This is a guide to any questions you might have.

Adrosaga is a global human testing and data platform that helps organisations test, validate and improve AI systems with real people. Companies can test deployed AI agents across countries, languages, demographics, devices and user groups, and can also use the same global contributor community to collect, generate, label, translate and review data for AI development.

Adrosaga enables organisations to put an AI agent in front of real people from precisely targeted audiences and capture structured feedback about how it performs. Testing can examine functionality, user experience, policy and guardrail conformity, compliance and performance across different audiences. Testing can be performed before deployment, during controlled releases and continuously after deployment.

Adrosaga can support a broad range of human AI testing, including functionality, user experience and usability, policy and guardrail conformity, compliance, localisation, accessibility, demographic performance, bias and fairness, hallucination and factual accuracy, instruction following, tone and brand alignment, human escalation, multi-step agentic workflows, regression testing and ongoing post-deployment monitoring.

Adrosaga tests AI systems with real people because real people interact with AI differently from developers, automated test systems and synthetic users. Human testing shows whether an AI actually works for the people it was designed to serve, how different audiences interpret its responses, whether unexpected behaviours emerge, and whether particular demographic groups experience different outcomes — evidence that is particularly valuable alongside automated and synthetic testing.

No. Adrosaga treats synthetic testing as an extremely valuable part of modern AI testing — but not a substitute for human testing. Synthetic testing delivers speed and scale, efficiently exploring large numbers of scenarios, but it cannot replicate the imperfections, cultural context and language nuance that shape how real people actually understand, use and experience an AI system. Nor can it deliver the same level of trust, confidence and assurance as a representative human audience tested at a suitable sample size, particularly across countries, cultures, minorities and broad demographic groups. Adrosaga's human testing is designed to complement synthetic testing, not replace it.

Yes. Adrosaga is well suited to ongoing functionality, user experience and compliance testing after an AI system has gone live. Repeated testing is important for detecting model drift and changes caused by learning or feedback loops, model updates, system-prompt changes, guardrail changes, RAG content or configuration changes, or modified tools and integrations.

An Adrosaga testing project can potentially be launched within 24 hours when the client has the key inputs ready: the target audience, the version of the AI agent or system to be tested, and the questions, scenarios or work items contributors will complete. More complex projects may need additional setup for integrations, credentials, methodology or specialist audiences.

Completion time depends on project scale, complexity, target audience, contributor expertise and response volume. Adrosaga is designed for speed and quality, with built-in automation for matching, distribution, communications, orchestration and data processing. Appropriately scoped tasks with an available audience can be completed within days.

Adrosaga helps organisations build independent evidence that an AI agent is ready to represent the company in front of real users. As AI agents become easier to build, the harder challenge is deciding whether they can be trusted to perform effectively, stay within company policies and guardrails, and meet compliance expectations. Adrosaga is independent of the team that designed and built the agent, reducing the risk of the development team effectively marking its own homework. By testing the agent with an appropriately sized and representative human audience and reporting the objective, methodology, audience, responses, findings and evidence, Adrosaga gives deployment approvers a stronger basis for a go-live decision.

Yes. Multi-step and agentic workflow testing is a core use case for Adrosaga, because real users don't always follow the sequence designers anticipate. Contributors can follow interactions across several steps, assess whether context is preserved, identify where workflows break down, and determine whether the final outcome matches the user's original objective — especially valuable for agentic systems that use tools, retrieve information, make decisions or perform actions.

Yes. Adrosaga can test whether an AI recognises when human escalation is required, escalates at the appropriate point, communicates the escalation clearly, transfers necessary context to the human operator, avoids unnecessary escalation and provides a satisfactory experience across the AI-to-human transition.

No. Adrosaga doesn't impose a limit on the number of tests you can run. You can submit as many testing projects as you need — each one is costed individually through the pricing calculator and in-platform pricing engine before you click Submit.

