Knockri

Knockri's Ethical AI page is a single-paragraph statement of principles (transparency, careful data selection, bias reduction, verification and validation) with no model documentation, governance structure or policy detail.

https://knockri.com/ethical-ai/

The science page works as an explainability statement: it says the models analyse only the interview transcript with NLP, detect no non-verbal or emotional cues, and avoid the practice of training on human evaluators' interview ratings.

https://knockri.com/science-and-dei/

Knockri's PCT application 'System and method for behavioral attribute measurement' (published 8 December 2022) discloses the method: responses are converted to text, transformer language models such as BERT or RoBERTa are fine-tuned on human-annotated behaviour data, and standardised scores are converted to percentiles to rank applicants.

https://patents.google.com/patent/WO2022251970A1/en

The Treasury Board Secretariat's Algorithmic Impact Assessment of its Knockri pilot (published February 2024) explains the scoring in plain language: audio is transcribed, an NLP model flags work behaviours, each question's score is the share of rubric behaviours present and the overall score is their average; it also says the production models are static, the algorithm is a trade secret and Knockri's technical documentation went to TBS rather than being published.

https://open.canada.ca/data/en/info/52b8574d-5c95-463b-b375-8edd092cea30

The reports page offers the white papers 'The Science of Knockri's AI-Based Selection Procedures' and 'The Knockri Approach to Legal Defensibility', but each 'Get The Report' button leads to a lead-capture form rather than a public download.

https://knockri.com/reports/

IBM's May 2026 case study says Knockri uses traditional machine learning for scoring and classification, uses LLMs for summarization, interpretation, orchestration and candidate support, and governs both through watsonx.governance.

https://www.ibm.com/new/product-blog/how-knockri-scaled-structured-hiring-with-ibm-watsonx

No model or system card, instructions-for-use document, ISO/IEC 42001 certification or downloadable technical documentation pack was found on the knockri.com sitemap pages (ethical-ai, science-and-dei, security, reports, privacy and product pages) or in web searches. Knockri also runs a Drata trust center at trust.knockri.com that is not linked from its site; it sits behind a Cloudflare bot challenge, so its contents could not be reviewed for this assessment.

Have a URL that disputes this?

No NYC Local Law 144 bias-audit summary or independent audit (BABL AI, DCI, ORCAA, Warden AI, Holistic AI, Credo AI) was found on any knockri.com sitemap page (including the security, science, reports and FAQ pages), in web searches pairing Knockri with 'bias audit', 'Local Law 144' and 'AEDT', or at trust.warden-ai.com/knockri (404). Knockri also runs a Drata trust center at trust.knockri.com that is not linked from its site; it sits behind a Cloudflare bot challenge, so its contents could not be reviewed for this assessment.

Have a URL that disputes this?

The TBS Algorithmic Impact Assessment says Knockri gave TBS documentation showing that its behavioural predictions perform equally across demographic groups and that its transcription error rates beat industry guidelines, but answers 'No' when asked whether the bias-testing information is publicly available.

https://open.canada.ca/data/en/info/52b8574d-5c95-463b-b375-8edd092cea30

Knockri's report on adverse impact and its content and criterion-related validation studies sits behind a lead form, and the page shows no figures.

https://knockri.com/the-knocki-approach-to-legal-defensibility/

Knockri describes an 842-participant study, run with Virginia Tech's Louis Hickman and presented at SIOP, that relates its automated scores to rated personality, cognitive ability, GPA and hireability; the write-up reports no subgroup or adverse-impact results.

https://knockri.com/blog/recruiting-tools/siop-annual-conference/

The Globe and Mail reported on 3 January 2020 that Knockri's software also analysed facial expressions and had been trained on applicant videos scored by client recruiters, which differs from Knockri's current transcript-only description; no independent audit has verified the current design.

https://www.theglobeandmail.com/business/article-canadian-employers-turn-to-ai-to-analyze-video-interviews-of-job/

TBS's published Algorithmic Impact Assessment of its Knockri pilot (Impact Level 1 under Canada's Directive on Automated Decision-Making) assesses effects on rights, equality, dignity, privacy and economic interests and documents an opt-out to an in-person alternative, a recourse route and HR review of every result, which makes it the closest public analogue to a deployer FRIA, although the deployer wrote it.

https://open.canada.ca/data/en/info/52b8574d-5c95-463b-b375-8edd092cea30

TBS's Privacy Impact Assessment summary for the Knockri pilot lists nine risks, including algorithmic bias and limited transparency, with mitigations such as HR advisers auditing the AI results.

https://www.canada.ca/en/treasury-board-secretariat/corporate/reports/access-information-privacy-reports-treasury-board-secretariat/privacy-impact-assessment-summary-using-artificial-intelligence-automate-candidate-evaluations.html

Knockri's government page says it 'successfully completed' the AIA and PIA with the Treasury Board of Canada Secretariat, which shows it is willing to support deployer impact assessments.

https://knockri.com/government/

No EU AI Act Article 27 FRIA template, deployer-obligation guidance or other EU AI Act material was found on any knockri.com sitemap page or in web searches pairing Knockri with the AI Act.

Have a URL that disputes this?

