Canditech Review: The Skills Assessment Platform's Evidence
On this page 10 sections
Canditech is a configurable pre-employment assessment platform. It can deliver coding, SQL, spreadsheet, written, video, cognitive, personality, and job-simulation exercises; connect results to applicant-tracking systems; apply proctoring controls; and, on higher plans, use AI to score some responses. Its strongest use case is replacing generic screening with a work sample that a hiring team has tied to the actual job.
The platform cannot prove by itself that a test predicts performance, treats groups fairly, or complies with every employment law. Those properties depend on the job analysis, questions, scoring rules, candidate population, accommodations, cutoff, monitoring, and human decisions used by each employer.
This review was checked on September 13, 2026. It is based on public product documentation and has not involved a paid account, a controlled candidate trial, a security audit, or access to Canditech’s private validation studies. Feature, security, integration, and customer-result claims are labeled as vendor-reported. Legal sections describe operational questions, not legal advice.
Product scope and the core workflow
Canditech lets an employer select or build an assessment, invite candidates, score submissions, review reports, and move results into a recruiting workflow. The current Canditech product tour lists more than 500 prebuilt test sections and question formats including multiple choice, open text, video, file upload, coding, SQL, Excel, and embedded Google Sheets.
The product is broader than a coding-test service. A hiring team can combine technical tasks with communication or scenario questions and can customize instructions, weights, timing, branding, and score thresholds. It can also import an existing assessment.
That flexibility is useful and dangerous. A realistic task can improve job relevance. A poorly designed custom test can add noise, exclude qualified candidates, or measure familiarity with the test format rather than ability to do the job. The platform provides tools; the employer still owns the inference from score to employment decision.
The most defensible workflow starts with a job analysis. Teams should identify important work behaviors, define what acceptable performance looks like, choose a sample that closely represents the work, pilot the task, and review results against later job performance. A library label such as “data analyst” is not evidence that a particular question is valid for every analyst role.
What a candidate can be asked to do
Canditech’s public documentation describes several assessment modes:
| Mode | What the platform supports | Main review question |
|---|---|---|
| Coding and frontend tasks | Code execution, hidden tests, project work, and AI-assisted scoring | Does the task match the production stack and seniority? |
| SQL and spreadsheets | Queries and work inside embedded data tools | Is the dataset representative without exposing company data? |
| Open text and video | Manual review or AI scoring of written content and transcripts | Is the rubric job related and accessible? |
| Cognitive and personality tests | Standardized sections from the test library | What validity evidence applies to this job and population? |
| AI-proficiency tasks | Embedded AI tools with prompts and activity available for review | Is AI use allowed in the real job, and what behavior is being scored? |
Canditech says customers can embed ChatGPT or Google Search inside an assessment and record the candidate’s activity. Its advanced-features documentation says these tools are available on the Enterprise plan.
This feature can make an assessment more realistic for jobs where employees are expected to research or work with AI. It can also create a misleading test if the employer bans tools during assessment but expects them on the job. A good work sample measures how the work is actually done, including which aids are permitted.
Candidate instructions should state what is recorded, how it is used, and which behaviors affect the score. Hidden monitoring may reduce trust and make accommodation requests harder.
AI scoring is configurable, not automatically objective
Canditech offers AI-assisted scoring for open-text, video, and coding responses on Enterprise plans. Its AI scoring documentation describes prebuilt agents and custom agents driven by employer-written rules. For video responses, Canditech says it transcribes speech and scores the text rather than appearance, background, tone, or accent. Reviewers can inspect and override scores.
Those design choices can reduce some irrelevant inputs, but they do not make a score neutral. Speech-to-text accuracy can vary by accent, language, audio quality, disability, and domain vocabulary. A text-only system can still reward a writing style correlated with education or culture. Employer-written criteria can encode vague or discriminatory preferences.
Automation can also create false consistency. Applying the same flawed rule to every candidate is repeatable, but not valid. Before using AI scores to reject applicants, an employer should test:
- agreement between trained human reviewers and the model;
- error patterns by job-relevant response type and protected group where lawful to measure;
- stability across model or prompt changes;
- sensitivity to small wording changes;
- override rates and reasons;
- the relationship between scores and later job performance.
The platform documentation does not publish the underlying model providers, versioning policy, validation sample, error rates, subgroup results, or independent audit for each scoring agent. A buyer should obtain those details under a data-processing and service agreement rather than assume the marketing term “objective” settles them.
Work samples have evidence, but local validation still matters
Work-sample tests have a long research history. A 2005 meta-analysis in Personnel Psychology found a mean observed correlation of .26 between work-sample scores and job-performance measures, rising to .33 after correcting performance-measure unreliability. The peer-reviewed paper also warned that earlier summaries had overstated validity.
Those figures do not validate Canditech as a whole. They aggregate different jobs, samples, criteria, and study quality. A simulation of the wrong work can be highly realistic and still predict little that matters.
The Society for Industrial and Organizational Psychology’s Principles for the Validation and Use of Personnel Selection Procedures explains that evidence can be criterion related, content based, or construct based. It also emphasizes job analysis, reliability, fairness, documentation, and appropriate interpretation.
For a customized Canditech exercise, content evidence may be the practical starting point: show that tasks and scoring cover important behaviors from the job. If an employer ranks or rejects large applicant pools, it should also examine criterion evidence and adverse impact over time.
Canditech says its psychometric team can help customers build assessments. That is a vendor service claim. Buyers should request the actual job-analysis record, validation report, reliability evidence, version history, and scope rather than treating the presence of a psychometrician as validation.
