Skills intelligence software: what it does, and what it reads

Skills software is sold on one promise: point it at your company and it will tell you what your people can do. Of the 139 products we reviewed, 18 try to work out skills on their own, without anyone typing them in. Most of the rest ask your people or test them, 14 sell data about the job market instead, and 32 publish too little to tell how they work.

Whether a product works skills out for itself, and what it reads when it does, is most of the buying decision, and a product's name will not tell you either. Two products that both call themselves skills intelligence platforms can work in completely different ways: one reads the work your people produce, the other sends out surveys and self-assessments and waits for the answers.

139
products reviewed
18
that work out a skill without being told

Two questions tell them apart: what the system reads, and whether a human has to type the skill in.

What you are buying

Two products wearing one name

There are four ways a skill can get into any of these systems. Of the 139 products, 107 say enough in public to be placed in one of the four or a mix of them, and which one a product uses decides what it costs, how it goes wrong, and who it suits.

Works it out

Reads the work

Reads what your people produced without being asked to, so nobody has to type a skill in.

What it reads
tickets · documents · codesupport cases · meetings
Asks

Someone types it in

Someone in your organisation sets up a skills framework, and people rate themselves against it or their manager rates them. This is the familiar skills matrix.

What it reads
self-rating · manager rating work output
Tests

Someone sits a test

A test the person sits establishes the skill, so you learn how good they are at what that test covers, and nothing beyond it.

What it reads
test scores · simulations anything untested
Doesn't look

Sells market data

It sells data about the wider job market, such as what employers advertise for and what people put on public profiles, rather than a record of what your own staff can do.

What it reads
job adverts · public profiles your staff’s skills

The complication is that at least half of the 18 products that work skills out on their own do not read your people's work. What they describe working from is CVs, job descriptions, HR records or data about the job market.

Nine vendors, not all of them among those 18, set out where their skills data comes from.

Where each vendor’s skills data comes from

9 vendors
  1. Beamery

    “Skills from Beamery AI are inferred skills based on the primary job title.”

  2. Oracle Dynamic Skills

    It draws on person profile, candidate, job requisition, job profile and course details. All of it sits in the HR system, and the only parts employees write are what they say about themselves, such as a career statement, rather than their work. In August 2026 Oracle announced an agent that will work out skills from connected work systems, with no release date given.

  3. Degreed

    A skill gets onto a person’s skills list in four ways: the employee adds it, it is picked up from their CV, the organisation assigns it from a role or skill standard, or Degreed’s AI suggests it.

  4. Nestor

    It offers self-review, manager review and endorsements from colleagues as ways to confirm how good someone is at a skill.

  5. SAP SuccessFactors

    Skills reach a person’s profile from their job role, the learning they work through, completed performance and 360 review forms, development goals and recruiting, and employees and their managers add and rate skills by hand. Suggesting skills from a CV, or from the achievements people note down, needs a separately purchased AI units licence. Everything on that list is an HR record or something written about the person, rather than the work they do.

  6. Neobrain

    Its AI reads CVs, career paths and LinkedIn profiles to build a picture of what a person knows, and drafts a first skills framework from organisation charts, job descriptions and other internal documents. Setting each person’s level, in Neobrain’s words, “is fundamentally human. It involves employee self-assessment and managerial validation.” Its assistant in Microsoft Teams collects feedback from managers and colleagues.

  7. Workday Skills Cloud

    It suggests skills from what is already in Workday: current and past job profiles, job history, completed projects and courses, and public feedback. The employee or an administrator then adds them. It reads no source outside Workday itself, so evidence of skills in tools such as Jira, GitHub or Slack reaches it only if you import it.

  8. Cornerstone

    Its Skills Graph attaches skills to each employee’s job title, using data from over 250 million profiles. A suggested skill does not count towards someone’s level until they confirm it, and levels come from self, manager and peer ratings plus a system estimate based on work history.

  9. Visier

    You can bring in an existing skills inventory, from partners such as TechWolf, Fuel50 and Lightcast or from almost any system or file. Its separate skills tools pick out skills from job adverts, CVs and job titles.

Where the skills come from a job title, as Beamery's do, what you have is a job-title lookup. That is useful, because it tells you what a person in that role probably knows, which makes it a reasonable starting point for a framework and a poor basis for deciding who to move into a role. If you buy expecting a product that reads your people's work and receive a job-title lookup, you find out when somebody asks the system what a particular person can do, and all it can offer is what their job title suggests.