Adrosaga needs three things to set up a testing project: your target audience, the version of the AI agent or system to be tested, and the questions, scenarios or work items contributors will complete. With those ready, a project can potentially launch within 24 hours; more complex projects may need extra setup for integrations or specialist audiences. Prefer a guided start? Adrosaga also offers a free 30-minute demo walking through test setup, contributor targeting and evidence output, with no commitment required.

Yes. Adrosaga can target testing using contributor attributes such as country, region, language, dialect, fluency, age, education, employment, industry experience, software skills, interests, accessibility requirements and other relevant characteristics. Different audience groups can be included within the same test and compared in reporting.

Yes. Adrosaga is designed for global testing. Contributors can be targeted by country, region, language, regional dialect and level of fluency. The Adrosaga platform itself is automatically translated into more than 100 languages, helping contributors participate in their preferred language. Different language, dialect and fluency groups can be included within the same test and the report can be filtered to compare their responses.

Yes. Adrosaga is built to detect differential AI behaviour across demographic groups — clients can structure the same test across multiple audiences and compare results, with reporting filterable by relevant characteristics to identify differences in usability, interpretation, performance or outcomes.

Yes. Adrosaga lets clients use their own employees, customers or other invited audiences rather than relying exclusively on the global contributor community. This gives clients access to Adrosaga's task orchestration, structured testing and reporting capabilities with audiences they already control, and the option to combine internal or customer audiences with independent external contributors when broader validation is required.

Yes. Adrosaga can target contributors according to language, regional dialect and level of fluency. Different fluency groups can be included within the same test rather than requiring separate projects, and the report can then be filtered by language, dialect or fluency to compare how different audiences experienced the same AI system.

In addition to deployment testing, Adrosaga can provide targeted human-generated data to train, evaluate and improve AI systems. Contributors can create, label, classify, translate, review and evaluate data according to defined project requirements.

Yes. Adrosaga allows organisations to commission new human-generated data specifically for their models, products or use cases rather than relying exclusively on publicly available datasets. Projects can be targeted to particular markets, audiences, expertise levels or demographic groups.

Yes. Adrosaga can allocate the same work item to multiple independent contributors. This enables clients to collect multiple human judgements, measure agreement, compare perspectives and use consensus or disagreement as part of the quality methodology.

Fraud detection is an ongoing process throughout the contributor lifecycle, not just a registration check. Adrosaga uses specialist fraud and identity technology, device and behavioural signals, duplicate-account detection, contributor risk profiles and human review of ambiguous cases. Risk information can continue to be updated as contributors use the platform.

Yes. Identity verification is a core part of Adrosaga's quality controls — contributors are screened through specialist identity and fraud systems, and KYC verification is required before their first payout.

Adrosaga combines community engagement, task-specific contributor ranking, quality tracking and fair contributor treatment. Contributors build experience for each type of task and are ranked accordingly, while badges, achievements and recognition encourage sustained participation and strong performance. Adrosaga also pledges to pay contributors fairly and promptly, provide transparency about work and compensation, and communicate fairly and respectfully. These practices, combined with audience targeting and multiple independent reviews where needed, help build a trusted community capable of increasingly reliable results.

Adrosaga builds task-specific experience profiles rather than treating every contributor as equally suited to every task. Contributor activity, experience and performance can be tracked across different work categories, helping identify people who have demonstrated stronger capability for particular task types and match future work accordingly.

Yes. Adrosaga is designed to orchestrate projects ranging from targeted specialist testing to large global programs involving thousands of contributors and hundreds of thousands of work items. This scale is designed to support multiple concurrent programs running in parallel, rather than one large engagement at a time.

Yes. Clients can use CSV and Excel import/export workflows to move large volumes of work items and results. Adrosaga supports mapping between client data structures and platform work-item structures, while API-based orchestration can support more automated or recurring workflows.

