Skip to main content
This page is the instructions for use for JobFit. In AI Act terms, PastaHR is the provider and your organisation is the deployer. It covers what JobFit is intended for, how it works, where its limits are, and what you need to do to use it responsibly.

​
Overview

The table shows where you can find the information required by Article 13(3) on this page.

​
Provider

JobFit is developed and operated by: Pasta Software Studio AG
Brauerstrasse 4
8004 Zurich
Switzerland For information on security and data protection, see trust.pastahr.com.

​
Intended purpose

JobFit is a decision-support tool for the initial screening phase of recruitment. It evaluates applications against a set of job-relevant factors, such as education, relevant work experience, language skills or certifications, so that recruiters can identify promising candidates in a large applicant pool more quickly and prioritise their review. JobFit is not designed to make hiring decisions:
  • JobFit never rejects a candidate automatically. No JobFit result, including No Fit, can trigger an automatic rejection.
  • The final decision on every application is always made by a person in your organisation.
  • All JobFit results are labelled as AI-generated and non-binding in the user interface.

​
Foreseeable misuse

The following uses go against JobFit’s intended purpose. For each one, you can see how PastaHR limits the risk and what your organisation should do.

​
How JobFit works

​
Factors

JobFit is configured per job profile or Flow. The assessment is based on a list of factors:
  • Fixed factors can be defined for a profile and are included in the assessment of every job connected to that profile.
  • Conditional rules can add, remove, refine or change the importance of factors for special cases, for example depending on the role, location or workload.
  • If no fixed factors are configured, the AI derives relevant factors from the job information, taking any configured rules into account.
  • Authorised users can review, change, remove or add factors at any time.
JobFit is controlled through the Flow associated with a job. For applications submitted through PastaHR, JobFit is enabled or disabled in that Flow. For applications received directly through your ATS, the job is linked to a PastaHR Flow using the designated ATS field, and JobFit must be enabled in the linked Flow.

​
Evaluation of each factor

Each factor is evaluated independently by a specialised prompt. The result for each factor is one of three outcomes:
  1. Applies: the candidate meets the factor.
  2. Does not apply: the candidate does not meet the factor.
  3. Cannot be evaluated: the available information is insufficient, unclear or contradictory.
For every factor, JobFit provides an explanation for the result and cites the relevant passages from the candidate’s information where available.

​
Combining the results

The individual factor results are combined using fixed, deterministic rules. The AI does not decide how the results are combined.
  • By default, all factors contribute equally.
  • An unclear factor contributes no points but remains part of the maximum possible score. Missing information can therefore lower the overall result.
  • If more than half of the factors are unclear, the overall result is Uncertain.
Otherwise, the result is classified by the share of the maximum score achieved:

​
Optional weighting

Weighting is an optional feature that is activated for an organisation on request. When it is enabled, factors can be classified as Must-have, Normal or Nice-to-have.
  • The numerical weights are calculated automatically from the composition of the assessment. Users do not enter numerical weights.
  • A higher importance category always outweighs all factors in lower categories combined.
  • A Must-have that does not apply results in a No Fit classification, unless the overall result is Uncertain, and lowers the candidate’s position in JobFit-based ranking. This classification is non-binding and cannot trigger an automatic rejection.
  • An unclear Must-have does not trigger the No Fit rule on its own.

​
Statuses that are not an assessment

  • Error: a technical failure occurred during the evaluation.
  • Pending: the evaluation is missing or outdated, for example after the factors were changed.
These statuses say nothing about a candidate’s suitability.

​
Input data

JobFit evaluates the following information from an application:
  • Answers given during the application, including screening questions
  • Content extracted from uploaded documents such as CVs and cover letters
  • Information on education, work experience, skills and qualifications
Document extraction supports PDF, JPEG and PNG files, including scanned documents. JobFit accepts input in all languages. Explanations are written in the language of the respective factor, while cited source passages remain in their original language. JobFit only evaluates the factors configured for the job. Adding factors or changing their importance does not cause any additional applicant data to be processed. For information on where and how data is stored and processed, see trust.pastahr.com.

​
Accuracy, testing and known limitations

​
Testing and fairness

  • JobFit is continuously evaluated against internal benchmarks that cover the complete workflow, from document extraction to the evaluation of individual factors, the handling of missing or contradictory information and edge cases.
  • The benchmarks are continuously expanded to reflect new scenarios and product developments.
  • Factor evaluation is tested for fairness. JobFit is designed to exclude sensitive attributes such as gender or ethnicity from the evaluation, and regular bias audits are carried out.
  • Applications with little information are specifically tested during development.