Reading work does not give the full picture either: Microsoft says the skills its People Skills engine works out “aren’t intended to be a comprehensive reflection of a person’s capabilities”. Getting people to confirm the results is a separate problem. In TechWolf's own T‑Mobile case study, where confirming was left to employees and never made compulsory, confirmation stands at around 32% across the company and closer to 70% in back-office functions, with the gap, in the case study’s words, “concentrated in frontline retail and care”. That deployment still worked from HR and learning records and job descriptions, and the case study describes reading work systems directly as the next step. Ask every vendor on your shortlist for the same numbers.

The questions to ask on a first call: what does the system read, and does a human have to type the skill in? Ask for the list of systems it would read in your organisation. If the answer is the HR system, CVs and job descriptions, you're buying something that reads what has been written about your people rather than the work they do. For several of the things companies use skills data for, that may be exactly what you need.
Which kind you need

You may not need software that works skills out for you

Whether any of this matters depends on what you want the data for. On two of these six uses the wrong kind of product hands you an answer that is confidently wrong, while on the other four you do not need software that works out each person's skills.

Move someone into a different role

Person-level

A job-title lookup will tell you what somebody in that role would typically know, which is not the same thing as what this person has done. If the two differ, you may not find out until after the move has been made.

Find who can do X, right now

Person-level

Either inference, meaning software that works skills out for itself, or a self-reported matrix can answer this, provided somebody keeps the matrix current. Both go out of date, but a matrix does it silently, because nothing on the screen changes on the day the information stops being true.

Prove certification currency to an auditor

Attestation

Inference does not help you here, because an auditor is not asking what a system concluded about somebody. They want a dated record with a name against it, and that is what matrix tools are built to keep: AG5, for example, tracks each certificate’s validity and renewals and keeps an audit trail of every action.

Size a skills gap across the org

Role-level

You need the shape of the gap rather than an accurate reading of each individual, and a job-title lookup will give you that. Workday, SAP, Oracle and Cornerstone each sell a skills module that reports on the skills across your workforce, so check what your contract already includes before you buy anything new. Cornerstone, for one, charges extra for the add-on that shows gaps across teams and the whole business.

Target learning by role

Role-level

The same suites use their skills data to recommend learning. If your suite sells its learning module separately, as Oracle and SAP do, you will need that module as well.

Build job architecture or pay bands

Role-level

You are describing roles rather than people at this point, so you do not need to know what each employee can do.

Inference is only worth paying for when the answer has to be about a person rather than a role. Three of those six uses are about roles rather than people, and a fourth needs a dated record with a name against it instead of something a system has worked out.

Availability

What you can buy, at your size

Which kind you need is only half the decision, because what is on sale depends on the size of your organisation.

How skills get in, and who the vendor sells to · 139 products · Sep 2026
How skills get in Enterprise
only
Enterprise
+ mid
Mid‑market Smaller Unclear Total
Software works it out 85023 18
Mixed 34100 8
A test 710302 22
People type them in 5181543 45
Market data only 84011 14
Couldn't determine 178214 32
reads your work asks your people tests them market data only

Of the 21 products aimed mainly at mid-sized companies, 15 ask employees to type their skills in, and among those that say publicly how they work, none works skills out entirely on its own. Most of the products that do work skills out are sold to large organisations: of those 18, eight sell only to enterprises and five more sell to enterprises and mid-sized firms. 365Talents, now Docebo's advanced skills product, is built for organisations that typically have 1,000 employees or more. The smallest customer TechWolf gives a size for has about 5,000 people. Gloat has supported organisations of all sizes, and the smallest company in its case studies has 4,200 employees. Eightfold publishes no price or minimum size, counts Fortune 500 companies among its customers, and has one US federal contract on record: a single-year Air Force contract worth about $330,000.

Two products break that pattern: Talentguide, whose paid plans start at 50 seats, and Microsoft People Skills, which works out skills from people's documents, emails, chats and meetings in Microsoft 365. Working skills out this way costs $0 extra, so far, with Microsoft Copilot or certain Viva licences, and Microsoft sets no minimum number of seats for those, but a standard Microsoft 365 licence on its own does not include it.

Its engine matches each person to a set of skill names, and a company loading its own skills list supplies a code, a name and a short description for each skill. Proficiency levels are not described in Microsoft’s People Skills documentation, so if you need to know how good someone is at a skill, put that question to Microsoft. People can use it to find colleagues by the skills they share on their profiles, and Microsoft says its performance may be limited for staff who spend little time in Microsoft 365, such as frontline and field workers. It also says People Skills is not designed for promotion or pay decisions, so it will not settle who is ready for a role.

The field

The vendors worth knowing about

There is no single ranking of these vendors, because a matrix tool and the automatic kind are not doing the same job.