Yes. Adrosaga is designed to support integration and orchestration through APIs as well as platform-based workflows and structured imports and exports. This can connect human testing and data workflows with existing AI systems, operational platforms and data pipelines and reduce manual administration for recurring programs.

Yes. Adrosaga can support ongoing tasks where new batches of content, test cases or work items are added on a recurring basis. As each batch becomes available, eligible contributors can accept and complete it. Clients can limit how many batches an individual contributor may complete, helping broaden the contributor sample, control participation and avoid over-reliance on the same people.

Yes. Adrosaga includes built-in automation for matching contributors to work, distribution, communications, orchestration and data processing, so large programs don't require proportionally more manual effort as they grow. Reporting is automated too — live results and methodology are available as testing runs, and Adrosaga's AI Data Scientist capability can automatically analyse large volumes of responses to surface themes, sentiment and differences across audiences. This runs alongside the platform, APIs and import/export workflows used to move data and work items at scale, keeping repeatable programs efficient to manage.

Adrosaga supports confidentiality controls, NDAs and controlled access to project information. Where contributors need access to a client's AI agent, application or test environment, controlled credentials or test accounts can be allocated to individual contributors. Credentials can be restricted to the intended environment and forfeited or disabled when the contributor completes the task.

Yes. A task can require a non-disclosure agreement before a contributor receives access to protected project information. Clients can use an Adrosaga NDA or provide their own NDA and confidentiality requirements.

Yes. Adrosaga is built to handle sensitive content responsibly — clients can flag tasks involving biometric, sensitive or potentially offensive material during project configuration, contributors are informed before participating, and contributor preferences can exclude categories of work they have opted out of.

For Adrosaga's initial platform launch, the client's AI agent remains separate from the Adrosaga platform. The client provides contributors with access to the specific version of the AI agent within scope, using controlled credentials or test accounts where required. Credentials can be assigned to individual contributors and forfeited or disabled when participation is complete. Adrosaga plans to introduce Bring Your Own Agent capability progressively in later phases to make this integration more seamless.

Adrosaga hosts platform data on AWS and MongoDB Atlas in the AWS Europe (Ireland) region. A small number of specialist providers, such as identity verification and payouts, may process data in other countries under contractual safeguards. These providers are listed in Adrosaga's sub-processor register at adrosaga.com/subprocessors.

Yes. Adrosaga encrypts data in transit with TLS 1.2 or higher, and at rest in databases, file storage and backups.

Only Adrosaga staff whose role requires it. Access is granted per person through single sign-on with MFA, and is removed when no longer needed.

Yes. Adrosaga logically separates each client's data within the platform and applies access checks on every request.

Not yet. Adrosaga is working towards both, and its practices are already aligned with them. Adrosaga's hosting providers, AWS and MongoDB Atlas, are certified to both.

Adrosaga keeps dependencies up to date, reviews code before release and prioritises security fixes. Vulnerabilities can be reported to security@adrosaga.com under Adrosaga's Responsible Disclosure Policy.

If a security incident occurs, Adrosaga investigates, contains and fixes the issue. Adrosaga notifies affected clients without undue delay and within legal or contractual timeframes, with details of what happened and what Adrosaga has done.

Yes. Adrosaga backs up the production database automatically every 6 hours and can restore it to any point in the last 7 days. Longer-term snapshots are kept for up to 12 months. Stored files are versioned, so earlier versions can be recovered.

Adrosaga is established in Singapore and complies with the PDPA, with the PDPC as its lead supervisory authority. Where the people whose data Adrosaga processes are elsewhere, it also meets the laws that apply to them, including the EU and UK GDPR, the Australian Privacy Act and the California CCPA/CPRA.

When a contract ends, Adrosaga gives clients 30 days to export their data. After that, Adrosaga deletes it from production systems within 30 days. Copies in backups are not accessed or used, and expire as backups rotate out, within 12 months at most. Billing records are kept for 7 years for tax and accounting.

