In this article, I will discuss language technology and robotic process automation, their interaction, and how they can provide business value with three practical examples. The explanation will be in terms that can be easily understood by non-technical business professionals.
Language Technology
Language Technology (LT) is a term used to describe natural language processing and computational linguistics in layman's terms. With a value of 20.98 billion in 2021, according to Fortune Business Insights, language technology (or NLP) is a key technology in the information age. Many of the apps and tools we use on a daily basis are built on this innovation. Language technology has a strong connection with artificial intelligence and is often considered a type of AI. This is because it involves training computational models and creating prediction systems to extract meaningful insights from large amounts of language data, similar to machine learning. Examples of language technology include search engines such as Google, advanced translation services like Deep-L, voice assistants (Siri, Alexa), word prediction in messaging systems on smartphones, dictation and accessibility features, mining customer interactions to improve products, and keeping up with changes in legal text or compliance regulations. This is just the tip of the iceberg.
Robotic Process
Automation Robotic process automation (RPA) is a technique to create, improve and implement software robots based on artificial intelligence that can access, modify and manage digital systems and data. This technology simplifies many manual tasks and gives employees more time to focus on higher-level tasks. By incorporating RPA solutions into projects, companies can further enhance their value streams and achieve a faster return on investment. RPA is an advancement of business process automation (BPA), which primarily relies on pattern-based methods and rule-based systems. By incorporating AI algorithms, BPA capabilities are significantly increased, hence the term RPA is used to describe this evolution. Usually, these automation techniques are applied to structured data, but by also incorporating natural language processing (NLP) methods, RPA is further enhanced to handle any form of raw data, making this field particularly exciting. According to Gartner, the RPA market grew by 38.9% in 2020 to reach a value of 1.9 billion, maintaining its position as the fastest-growing innovation in the enterprise software market. Forrester predicts that it will reach a value of 22.5 billion by 2025. These different projections show the broad agreement on the exponential growth potential of the RPA market.
It's worth mentioning that RPA is often confused with low-code or no-code solutions, but that is not a precise characterization. RPA is not restricted to low-code or no-code scenarios. It can be a process pipeline written in Python, Java, or any other programming language, running on the backend, with automatic task schedulers and triggering software controlling its execution based on specific events. Technically, with enough time and technological maturity, RPA may not even require a GUI or human intervention in some cases, even now.
For more information, check out my previous article, where I also discuss implementing an NLP system for supporting systems engineering tasks, which is another form of RPA that we are implementing at Itemis.
Poised for growth
Currently, RPA solutions are implemented across many industries, with BFSI (banking, financial services & insurance) and healthcare being currently the largest segments for this technology. By now, even the legal sector is benefitting from solutions that grant legal professionals a more streamlined experience during research and (legal) information lookup. By adding NLP & AI methods, real world entities in text documents can be identified and highlighted automatically, even mapped together by letting an algorithm identify the relations between them. This takes a big load off the to-do list of knowledge workers, allowing them to redirect their attention to other, maybe less repetitive tasks. Shortages of available talent pools in some business segments based on country will also benefit from software (and hardware) robots.
Around 80% of the data available online is unstructured data, in other words, textual data written in natural language by human authors. This is a vast array of information. With the proper language technology tooling in place, combined with solid RPA methods, businesses can tap into this data to speed up research and the acquisition of valuable business insights.
There is a subfield of natural language processing called information extraction, which, based on a pipeline-like architecture, applies NLP methods to gather insights from text. This happens in a stepwise process, where initially, linguistic information about the text is predicted with AI methods. These new predictions can then serve as a basis for gathering higher-level semantic features within text, to the point where the text itself can be transformed into a visual mind-map-like graph of interconnected nodes, representing the key insights of the text graphically. In a world where we often cannot deal with the sheer amount of text data, this type of technology can help us represent the information in a much ore digestible format, which can rapidly increase work velocity. Language technology can also be used to quickly summarise the key aspects of a larger text file into a small paragraph in real-time, a task called text summarisation. It can also be used to generate new text (natural language generation) with high grammatical accuracy, which is a technique that is also used in chatbots.
So instead of endlessly reading documents and trying to mentally connect the dots across documents, may that be for research or any other reason, language technology combined with RPA can let you focus on the things that matter most. It is of no surprise that, gradually, more businesses are tapping into the potential of these innovations, and why the market size is gradually increasing higher up across the billions.
Increasing business value with language technology and RPA
Now let’s get to the good part with 3 interesting use-cases.