Reads your people's work

Built to read work

TechWolf publishes no price or minimum and is built for large enterprises. By default, employees confirm their skills through its assistant in Microsoft Teams or Slack, and the results go into Workday, SAP and Visier. It connects to work systems such as Jira, ServiceNow and Teams as well as to HR and learning systems. At T‑Mobile, where 78% of the skills it suggested for jobs were approved, it worked from HR and learning records and job descriptions, so ask which of your own systems it would read. Microsoft People Skills does much the same inside Microsoft 365 if you pay for Copilot or certain Viva licences. If you want the automatic kind, start with these two.

Head to head: TechWolf vs Microsoft People Skills. The decision comes down to whether you need the skills worked out from people’s work to reach your HR system, which TechWolf does and Microsoft People Skills does not.

Talent platforms and suite modules

Check what it reads

These are substantial products, and large employers use several of them for internal mobility and job architecture, but check what each one reads before you trust a skill on someone’s profile. Beamery’s AI, for example, adds skills based on the person’s primary job title. Gloat builds profiles from HR records, uploaded CVs or LinkedIn profiles, learning and recruiting systems and optional ratings, and its agents add skills from the work itself, such as projects and code. If you already own one of the suites, its skills module is the sensible first thing to look at, though the terms differ from one suite to the next, and some charge extra for parts of theirs.

Asks your people · 45 in total

Skills matrix

This is how 15 of the 21 products aimed mainly at mid-sized companies work, and for most mid-market problems it is the right choice. AG5 suits manufacturing and regulated frontline work. Skills Base gives you a straightforward inventory, and MuchSkills builds one from employees rating their own skills, with managers checking the ones that matter most. SkillsTX is built on SFIA, the IT and digital skills framework, for IT and digital staff rather than the whole workforce. Fuel50 began as career-pathing software and now pairs a skills inventory, which starts from employees rating their own levels, with a talent marketplace. TalentGuard has employees rate their own skills, makes a manager confirm or change every rating, and measures readiness against the role each person is aiming at.

Tests people · 22 in total

Assessment

Workera’s tests centre on technical subjects such as machine learning, AI, data science and software engineering, plus data literacy. Its plans start at $40 a user a year, and the US Air Force paid $910,000 for 7,000 Workera licences for one year, or $130 each, so ask Workera what an organisation of your size would pay. iMocha does both: it offers over 10,000 tests across technical, functional and leadership skills, and it also fills in a person’s profile from their CV, certificates, courses, project records, performance records, ratings by the person and their manager, and work tools such as Jira and GitHub. Kahuna handles frontline competency in the places where an auditor will eventually ask to see it.

Money

What it costs

Almost nobody publishes a price, and the few public figures need a second look before you rely on them.

Beamery Not published no price on beamery.com, and no US federal contract on record · checked 14 Sep 2026
Visier Not published no pricing page on visier.com · checked 14 Sep 2026
TalentGuard $1.60/user/day published on talentguard.com, one product line · Sep 2026
Workera contract data US federal contract records · USAspending.gov
Lightcast contract data US federal contract records · USAspending.gov
Oracle Dynamic Skills $0 extra, so far no additional cost for Oracle Global HR customers in Oracle’s 2023 guide, and in its core HR subscription “where applicable” under its current contract terms · Oracle’s guides and service descriptions · checked 14 Sep 2026
Microsoft People Skills $0 extra, so far comes with Copilot or certain Viva licences, not with Microsoft 365 on its own · Microsoft Learn · checked 14 Sep 2026
SAP skills inference Extra licence Needs a separately purchased AI units licence · SAP SuccessFactors admin guide · checked 14 Sep 2026
Workday Skills Cloud Not published switched on automatically for customers on Workday’s Universal Main Service Agreement (UMSA); customers on its Main Service Agreement (MSA) must first sign an Innovation Services order form, and Workday does not say whether that costs extra · checked 14 Sep 2026
Cornerstone add-ons Not published a basic skills library comes with Cornerstone’s learning and talent products, but Skills Architect and People Graph are sold as add-ons · checked 14 Sep 2026
TechWolf Not published no pricing page on techwolf.ai · checked 11 Sep 2026
Directory listings will mislead you here. TalentNeuron shows as “$1.00 per feature per year” on two comparison sites, while a one-year US Department of Commerce contract for ten TalentNeuron users came to $58,827. Treat anything on Capterra or SoftwareAdvice for this category as unverified.