Adrosaga's Data Protection Officer is Daniel van Niekerk, contactable at privacy@adrosaga.com. Adrosaga responds to data subject requests within 30 days.

Yes. Adrosaga can complete security questionnaires and share further detail under NDA.

Adrosaga provides built-in reporting showing the project objective, testing methodology, progressive project statistics, contributor responses and resulting findings. Clients can filter the report by relevant audience attributes and other dimensions, and can export all underlying raw data for their own analysis, audit or integration with business intelligence systems.

Yes. Adrosaga's built-in Data Scientist capability can analyse large volumes of structured responses and free-text feedback. It can identify sentiment, themes, differences between groups and audience segments and produce plain-language summaries and recommendations, while clients retain access to the underlying responses and raw data.

Yes. Adrosaga has a transparent pricing policy. Our public pricing calculator lets you estimate a project's likely cost before creating a task, and once you're building it in the platform, the pricing engine recalculates in real time as you adjust requirements. Before you click Submit, you'll see an accurate cost for exactly what you've configured. See the pricing page for current plans and rates.

Adrosaga offers a free, pay-as-you-go option alongside paid subscription tiers that scale with your testing and data needs, adding contributor capacity, crowd credits and support as you grow, with flexible billing options. See the pricing page for current tiers, inclusions and rates.

No. Creating an Adrosaga account does not itself create a payment commitment. Clients can register for a free account and explore the platform without committing to paid work — a payment obligation only begins when the client registers for a paid subscription or, where applicable, clicks Submit on a paid task.

Yes. Clients on an Adrosaga subscription can use their own employees as an internal testing audience without contributor reward fees because Adrosaga does not need to pay those employees as external contributors. To qualify as an internal company audience, staff must register using an email address on the same approved company domain as the client organisation.

Adrosaga's pricing scales with your tier and billing option — contributor capacity, crowd credits and support increase as you move up. When you access our global crowd, Adrosaga provides a pricing calculator that estimates your cost up front and recalculates in real time as you configure a task. See the pricing page for details.

Yes. Adrosaga lets you create a free account and explore the platform yourself — registering doesn't create a payment commitment, so you can look around and set up your first project before deciding whether to submit a paid task or subscribe to a paid tier. The pricing calculator and in-platform pricing engine let you confirm project cost as you build a task. Prefer a guided walkthrough instead? Adrosaga also offers a free 30-minute demo covering test setup, contributor targeting and evidence output, with no commitment required.

Adrosaga's Crowd Credit expiry depends on the contract type. Month-to-month subscriptions don't include Crowd Credits, so contributor work is billed at the rate per the Pricing Calculator. On a 12-month term billed monthly, Crowd Credits are issued each month on a rolling 90-day expiry, so each month's credits must be used within 90 days of being issued. On an annual subscription paid in advance, Crowd Credits are issued for the full 12-month term and expire 365 days after the contract start date — for example, Crowd Credits issued on 1st January expire on 31st December of the same year.

Adrosaga can test whether an AI behaves consistently with company policies, intended behaviours and guardrails. If the global contributor community is used, the relevant policies, guardrails or evaluation criteria must be shared as supporting documentation so contributors know what conformity means. If those materials are confidential and cannot be shared externally, the client should use an internal audience such as its own staff.

Adrosaga's planned Regulation Library will help clients explore relevant regulations and identify testing and evidence considerations for their AI deployment, based on factors such as jurisdiction, industry, intended use, affected people and risk classification. Clients should obtain appropriate legal or regulatory advice for definitive interpretation.

Yes. Adrosaga tests your AI against external regulations and internal policies, brand requirements, operating procedures, safety rules and guardrails, and captures the evidence from that testing. If those internal materials are confidential, the test can use an approved internal employee audience instead of the external contributor community. Adrosaga's role is the testing and evidence — certifying compliance is your organisation's or your compliance consultant's responsibility.