Says scoring analyses only the interview transcript and detects no non-verbal or emotional cues, and that race, gender, age, ethnicity, accent, appearance and sexual preference do not contribute to a score.

https://knockri.com/science-and-dei/

Knockri's patent describes its training data as interview transcripts labelled for work behaviours by at least two human annotators, with disagreements settled by subject-matter experts, using a behaviour taxonomy adapted from the O*NET work-activity framework.

https://patents.google.com/patent/WO2022251970A1/en

The candidate FAQ lists the data stored (name, email, phone number, and video, audio or text responses) and says it is kept until the employer instructs a purge or the candidate asks for deletion, with no default retention period.

https://knockri.com/candidate-faq/

The security page claims ISO 27001:2013 certification (an edition since replaced by ISO/IEC 27001:2022), describes GDPR measures, and lists data Knockri does not collect, such as health, financial, date-of-birth and government-ID data.

https://knockri.com/security/

The privacy notice (updated 26 February 2025) says services are hosted in the USA, Canada and the EU and relies on Standard Contractual Clauses, but it still cites the invalidated Privacy Shield framework, ties candidate-data retention to customer agreements and does not say whether candidate responses are used to train models.

https://knockri.com/privacy/

Knockri announced a SOC 2 Type II examination in February 2023 without naming the audit firm.

https://knockri.com/blog/press/knockri-achieves-soc-2-type-2-certification/

The homepage FAQ says every candidate gets a proctoring 'Trust Score' that estimates the likelihood of cheating using active proctoring and 'AI-driven classifiers', but it does not say which signals feed the score.

https://knockri.com/

Scores are broken down by skill and behaviour so recruiters can see 'the rationale behind the score', and shortlists can be filtered by overall score or by individual skill.

https://knockri.com/shortlisting-transparent-scoring/

The candidate FAQ says employers can set responses to be scored by AI or reviewed by their talent team, and that 'final decisions are always made by people, not AI'.

https://knockri.com/candidate-faq/

TBS's AIA reports that the system allows human overrides and logs them, keeps an audit trail naming the model version behind each result, can give reasons, and gives HR a Behaviour Insight report of behaviours shown and missed plus the audio of every response.

https://open.canada.ca/data/en/info/52b8574d-5c95-463b-b375-8edd092cea30

A 'bias filter' lets recruiters view candidate results by ID rather than by name.

https://knockri.com/fair-hiring/

The behavioural-assessments page says recruiters can 'automatically send invites to schedule interviews based on the rank order of top scoring candidates', an option that lets the score alone trigger the next-stage invitation.

https://knockri.com/behavioural-assessments/

The responsible disclosure policy sends vulnerability reports to security@knockri.com; there is no bug bounty, and knockri.com/.well-known/security.txt returns 404.

https://knockri.com/responsible-disclosure-policy/

Knockri's Atlassian Statuspage, which status.knockri.com also points to, shows 'Page Inactive' and can be viewed only by team members; its API returns 401.

https://knockri.statuspage.io/

TBS's AIA says Knockri's production algorithms are static and that a log of model and system changes is kept, but that log is not public.

https://open.canada.ca/data/en/info/52b8574d-5c95-463b-b375-8edd092cea30

IBM says watsonx.governance provides oversight, risk controls, traceability and auditability across Knockri's workflow, but no monitoring results are published.

https://www.ibm.com/new/product-blog/how-knockri-scaled-structured-hiring-with-ibm-watsonx

No public log of model updates, bias or performance monitoring dashboard or incident history was found on knockri.com, on its public help centre at knockri.atlassian.net or on the inactive Statuspage.

Have a URL that disputes this?

A detailed candidate FAQ, also available in French, covers what is assessed, AI versus human review, the data stored, retention, accommodations and 24/7 support.

https://knockri.com/candidate-faq/

The product overview describes building assessments from video, audio or written questions, giving accommodations and extensions, and scoring candidates against a customised skills framework with questions vetted by an I/O psychologist.

https://knockri.com/product-overview/

Pricing is public: Growth costs $799 a month and Business $2,499 a month, both billed annually and covering 2,500 and 10,000 AI-evaluated interviews a year, while Enterprise is priced on request.

https://knockri.com/pricing/

A public Jira Service Management help centre hosts Candidate Care resources and a Recruiter Support portal.

https://knockri.atlassian.net/servicedesk/customer/portals

The integrations page shows connectors for Workday, SAP, Oracle, iCIMS, Greenhouse, ADP Workforce Now, BambooHR and Avature, including bulk invites and data import and export.

https://knockri.com/integrations/

Since May 2022 Knockri's behavioural video assessment has been listed on the SAP Store with integration into SAP SuccessFactors Recruiting.

https://knockri.com/blog/press-releases/knockri-now-available-on-sap-store/

The Fair Hiring page links to the form-gated reports and still shows a placeholder case-study teaser reading 'increase racial & gender diversity by XX%'.

https://knockri.com/fair-hiring/

No public DPA, subprocessor list, guidance for deployers on NYC Local Law 144 or the Illinois AI Video Interview Act, candidate-notice templates or EU AI Act deployer documentation was found on any knockri.com sitemap page.

Have a URL that disputes this?