford to put data scientists in every business unit. The only answer to the talent shortage and the need to create a data driven enterprise is to automate the analysis of data and explanation of that analysis. But if you want to quickly explain data and what actions to take in response to data, you need to use language. To put it in technical terms, Human being’s “user interface” is language. NLG technology is critical as we move forward because it is the only technology that turns data into human language.
It’s important to define what we mean by Natural Language Generation and how this differs from a template. Anyone who’s used Mailchimp is familiar with a template. For example a template could say the following “ Dear <>, I wanted to email you about <> and your marketing needs.” A template is a simple and low-cost way to generate simple descriptive text that doesn’t change based on variables, that is to say this sentence only changes by adding a name and a company name. Some so-called NLG vendors try to market template software, calling it NLG and charging thousands for content that could be generated with a simple template. Natural Language Generation works differently. NLG software is programmed with the rules of grammar of the languages in which the software writes. This means that the text completely changes as the data changes behind the text. Enterprise NLG Software companies also include some sort of analysis or reasoning process before the text is written. The end result is that true NLG software can generate summaries of data as well as explanations and even suggestions of the next best steps. These abilities mean that NLG software can write narrative based on complex data sets and following complex rules. This ability is critical in financial services where the narrative is based on complex data sets, which often come from different locations.
Whenever you talk about a software in the AI space, you need to take a moment to separate the myth from the reality.
The truth is that Natural Language Generation needs to be configured for each use case you have in mind. There is no magic Plug and Play NLG. Some vendors will sell preconfigured application but those were still configured and if you want those applications to use your business’s means of analysis, or your business’s language and way of writing, then these off the shelf applications will need to be rebuilt and reconfigured.
Machine Learning is one of the biggest buzzwords of the day, and you can see this in the marketing of certain NLG companies who say “yes, of course we do Machine Learning, we do Language Understanding, we do it all”. The truth is that the most advanced NLG solutions actually employ a hybrid approach. That is to say they use machine learning to suggest data analysis rules, which a human user approves of rejects. This hybrid approach speeds up development time but ensures that whatever text is written is fully traceable. Traceability is critical so that you can show WHY advice is given and so that the software can explain WHY as well. Purely Machine Learning driven NLG is what we call a “black box” technology. This means that advanced algorithms analyse the data and reach a conclusion for a reason that the user doesn’t understand and which the software can’t explain in the written text.
You will need your NLG application integrated into your tech
stack. There will also be work to do to connect your data
sets to the NLG software. Some NLG vendors are much
more advanced than others when it comes to connecting to