1. Market & business intelligence
Having a solid marketing strategy and gathering market intelligence is quintessential for a business to create products and services that have enough customer demand and can pose as a solid stream of income for the organisation. But doing market research is known to be a long, involved process. Why not support your marketing team with language technology and RPA services to gather insights across documents and sources in a way no human being possibly could? Sometimes there are important connections between different insights across documents that would go unnoticed when research is done completely manually. By using LT and RPA, these connections can be identified quickly, further enriching the knowledge at the disposal of your marketing team, helping the business by enabling more informed decision-making. By using web scraping techniques, large amounts of data can be gathered, preprocessed and fed through an RPA pipeline to identify market drivers, expert opinions, statistics and even correlate them to measure similarity and dissimilarity between expert opinions and calculated forecasts, for instance.
Just as business intelligence provides critical information on the micro-level, market intelligence can provide critical information on the macro-level. By laying out a good strategic plan for information extraction efforts alongside solid implementation, LT and RPA can enrich your business with a holistic approach for gathering insights and making data-driven decisions. This can greatly increase accuracy of the very outdated SWOT analysis charts (strength, weaknesses, opportunities, and threats) to give an example, by going from SWOT analysis to SWOT analytics.
Note that SWOT analysis is an outdated concept as already mentioned. Organizations often spend much time analyzing their internal drivers (strengths and weaknesses) and therefore often not enough time analyzing external drivers (opportunities and threats) which, in economies marked by high levels of disruption, can be detrimental. With a data-driven approach based on LT and RPA, a system can be used to automate SWOT, which has no inherent proclivity to imbalance internal or external drivers (but can be tuned to focus more on macro-drivers in a volatile market for instance). Also, by having the capacity to parse through large amounts of data within a very short time, SWOT can be reborn, like a proverbial phoenix from the ashes, by replacing the analysis with automation and further extending the SWOT idea by integrating advanced analytics.
2. Resume evaluation
Hiring top candidates for a company is an essential business driver. After the covid-shift to home office and the new post-covid mixed-model that many companies now follow, a lot of the classic recruitment techniques have lost their traction.
Note that at Itemis we support a fully remote model, which is great for people in areas without access to an Itemis office space. So, if you like this model, why don’t you explore our current job postings, and perhaps you’ll stumble upon your dream job.
When dealing with lots of applicants in the hundreds, it takes a lot of time to scour through applicant CVs to select the best ones for an interview. Given possible time constraints, some may have to read quickly through a set of applicants, which may lead to poor decision-making, especially if the reviewer happens to be an engineer or business professional who is usually not involved in making HR decisions. In such a scenario, it would be useful to implement an extraction system that scours through hundreds of applications and extracts top applicants based on techniques found in semantic search engines: This means that, instead of simply looking for connecting textual matches between the job posting and a CV, the system also compares keywords from the job posting to the larger semantic context of the CV. An AI-powered system can also compare keywords from the job posting to synonyms in the CV by utilizing semantic similarity techniques, by also incorporating the larger context of the CV file into the equation. This is a state-of-the-art approach that can be a boon to recruiters through making data-driven decisions, potentially even reducing human bias when the AI model is trained on a neutral, balanced, and representative dataset, while leaving the in-person decisions to the interviewer during & after the meeting. When the dataset has negative speech, then of course the AI model may become biased. Therefore, it is essential to treat datasets as “first-class citizens”, or ideally: Treat your data as a fully-fledged product to enable the right mindset when developing an AI strategy for your business.
After the system filtered candidates, the interview phases can be left to the recruiters for doing the in-person decision making, augmenting HR management with a data-driven approach. This benefits both applicants and HR, striking a powerful balance between manual labor and automation, by empowering HR professionals to focus on tasks that cannot be so easily automated.
In many ways managing datasets is like eating your vegetables at lunch: Some people simply don’t like doing it. You’ll survive, but it will become detrimental to your system when your datasets aren't clean, balanced, shuffled and representative of their respective statistical population.
3. Employee & customer sentiment analysis
Sentiment analysis is a popular and well-established concept in language technology. In short, it uses AI to analyze the sentiment of text input. This may be in terms of a movie review, product review, employee feedback, student evaluation, or political opinion, although this is not an exhaustive list. A sentiment system can help ranking a large assortment of opinions on a scale, such as from a numeric positive scale to a numeric negative scale. The information within these reviews can then be mined to make faster improvements to a product or service (customer satisfaction & retention) or to the internal workings of an organization (employee satisfaction & retention).
Most of the focus of developing such AI systems is still on the for-customer side and less on the for-employee side. For a profit-based business this initially sounds sensible: Development for the sake of customers means potentially more profit, but for employees it nearly always means more cost. However, satisfied employees tend towards boosting the business. By reducing fluctuation and increasing productivity, business-specific ‘incentives to do well’ are augmented with a data-driven approach.
More granularly, it can also help with employee review talks, their skill development or during re-organizing teams. By cross-examining the analytics data produced with such a system, aggregate data models can be created to identify insights otherwise invisible to the human eye, especially within a larger enterprise setting.
Comments