If the skills tools sit inside a suite you already own, check your contract before you assume they are included, because SAP and Cornerstone both charge for parts of theirs. Either way, budget for the time: TechWolf says it takes three to six months before every job and every employee has a confirmed set of skills, and that only two to six weeks of that is technical work; the rest goes on confirming the results and on change management.

Post-purchase

What happens after companies buy

Check the vendor's position as carefully as the product, because since 2024 five of the companies in this category have shut down, been bought, or cut staff.

2025AstrumU shut down after running out of money. It had spent seven years building a way to pick out and verify people’s skills from several sources of data.
Jul 2025Retrain.ai closed and put its platform up for sale after raising $34m.
Jan 2026365Talents was bought by Docebo in a deal that Docebo’s filing values at $61.3m, including later payments that depend on results. Docebo expects it to bring in about $9m of revenue between the sale and the end of 2026.
Mar 2026Gloat let staff go: citing unnamed sources, the Israeli business daily Globes put the cut at 20% and reported that some customers were not renewing contracts for Gloat’s main product, an internal marketplace for jobs and projects. Gloat said fewer people went than Globes reported, but gave no number.
2024 →SkyHive sold its skills technology and team to Cornerstone, and the product now sits inside Cornerstone’s platform as SkyHive by Cornerstone.

There is a quieter kind of failure too: hand-built frameworks can stall, and because a few companies published theirs, you can watch it happen. Etsy published an engineering career ladder in October 2019, accepted two small typo fixes later that month, and has not updated the public version since; it is now archived. More than 20 people made 69 changes to Monzo's public framework, but the last change to the framework text in its main published version was in September 2019. At the Financial Times, the engineering progression framework had lived in a spreadsheet that split into at least six competing versions naming roles that no longer existed; a working group of five FT engineers rebuilt it, released a first trial version roughly fifteen months after starting, and their version is still being maintained seven years on.

Whatever you buy or build, someone has to own it on an ongoing basis. At the FT, one of the engineers from that original working group was still updating the framework in January 2026.

Build versus buy

If you're thinking of building it

The taxonomy, the standard list of skills and occupations that people get matched against, costs nothing: the European Union's ESCO classification publishes 13,939 skills across 28 languages, and the US O*NET database covers 1,016 occupations. Both are free to download and free to use as long as you credit the source, and O*NET is updated quarterly. So when a pitch treats the taxonomy as the hard part, bear in mind that the list itself is something you can download this afternoon.

The hard part is working out who has which skill, and both the research and TechWolf's own rollout put numbers on how hard that is.

50–65%
the scores of the leading methods in a 2025 comparison, counting how many of the right skills they placed in their top five suggestions for sentences from job adverts
Under half
how closely two experts’ choices of skills matched when both labelled the same job advert, averaged across a team of twelve AI specialists
About 32%
the company-wide level of employee confirmation at T‑Mobile, where staff were invited but not required to confirm the skills suggested for them, in a case study TechWolf published itself

When researchers tested GPT-3.5 and GPT-4 on six sets of job-advert data in 2024, the AI models fell well short of older software built for the task on five of the six. Part of the problem is the answer key itself: the researchers behind the top-five scores and the expert comparison say that overlapping skill names make the correct answers hard to pin down. In published studies, trained reviewers agreed well on which words in a job advert named a skill, but when experts had to label the same advert with skills from a standard list, their choices overlapped by less than half.

There is also a legal consequence to weigh: under the EU AI Act, software that infers skills for employment decisions puts a set of obligations on whoever provides it, and building your own moves those from the vendor onto you.

Before you decide to build, compare it with buying: in TechWolf's own T‑Mobile case study, skills were suggested for every employee in under six months.

What building actually looks like, agent by agent, with the data each one needs and the person who decides: The HR Agent Map

Practical

Where to go next

Sources and method 139 products reviewed between 1 and 2 September 2026, with prices and vendor documents re-checked between 11 and 14 September 2026. Vendor documentation and pricing pages were read directly wherever the vendor's site allowed it; several sites blocked access. Contract figures from USAspending.gov federal records. Vendor status from Globes, Calcalist and GeekWire, Docebo's filings and press release, and Cornerstone's own announcement. Career-framework figures counted from the public GitHub repositories, with the Financial Times history taken from an account published by one of the engineers who rebuilt it. Taxonomy figures from the skills and occupation lists that ESCO and O*NET publish. Accuracy figures from three research papers (SkillSpan, 2022; a 2024 study of large language models by Nguyen and colleagues; and the 2025 paper that introduced the Skill-XL test) and from TechWolf's T‑Mobile case study.

For 32 of the 139 products we could not tell from public material how skills get in, so we recorded them as unclear rather than guessing.