Proctoring creates a tradeoff, not a “cheat-proof” result
Canditech advertises controls including copy and paste restrictions, tab or cursor tracking, randomized questions, device and location monitoring, webcam snapshots, time limits, plagiarism detection, and signals intended to detect ChatGPT use. The current proctoring help article explains that employers choose which controls to enable and that candidates receive notices about monitored behavior.
No remote assessment is literally cheat proof. A second device, another person, accessibility software, remote desktop, paraphrasing, or a model outside the monitored environment can defeat individual signals. Aggressive controls can also flag legitimate behavior, such as switching to assistive technology, looking away while thinking, using multiple monitors, or losing connectivity.
Employers should define what counts as misconduct before reviewing flags. A monitoring signal should trigger review, not an automatic accusation. Teams should offer an appeal path and an alternative format when a disability, privacy constraint, or technical limitation prevents use of a proctoring method.
For many roles, allowing tools and designing a harder, role-specific task is more realistic than trying to reconstruct a closed-book exam at home. Canditech’s embedded-AI mode points in that direction, though the employer still needs a scoring method that separates useful judgment from copied output.
Pricing is public at a reference volume
Canditech now publishes reference prices based on annual candidate volume. At the 100-candidate selection shown on September 13, 2026, the pricing page listed:
| Plan | Displayed price | Selected limits and features |
|---|---|---|
| Team | $150 per month, billed yearly | Two active assessments, eight users, core question types |
| Pro | $200 per month, billed yearly | Unlimited assessments and users, ATS integration, coding, SQL, frontend, and Sheets |
| Enterprise | Contact sales | AI auto-scoring, embedded AI tools, API, SAML, webcam snapshots, and psychometric support |
These are current vendor prices for the selected volume, not a universal quote. Candidate allowances, taxes, implementation, services, renewal terms, and volume changes can alter total cost. Buyers should preserve a dated proposal and ask how retakes, abandoned sessions, test candidates, and duplicate invitations count.
Cost also includes candidate time and hiring-manager work. A 90-minute exercise sent to 1,000 applicants consumes 1,500 candidate hours even if software scoring is automatic. The assessment should appear late enough that the employer has a real interest in the applicant, or be short enough to justify broad screening.
Integrations reduce copying but do not remove governance
Canditech says it connects with more than 40 applicant-tracking systems. Greenhouse’s own Canditech integration guide confirms that Greenhouse Recruiting can add a Canditech stage and receive assessment results.
An integration can reduce manual invitations and transcription errors. It can also turn a score into an automatic workflow decision. Teams should inspect which data fields move between systems, who can see them, how long they remain, what happens after deletion, and whether a rejected candidate can obtain an explanation.
Automation rules deserve special attention. A cutoff may look harmless in the assessment tool but become decisive when the ATS automatically rejects everyone below it. Version changes can make scores from different candidate cohorts incomparable. Recruiters should log assessment version, scoring version, cutoff, and override for each decision.
Security statements need documents behind them
Canditech’s trust center says data is hosted on AWS infrastructure in the European Union, encrypted with AES-256 at rest and TLS 1.2 or 1.3 in transit, and covered by an ISO/IEC 27001 certification. It also says the company performs penetration testing at least annually and offers a data-processing agreement.
These are vendor-reported controls. The public page does not include the certificate scope, current audit dates, statement of applicability, penetration-test report, subprocessor list, incident history, recovery tests, or retention defaults. Buyers handling applicant video, identity, location, and assessment data should request those materials.
Security review should follow the data, not the logo. Ask where candidate content, transcripts, webcam images, model prompts, backups, logs, and support exports are stored. Ask which subprocessors receive them and whether any provider can train a model on the content. Confirm deletion times and what happens when an employer closes an account.
Employment-law duties remain with the employer
In the United States, a vendor does not absorb an employer’s duties under Title VII, the Americans with Disabilities Act, or applicable state and local rules. The EEOC has said that automated systems used to make or inform selection decisions can create unlawful disparate impact. Its agency history and AI guidance summary also notes that passing the four-fifths rule does not guarantee a procedure is lawful.
New York City’s Local Law 144 applies to covered automated employment decision tools. The city’s official AEDT page says covered employers and employment agencies must arrange an annual bias audit, publish specified information, and give required notices before use. Whether a configured Canditech workflow falls within the law depends on how it substantially assists or replaces discretionary decisions.
An employer should therefore document:
- the job analysis and reason for each assessed construct;
- validation and reliability evidence for the configured test;
- notice, consent, retention, and accommodation processes;
- adverse-impact monitoring and the population used;
- human review, appeal, and override rules;
- version changes to content, scoring models, and cutoffs.
Buying software does not complete this work.
Who should shortlist Canditech
Canditech is a plausible shortlist candidate for teams that need customized work samples across several job families and want technical tasks, written responses, video, spreadsheets, AI-use exercises, and ATS integration in one system. Its public pricing makes an initial comparison easier than it once was.
It is less compelling for a team that wants a fully validated off-the-shelf test with published norms and independent technical manuals for a narrow occupation. It may also be excessive for low-volume hiring where a structured interview and a carefully reviewed work sample can be administered without another platform.
A useful pilot should test one role and one decision point. Run the same sample through trained human review and any automated scoring. Track completion, candidate feedback, accommodations, subgroup outcomes where lawful, reviewer time, overrides, and later job evidence. Do not begin by automating rejection across the company.
The platform’s value depends on the quality of the assessment wrapped around it. Canditech can make a well-designed work sample easier to deliver and review. It can also make a weak test easier to scale. Procurement should be based on which of those two outcomes the employer can demonstrate.
Source note
Sources were checked on September 13, 2026. Product, price, security, integration-count, and customer-result statements from Canditech are vendor-reported. Greenhouse confirms its own integration. Research and professional standards support careful use of work samples but do not validate Canditech’s complete platform. Legal requirements vary by location and configuration.