This document is a question bank for procurement teams shortlisting an AI-driven assessment vendor such as Selection Lab. Use it to draft RFP and tender questions for hiring assessments. It assumes you are assessing a platform that screens, tests or scores candidates and that you need evidence you can defend in an audit.
Each question carries a priority: M (mandatory, pass or fail with evidence) or D (desirable, scored). Require every answer to cite an appendix. A statement without an attached artefact counts as a claim, not evidence.
Attach three conditions to the bid:
Request one compliance pack containing:
Rationale: a missing item at bid stage is a reliable predictor of a missing item at audit stage.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Which constructs does each assessment measure (soft skills, hard skills, intelligence or potential, culture fit)? | M | Assessment catalogue with construct definitions | Defines what is being scored |
| What are the predictive validity coefficients, and against which outcomes (performance, turnover, retention)? | M | Signed validation report with sample sizes | Shows scores relate to job outcomes |
| What internal consistency and test-retest reliability applies to each assessment type? | M | Reliability tables per assessment | Unreliable scores cannot be valid |
| Describe the norm groups: size, demographics, geography, recency. | D | Norm group documentation | Prevents misleading comparisons |
| How do raw scores become interpretable outputs? | D | Scoring manual | Supports consistent decisions |
| How are cheating, random responding and anti-fraud controls handled? | M | Proctoring and detection description, completion and drop-off data | Protects test integrity |
Selection Lab's deck reports customer outcomes (27% fewer drop-offs, March 2025; 21% lower early turnover, January 2024). Treat such figures as a starting point and ask for the method behind each one, as you would with any vendor.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Provide subgroup analyses and the adverse impact method (impact ratios). | M | Adverse impact report, remediation process | Core fairness evidence |
| What is explainable: scores, rationale, item-to-competency mapping? How is it shown to recruiters and candidates? | M | Sample recruiter and candidate outputs | Enables contestability |
| How can recruiters review and override outputs, and is this logged? | M | Workflow screenshots, audit log sample | Prevents rubber-stamping |
| How are model or test changes controlled? | D | Change log and regression testing policy | Keeps validity evidence current |
| How are accessibility needs and multilingual delivery handled, and how is fairness monitored over time? | D | Accessibility statement, monitoring schedule | Ongoing fairness |
The UK Information Commissioner's Office (ICO) guidance on AI and automated decision-making points to transparency, contestability, meaningful human review and bias monitoring in recruitment automation. For GDPR Article 22 purposes, ask the vendor to state whether any decision is solely automated.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Where is personal data stored, and does any processing occur elsewhere? | M | Residency statement, sub-processor locations | Transfer risk |
| What lawful basis and consent flow apply to each processing purpose? | M | Consent screens, privacy notice | Lawful processing |
| What are the retention periods by data type? | M | Retention schedule | Storage limitation |
| How are DSAR and erasure requests handled and confirmed? | M | Process document, deletion confirmation sample | Rights handling |
| Does the vendor supply inputs for employer DPIAs? | D | DPIA input pack | Reduces your workload |
| How is personal information removed from conversation data, and are local LLMs used? | D | Technical description | Limits exposure |
Selection Lab states that personal data is stored in Frankfurt, that consent and retention periods are set per processing purpose, and that results are shown to the candidate first, with consent requested again before sharing. It also states that local LLMs remove personal information from conversations. Verify each statement against the DPA and sub-processor list.
Request the supported ATS list and proof that candidates, reports and scores write back accurately (M). Selection Lab states that candidates and reports can be viewed and updated inside the ATS and that attributes are filled automatically. Ask for a live write-back test using your own ATS sandbox.
For ATS integration via API and webhooks, request:
Selection Lab states that SmartChat responds to every candidate within 10 seconds. Ask for the measurement method and the latency under peak load.
The European Commission's "AI Act | Shaping Europe's digital future" page states that from 2 December 2027, high-risk AI systems face strict obligations before market placement, including risk assessment and mitigation, human oversight and record-keeping. Recruitment systems are within the scope of employment-related high-risk use. Request:
| Question | Level | Evidence required |
|---|---|---|
| Map each assessment use case to AI Act obligations. | M | Classification memo |
| Provide lifecycle risk assessment (bias, data quality, robustness, cybersecurity). | M | Risk management file |
| Describe controls for stopping or overriding outputs. | M | Human oversight design |
| Describe logging and traceability. | M | Log specification |
| Supply deployer documentation. | D | Instructions for use, technical documentation pack |
Score mandatory items pass or fail. A fail without a dated remediation plan removes the vendor from the shortlist. Score desirable items from 0 to 5.
Suggested weights: privacy and security 30%, validity and reliability 25%, fairness and explainability 20%, EU AI Act readiness 10%, integration 10%, commercial terms 5%.
Grade each answer by evidence type:
Cap the score of any Grade C answer at half marks. Apply the same standard to every vendor on the shortlist, Selection Lab included.
