In this article, we’ll describe what skills inference can do, and how it can enable the building of skills frameworks, speeding the adoption of a skills-first approach.
A critical component of building a skills-first organisation is the ability to understand, manage, and monitor the skills within your workforce.
Ultimately, organisations need to know the spread of skills across their workforce, common definition and language to understand and describe those skills, and they need to know the proficiency levels of skills within the business. .
With a skills framework that can provide this level of insight, organisations are able to put a number of use cases into action. These include mapping career paths, improving training and performance management, hiring more effectively, and tracking pay progression data for pay transparency legislation.
The challenge for many organisations in reaching this stage has been that surfacing skills from a range of sources, often spread around different teams and in varied formats, has proved to be a barrier. This is where skills inference can make the difference.
One of the biggest challenges for organisations has been the identification of existing skills across the business. 75% of organisations don’t have a centralised skills taxonomy which allows them to at least have a view of skills.
In addition, skills taxonomies are large data-sets of skills ‘tags’. They can provide information of the skills that exist, but lack the detail, such as proficiency, which is required for many use cases.
Another common barrier has been the significant resource, effort, and cost required to build out skills data and the level of descriptions required to support desired use-cases. This includes the complex review process needed to consolidate data from spreadsheets, and to gain review and approval from across the business.
For these reasons, organisations have often struggled to get skills initiatives off the ground, or have only partially achieved their goals.
Skills inference can enable organisations to overcome some of these barriers, reduce costs, and speed the process.
Skill Inference is a method of extracting skills from text, using natural language processing, and more specifically. semantic data analysis - the process of analysing data to extract meaning and insights.
The aim of skills inference is not necessarily to collect data, but to detect skills.
Job data is extracted and deconstructed into component parts, each of which has meaning and can inform the context of the skills.
These component parts are then processed by an NLP pipeline using Token Extraction and Named Entity Recognition techniques.
This job data is then processed to identify skill synonyms and skill concepts. From this the skills are mapped and the collection of skills is classified, with further processing to normalise and classify skills.
The skills data can now be enriched with category classification, so skills are classified according to their type - soft skill, technical, professional skill, and so on.
Classifications also include detecting the level of seniority, industry and occupation. This is classified against recognised industry standards such as ISCO, O*NET and SOC.
These inferred skills are now given a confidence score based on frequency analysis against our job model, co-occurrence patterns (interconnectivity) of groups of skills and historical usage data.
The key to effective skills inference is in the contextual validation and refinement of the data. It’s vital to interpret that skills data to make it meaningful.
Anyone can buy in a skills taxonomy and plug it into a set of data, but this is likely to leave you with an enormous amount of data which is neither meaningful or useful.
It’s important to have confidence that the skills inference has found good skills. This is a challenge, as the inference process is finding every skill it can, which means the initial result is a large data set of skills.
It is also easy to infer poor quality skills data, because the nuance is in understanding the full sentence and the context of the job. For example, some words in skills data, such as ‘support’ may have several applications with different meanings, from help desks to supporting business strategy.
Some of these will be very relevant but sometimes there are multiple variations of skills, meaning you may have some duplication and overlapping of skills - financial analysis and financial data analysis for example.
Skills data may also be too generic to be relevant or to add enough value to be truly useful.
Consistency is also vital. For example, job descriptions may be written in different ways across teams, with varying language used to describe jobs and skills.
This can mean there is inconsistency and a lack of standardisation of skills.
To potential issues around context, consistency, and standardisation, there is a process of training the data to decide which skills are right and wrong.
Skills inference is really sensitive to the way the content has been written, while not all job descriptions reference the actual skills required, this is assumed by the context of the job and its level.
What’s required is a benchmark showing what good looks like, so that the skills data you have surfaced can be tested and improved to produce valuable and accurate data.
For this, external data is used to benchmark the role and suggest additional skills based on other similar roles, or skills that are more relevant in the market.
This means you can be confident about the data, knowing that the skills data surfaced has been validated, compared against external benchmarks, and accurately describes the range of skills across the business.
RoleMapper's Skills Innovation Partnership can help to fast-track the shift to skills by co-creating and building innovative AI and technology solutions to support people strategy and overcome process challenges.
Role Mapper Technologies Ltd
Kings Wharf, Exeter
United Kingdom
© 2025 RoleMapper. All rights reserved.