Yes. Adrosaga is structurally independent from the team that builds or operates the AI — evaluators have no involvement in development and no stake in the outcome. That separation is what regulators, auditors and boards look for when deciding whether AI validation evidence can be relied on, rather than treated as a self-assessment. Adrosaga produces the evidence that feeds directly into your team's regulatory filing, not an internal attestation you'd have to file yourselves without third-party backing.

No. Adrosaga is a human validation and evidence capture service, not a compliance consultancy. Adrosaga produces the independent testing evidence; your compliance or legal team applies it to your framework.

Adrosaga doesn't give regulatory advice. It gives you the independently produced data that supports it.

No. Adrosaga does not build AI; it tests AI systems and gives you the evidence your compliance team can take upstream. Structural independence from AI system builders is what makes the evidence count.

Adrosaga is configurable, so clients can ask whatever questions are appropriate for the AI system, testing objective and intended audience. Clients do not need to design a test from scratch: once they identify the objective and test type, Adrosaga can provide templates and question libraries to help guide the testing process.

Yes. Version control is built into how Adrosaga runs tests — clients provide access to the specific environment or version of the AI agent in scope, which matters because model, prompt, RAG, guardrail, tool and integration changes can alter behaviour between versions and affect the value of repeat testing.

Yes. Combining test types within one project is a core Adrosaga capability — a single structured test can evaluate several dimensions of an AI system, for example functionality, user experience, policy conformity, human escalation and demographic differences, while collecting separate structured responses for each objective.

Traditional data-labeling platforms are typically used to collect, annotate or review data used to train and improve AI systems. Adrosaga can support human-data workflows, but its deployment-testing proposition is different: organisations can test live or pre-launch AI systems with representative human audiences, structured scenarios and targeted demographic, geographic and linguistic groups. The outcome is evidence about how the deployed AI behaves with real people, rather than only a labelled dataset for model development.

Adrosaga solves a different problem than synthetic AI testing tools. Synthetic tools use automated or simulated evaluation to test AI behaviour at speed and scale, but they can't replicate the imperfections, cultural context and language nuance that shape how real people actually experience an AI system. Adrosaga complements that with representative human evaluation across demographics, countries, languages and levels of fluency, delivering a level of trust, confidence and assurance that speed-and-scale synthetic testing alone can't achieve. This adds evidence about how real users interpret, experience and complete tasks with an AI system, including where outcomes differ between audience groups. The two approaches can be used together rather than treated as substitutes.

User research platforms are useful for understanding user needs, attitudes, usability and broader experience. Adrosaga is designed around structured testing of AI systems: defining test objectives, selecting representative audiences, running scenarios against a specific AI deployment and producing structured findings that can support product, risk and governance decisions. If the primary need is open-ended discovery or general user research, a dedicated user research platform may be the better fit. If the need is repeatable human validation of AI behaviour and outcomes, Adrosaga is designed for that use case.

Specialist AI evaluation vendors vary significantly, so the comparison should be made against the specific vendor. Adrosaga's core distinction is representative human testing: organisations can define target audiences by geography, language, dialect, fluency, demographics and other relevant characteristics, run structured scenarios against a specific AI deployment, and compare outcomes across groups. For vendors focused primarily on automated or synthetic evaluation, the approaches may be complementary rather than interchangeable.

Use Adrosaga when the question depends on how real people understand, trust, experience or act on an AI system, especially across different countries, languages, demographics or levels of fluency. Synthetic testing remains useful for automated coverage, repeatability and scale. Adrosaga adds representative human evidence for questions that require interpretation, context or real-user outcomes. In most testing programs, the strongest approach is to use both.

Adrosaga is not intended to replace an internal QA team. Internal QA remains important for product quality, functionality, regression and system knowledge. Adrosaga is useful when you need an independent testing layer, representative external or internal audiences, or structured human evidence that sits outside the team that built and validated the system. The two approaches are complementary: internal QA can validate the product against internal requirements, while Adrosaga can add independent deployment-level testing with target user groups.

