How the data gets from a careers page into your dashboard

A RoleDemand run has eight fixed stages: collect from the employer's applicant tracking endpoint, track the posting against yesterday's state, clean it into a canonical shape, classify function and seniority, resolve the location, scope the result to your grant, run quality checks, then rebuild and publish the dashboard.

1. Collect

   One adapter per applicant tracking platform calls the employer's own public endpoint and returns the raw postings. Network policy, retries and per-domain politeness live in one place, so no adapter can decide to be impolite on its own.

2. Track

   Each posting is matched against a local lifecycle database, which is what turns a list of open roles into the two facts that matter: what opened since yesterday, and what closed.

3. Clean

   Titles, departments and locations are normalised into one canonical frame with a fixed column list. A run whose columns do not match the schema fails rather than shipping.

4. Classify

   Function and seniority are resolved by a cascade: a language model first, a trained classifier second, hand-written rules last. The fallback order is fixed, and each posting records which of the three decided it. Skills are read out of the description against a per-market vocabulary, so that column fills whether or not the language model ran.

5. Resolve locations

   Location text is exploded into tokens and resolved against a shared gazetteer, so one employer writing "London / Hybrid" and another writing "Greater London" land in the same place.

6. Scope

   Your grant is expanded into an explicit list of modules, sectors and employers. Anything outside it is dropped before the dashboard is built, so it never exists in the file your browser downloads.

7. Check

   A quality report runs over every output: classification coverage, per-employer collection status, circuit-breaker cooldowns, sudden drops to zero, and company-name hygiene. Threshold breaches warn loudly and schema mismatches stop the run.

8. Build and deliver

   The dashboard bundle is rebuilt for your scope and published behind a signed cookie. The insight pack is written from the same cycle's numbers.

Every day. The run is automated and scheduled, so the dashboard your consultants open in the morning reflects what the tracked employers were advertising on the previous cycle rather than whenever someone last remembered to update a spreadsheet.

A cycle produces the dashboard and the insight pack together, so the written commentary is never describing a different day's numbers from the ones on screen.

RoleDemand has 34 applicant tracking adapters, covering Greenhouse, Lever, Workday, Ashby, Teamtailor, SmartRecruiters, Personio, Workable, iCIMS, SAP SuccessFactors and 24 more. All 34 were probed against live public tenants on 6 August 2026 and all 34 returned a valid response.

• Ashbyashby
• BambooHRbamboohr
• Breezy HRbreezy
• CareerPuckcareerpuck
• JSON-LD behind a bot checkcf_jsonld
• Comeetcomeet
• Eightfold AIeightfold
• Gemgem
• Greenhousegreenhouse
• HiBobhibob
• Homerunhomerun
• iCIMSicims
• JazzHRjazzhr
• Jobvitejobvite
• JOIN.comjoin
• Generic JSON-LDjsonld
• Leverlever
• Manatalmanatal
• Oracle Cloud Recruitingoracle_orc
• OrangeHRMorangehrm
• PeopleForcepeopleforce
• Personiopersonio
• Phenom Peoplephenom
• Pinpointpinpoint
• Recruiteerecruitee
• Ripplingrippling
• SmartRecruiterssmartrecruiters
• SAP SuccessFactorssuccessfactors
• Oracle Taleo Business Editiontaleo
• Talexiotalexio
• Teamtailorteamtailor
• Workableworkable
• Workdayworkday
• Zoho Recruitzoho_recruit

An employer on a platform outside this list is still reachable. The two generic readers in it handle any careers page that publishes structured job data of its own, and adding a named platform means adding one file to the pipeline.

The run keeps going. A per-employer circuit breaker takes a failing source out of the rotation instead of letting it stall the cycle, the quality report names the employer and the failure, and repairing the source is part of what the monthly fee covers.

Sources move. A company migrates from one applicant tracking system to another, or renames a board, and the endpoint that worked yesterday returns nothing today. That is the ordinary failure in this business, which is why repairing it is written into the price rather than billed as a change request. See what each plan covers.

Scoping happens when the dashboard is built, not when it is requested. Your grant is expanded into an explicit list of employers, sectors and modules, and anything outside that list is removed before the file is produced, so an employer you are not entitled to see does not exist in the bundle your browser downloads.

The same rule applies to modules. A section of the dashboard your agency has not bought is absent from the build rather than hidden in the interface. The full description of what is collected and what is refused is on the data sourcing and ethics page.

Nothing. You name the market and the employers once, at the start. After that the run is automated, the quality checks are ours, and repairing a source that moves is part of the service.