This document is a question bank for procurement teams shortlisting an AI-driven assessment vendor such as Selection Lab. Use it to draft RFP and tender questions for hiring assessments. It assumes you are assessing a platform that screens, tests or scores candidates and that you need evidence you can defend in an audit.
Each question carries a priority: M (mandatory, pass or fail with evidence) or D (desirable, scored). Require every answer to cite an appendix. A statement without an attached artefact counts as a claim, not evidence.
Attach three conditions to the bid:
Request one compliance pack containing:
Rationale: a missing item at bid stage is a reliable predictor of a missing item at audit stage.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Which constructs does each assessment measure (soft skills, hard skills, intelligence or potential, culture fit)? | M | Assessment catalogue with construct definitions | Defines what is being scored |
| What are the predictive validity coefficients, and against which outcomes (performance, turnover, retention)? | M | Signed validation report with sample sizes | Shows scores relate to job outcomes |
| What internal consistency and test-retest reliability applies to each assessment type? | M | Reliability tables per assessment | Unreliable scores cannot be valid |
| Describe the norm groups: size, demographics, geography, recency. | D | Norm group documentation | Prevents misleading comparisons |
| How do raw scores become interpretable outputs? | D | Scoring manual | Supports consistent decisions |
| How are cheating, random responding and anti-fraud controls handled? | M | Proctoring and detection description, completion and drop-off data | Protects test integrity |
Selection Lab's deck reports customer outcomes (27% fewer drop-offs, March 2025; 21% lower early turnover, January 2024). Treat such figures as a starting point and ask for the method behind each one, as you would with any vendor.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Provide subgroup analyses and the adverse impact method (impact ratios). | M | Adverse impact report, remediation process | Core fairness evidence |
| What is explainable: scores, rationale, item-to-competency mapping? How is it shown to recruiters and candidates? | M | Sample recruiter and candidate outputs | Enables contestability |
| How can recruiters review and override outputs, and is this logged? | M | Workflow screenshots, audit log sample | Prevents rubber-stamping |
| How are model or test changes controlled? | D | Change log and regression testing policy | Keeps validity evidence current |
| How are accessibility needs and multilingual delivery handled, and how is fairness monitored over time? | D | Accessibility statement, monitoring schedule | Ongoing fairness |
The UK Information Commissioner's Office (ICO) guidance on AI and automated decision-making points to transparency, contestability, meaningful human review and bias monitoring in recruitment automation. For GDPR Article 22 purposes, ask the vendor to state whether any decision is solely automated.
| Question | Level | Evidence required | Rationale |
|---|---|---|---|
| Where is personal data stored, and does any processing occur elsewhere? | M | Residency statement, sub-processor locations | Transfer risk |
| What lawful basis and consent flow apply to each processing purpose? | M | Consent screens, privacy notice | Lawful processing |
| What are the retention periods by data type? | M | Retention schedule | Storage limitation |
| How are DSAR and erasure requests handled and confirmed? | M | Process document, deletion confirmation sample | Rights handling |
| Does the vendor supply inputs for employer DPIAs? | D | DPIA input pack | Reduces your workload |
| How is personal information removed from conversation data, and are local LLMs used? | D | Technical description | Limits exposure |
Selection Lab states that personal data is stored in Frankfurt, that consent and retention periods are set per processing purpose, and that results are shown to the candidate first, with consent requested again before sharing. It also states that local LLMs remove personal information from conversations. Verify each statement against the DPA and sub-processor list.
Request the supported ATS list and proof that candidates, reports and scores write back accurately (M). Selection Lab states that candidates and reports can be viewed and updated inside the ATS and that attributes are filled automatically. Ask for a live write-back test using your own ATS sandbox.
For ATS integration via API and webhooks, request:
Selection Lab states that SmartChat responds to every candidate within 10 seconds. Ask for the measurement method and the latency under peak load.
The European Commission's "AI Act | Shaping Europe's digital future" page states that from 2 December 2027, high-risk AI systems face strict obligations before market placement, including risk assessment and mitigation, human oversight and record-keeping. Recruitment systems are within the scope of employment-related high-risk use. Request:
| Question | Level | Evidence required |
|---|---|---|
| Map each assessment use case to AI Act obligations. | M | Classification memo |
| Provide lifecycle risk assessment (bias, data quality, robustness, cybersecurity). | M | Risk management file |
| Describe controls for stopping or overriding outputs. | M | Human oversight design |
| Describe logging and traceability. | M | Log specification |
| Supply deployer documentation. | D | Instructions for use, technical documentation pack |
Score mandatory items pass or fail. A fail without a dated remediation plan removes the vendor from the shortlist. Score desirable items from 0 to 5.
Suggested weights: privacy and security 30%, validity and reliability 25%, fairness and explainability 20%, EU AI Act readiness 10%, integration 10%, commercial terms 5%.
Grade each answer by evidence type:
Cap the score of any Grade C answer at half marks. Apply the same standard to every vendor on the shortlist, Selection Lab included.