Yes. Adrosaga is designed to sit alongside automated evals, synthetic testing, internal QA and other AI testing tools rather than require teams to replace them. It adds representative human testing, audience targeting and structured findings that can be used to examine how a specific AI deployment performs with real people. This lets organisations combine automated coverage with human evidence where interpretation, usability, guardrails, fairness or real-world outcomes matter.

Adrosaga adds a layer of independence internal staff cannot replicate: contributors who did not build the system, are not invested in its success, and represent audiences outside your organisation. This produces evidence of how real, external users across demographics, languages and markets actually experience your AI, rather than how colleagues familiar with the product experience it. Internal staff remain essential for product knowledge and functional QA; Adrosaga adds the independent, third-party evidence that regulators, boards and clients increasingly expect before an AI system goes live.

Adrosaga exists because internal validation alone carries an inherent conflict of interest: when it's the only evidence used to approve a system for deployment, the same team that built the AI is also grading it. Internal teams can and should validate their own work as part of standard development practice, but Adrosaga adds an evaluation layer outside the development team, applying the same separation-of-duties principle used in risk and audit functions. This doesn't replace internal validation; it adds the independent check a self-assessment alone can't provide.

Internal teams often share the same assumptions, context and familiarity with the product that shaped its design, which can make it harder to notice where an AI system fails for users who don't share that context. Adrosaga's contributors approach the system without that familiarity, sourced to represent the demographics, languages and markets the deployment will actually serve. This surfaces usability, interpretation and edge-case issues, such as confusing prompts, culturally specific misunderstandings and unexpected user paths, that a team close to the build is less likely to encounter.

Adrosaga's parent-child account structure gives consulting firms a parent account with configurable access across their clients' linked child accounts. A firm can configure the parent account for oversight and monitoring only, or enable full edit access, so a consultant can track compliance and deployment status across clients and, where permissions allow, work directly within a client's account rather than switching between separate, disconnected logins. This is built for firms managing AI validation and compliance obligations across more than one client at a time.

Yes. Adrosaga's parent-child account structure is designed for this: a consulting firm's parent account gives oversight, monitoring capabilities and validation tools across each linked child account on a user by user basis, so deployments and compliance status across multiple clients can be tracked from one place rather than switching between separate accounts.

No. Adrosaga's parent-child structure gives child accounts no visibility into each other, so one client cannot see another client's data even under the same parent account.

Configurable permissions determine what parent account users can access and control across their linked child accounts, so a consulting firm's own team can be granted the oversight they need without exposing one client's data to another.

No. Adrosaga places no cap on the number of child accounts a consulting firm can link under a single parent account, so the structure scales as a firm's client base grows.

Adrosaga supports either option. Billing can be configured based on the relationship between a parent account and its child accounts, consolidated at the parent level or billed separately per child account or per project within that child account, so a consulting firm can match billing to how they actually work with each client.

Yes. Adrosaga lets a child account be unlinked from its parent account when a consulting firm's engagement with that client ends, so the relationship doesn't have to remain in place after the work does. The client doesn't lose their account history or access when this happens — if they want to keep using Adrosaga after unlinking, they can, but they'd need to accept billing and control of their own account.

Yes. Adrosaga's child accounts are built to cover all of a client's crowd testing needs, not just a single AI deployment — a client can test multiple AI agents or systems under the same child account, with each deployment managed and reported on separately. Child accounts are structured around the client relationship rather than any individual deployment, so a consulting firm doesn't need to create a new child account every time a client wants to test another AI system.

Yes. Adrosaga lets a child account be linked to more than one parent account at the same time. Each parent-child link carries its own configurable permissions, so every parent account only sees and manages the scope of work it's responsible for, without visibility into the client's other engagements.