​
Known limitations and risks

JobFit can make mistakes. The following risks are known, together with how PastaHR mitigates them and what you can do.
Risk: the AI can evaluate a factor incorrectly, for example by misinterpreting a qualification or overlooking relevant experience.How PastaHR mitigates it:
  • Each factor is evaluated individually, and the results are only combined deterministically.
  • Every evaluation provides a reason and cites the underlying source in the application, where available.
  • Only tested prompts are used, and outputs are continuously monitored for accuracy.
  • When weighting is enabled, the importance category and outcome of every factor remain visible, so you can see whether a Must-have caused the overall classification.
What you can do: check the explanation and cited source for every factor that is critical for your decision.
Risk: information provided by the candidate or extracted from documents can be incomplete, inconsistent or incorrect, which leads to a flawed assessment. Unclear factors count towards the maximum score without adding points, so a low result can mean that information is missing rather than that the candidate is unsuitable.How PastaHR mitigates it:
  • You always have access to the original documents and answers.
  • Data sources are cited in the assessment.
  • The application asks for key information in a structured way, in addition to uploaded documents, so the assessment relies less on documents with unpredictable content.
  • Applications with little information are specifically tested during development.
  • JobFit cannot reject candidates automatically.
What you can do: look at unclear factors and the original application, especially for candidates with a low result, and flag inaccurate assessments with the PastaHR support so they can be used for improvements.
Risk: the AI could unintentionally favour or disadvantage certain groups of candidates because of hidden biases in training data or prompt design.How PastaHR mitigates it:
  • Factor evaluation is tested extensively for fairness before it is made available.
  • Sensitive attributes such as gender or ethnicity are excluded from the evaluation by default.
  • Bias audits are carried out regularly.
What you can do: use only job-relevant factors and report results that seem unfair to PastaHR support.
Risk: if the factors are derived from a job description that doesn’t capture the role’s real requirements, strong candidates can receive a lower result.How PastaHR mitigates it:
  • You can define fixed factors at profile level instead of relying only on the job description.
  • Conditional rules adapt the factors to specific roles, locations or workloads.
  • You can review and adjust the factors and their importance at any time.
  • Assessments remain supportive and cannot reject candidates automatically.
What you can do: review the factors of each job before you start using JobFit for it.
Risk: a factor can be wrongly set as a Must-have or otherwise given too much influence, so suitable candidates receive a lower result or a No Fit classification.How PastaHR mitigates it:
  • Importance categories are configured explicitly and are visible to your users.
  • Numerical weights are calculated deterministically, not generated by the AI.
  • The category, outcome, explanation and cited evidence are visible for every factor, and the configuration can be corrected at any time.
  • A No Fit classification is non-binding and cannot reject a candidate automatically.
What you can do: use Must-haves only for genuinely mandatory requirements, such as a required work permit.
Risk: if a provider updates an underlying AI model, its behaviour can change and assessments can become inconsistent.How PastaHR mitigates it: PastaHR only uses tested and validated model versions. Change management and regression testing ensure stability (see Changes and model versions).
Risk: candidates may perceive AI assessments as unfair or opaque, which can reduce their trust in the recruitment process.How PastaHR mitigates it: AI outputs are clearly labelled, and candidates keep their right to contest decisions.What you can do: tell candidates transparently that AI is used to support the screening.
Risk: applications can contain text that tries to influence the AI’s evaluation.How PastaHR mitigates it: JobFit uses technical safeguards, including input sanitisation and system prompts that are difficult for user-provided text to override.

​
Human oversight

JobFit is built so that people stay in control of every decision:
  • No automatic rejections: no JobFit result can reject a candidate.
  • Explanations and sources: every factor result comes with an explanation and, where available, a reference to the source in the application.
  • Full transparency of the result: the outcome, importance category and explanation of every factor remain visible, so you can see which factor led to the overall classification.
  • Clear labelling: all assessments are persistently labelled as AI-generated and non-binding.
  • Configuration control: your organisation decides which factors are evaluated and how important they are.

​
What your organisation needs to do

As the deployer, your organisation is responsible for using JobFit in line with these instructions:
1

Assign trained people

Make sure the people who work with JobFit understand how it works, its limitations and the risk of relying on it too much. PastaHR supports you with training.
2

Review factors before use

Check that the factors for each job are job-relevant, precise and don’t relate to sensitive attributes. Use Must-haves sparingly.
3

Review results, not just scores

Read the factor explanations and check critical findings against the original application before making a decision.
4

Report problems

Report suspicious, inconsistent or wrong results to PastaHR support so they can be investigated.

​
Changes and model versions

  • PastaHR only uses tested and validated versions of the underlying AI models. Changes made by the model providers are not adopted automatically.
  • Model changes go through change management and regression testing against the internal benchmarks before they are released.
  • Changes to the factors of a job do not apply retroactively. Affected results are marked as pending until the applications are re-evaluated.

​
Operation, lifetime and maintenance

  • JobFit is operated as a hosted PastaHR service. You don’t need dedicated AI hardware or a local model installation.
  • PastaHR handles technical maintenance and software updates.
  • JobFit is intended for ongoing operation. No end-of-support date is currently planned. Availability is subject to your agreement with PastaHR.

​
Records and traceability

JobFit records every evaluation so that results can be traced and checked later.

​
Assessment of a single application

Each assessment is stored with the application. It contains the evaluated factors, their importance category, the outcome of each factor, the explanation, the cited sources and the overall result. You can see the full assessment in the JobFit view of the application and download it as a PDF.

​
System logs on request

PastaHR logs all JobFit evaluations of your organisation. On request, PastaHR provides you with an export of these logs. For every evaluation, the logs include:
  • The timestamp of the evaluation
  • The evaluated factors and their importance category
  • The outcome and explanation of each factor
  • The overall result
To request an export, contact support@pastahr.com.

​
Retention

Assessments and logs follow the deletion settings for personal data that your organisation configures in PastaHR. They are deleted together with the application.

AI Screening

Configure JobFit profiles and factors.