ZynoviqStart small. Prove it on your own system.
Every engagement begins with one workflow and one domain. You see the quality of our work on your AI before committing to anything larger.
The ladder
| Step | Package | Scope | How it is priced | Best for |
|---|---|---|---|---|
| 0 | Free sample test | 25 synthetic cases, one domain, about two weeks | Free | Anyone whose AI handles Indian GST |
| 1 | AI Quality Diagnostic | About 100 cases on one workflow, about four weeks; failure analysis, executive report, regression set | Fixed price, quoted | A first paid look at one workflow |
| 2 | Capability assessment | 300 to 500 cases across a domain; your baseline | Fixed price, quoted | A baseline before launch or a major change |
| 3 | Continuous evaluation | Quarterly re-tests, re-tests after notified releases, a release gate you set | Annual subscription; extra releases per run | AI that is load-bearing |
| 3+ | Regulatory Regression Service | Re-runs after rule changes, with an impact note | Per system per year, quoted | Tax-tech and ERP vendors, enterprises, government vendors |
| 4 | Red-team engagement | Professional adversarial scenarios; failures become regression tests | Fixed engagement plus optional recurring re-tests | Agents and systems that act |
| 5 | Enterprise AI quality programme | Several domains, teams and systems; managed delivery | Annual contract, quoted | Organisations running AI across business units |
| Licence | Benchmark licence | Use across your product's customers; versions refreshed | Annual licence, quoted | Software vendors, SAP partners, buyers comparing vendors |
| Data | Data Mission or expert case subscription | Commissioned datasets, or a recurring stream of newly certified cases | Fixed price or subscription, quoted | AI labs, model builders, agent companies |
Each package is complete in itself. You can stop after any of them.
Add-ons at any step
- Improvement data. Reasoning, preference or edge-case data aimed at exactly what failed, in your format. Never for a system whose benchmark score we publish. Improvement data.
- Expert review queue. Route low-confidence or high-value decisions to a qualified professional for a written answer with the rule, next business day. Expert review queue.
The four-week diagnostic, week by week
- Week 1 — you choose the workflow
the one where a wrong answer would cost most. We agree the domain (GST live; payroll and company law on request), the test environment and the acceptance criteria in a short specification.
- Weeks 1 to 2 — we build the case set
about 100 synthetic cases written by qualified professionals, each solved independently by two, reviewed and adjudicated, and tied to the law with a valid-as-of date.
- Weeks 2 to 3 — your AI is evaluated
in your test environment. Structured fields are marked by fixed rules; free-text reasoning is marked by qualified professionals.
- Week 3 — failures analysed
every failure with the correct answer and the rule behind it, grouped by cause: outdated rule, missed exception, wrong fact, wrong action.
- Weeks 3 to 4 — you receive the results
scores by capability, critical failures first, an executive report, the case-level file and a reusable regression set you keep, with a testing record for your evidence file. See a sample report.
- Then one question
shall we run this every time you change your AI?
How we price
We price against the professional work involved, never by the hour, by headcount or by compute.
- Complexity. Cases built around provisos, circulars and mixed facts take senior professionals longer than routine ones.
- Expert level. The capability and seniority required of the professionals who write, solve and adjudicate.
- Review depth. How many independent solvers and reviewers, and whether high-risk items get an authoritative-source check or an independent calculation.
- Rights. Creation and licensing are priced separately. Exclusivity is priced separately because it forecloses reuse. Custom cases built only for you cost more than cases drawn from the shared pool.
- Number of systems. Subscriptions and the Regulatory Regression Service are priced per system; benchmark licences per product.
- Answer format and delivery. Cases in SAP or other ERP document form need an extra specialist check; a licensed pack you run yourself is priced differently from a run we operate.
Acceptance criteria are agreed before work starts, and you receive a fixed price and a fixed deliverable list before anything begins. Fixed-price engagements are invoiced half on signing and half on delivery, payable within 30 days; subscriptions and licences annually or quarterly in advance. Prices exclude GST, which is added to invoices as required. Quotes are in INR for Indian buyers and USD elsewhere.
Four documents, no surprises
| Document | What it settles |
|---|---|
| Mutual NDA | Confidentiality both ways, from the first scoping call |
| Master services agreement | The standing terms between us: liability, confidentiality, intellectual property, term |
| Order form or statement of work | One per engagement: scope, acceptance criteria, rights, correction obligations, timeline and price |
| Data processing terms | Roles, what we receive (endpoint credentials, AI outputs), where it is stored, retention and deletion at close |
Licences state who may access, redistribute, publish or train on what. Benchmark licences also fix the version you may cite and how results may be described. Licences.
Land and expand
One workflow → one benchmark → more benchmarks → regression → red team → more business units.
Most teams start with the workflow where a wrong answer would cost most. Its case set becomes a regression set; further workflows add further benchmarks; red teaming follows once the basics hold; and the same approach then extends to other teams and systems. You decide at each step whether the next one is worth it.
What a scoping call covers
Thirty minutes, no preparation needed:
- What your AI does, and which decisions it makes on its own
- Where a wrong answer would cost you most
- How we can reach your AI in a test environment
- Which domain and topics to test first
- A fixed price and deliverable list, sent after the call
You will leave knowing whether a test is worth doing. If it is not, we will say so. Selling AI software? Ask about a benchmark licence covering all your customers. Software vendors and partners.
Packages — FAQ
What is the best first step?+
The free 25-case sample test if your AI handles Indian GST. It shows the quality of our work and the size of your AI's problem before anything is signed. Otherwise, the AI Quality Diagnostic on one workflow. Test your AI.
How much does it cost?+
The sample test is free. Everything else is fixed-price or subscription, quoted against an approved scope and deliverable list. We never publish amounts, and we never price by the hour, by headcount or by compute.
Do we have to sign a long contract?+
No. Each step is a complete piece of work you can stop after. Only continuous evaluation, the Regulatory Regression Service, benchmark licences and subscriptions run for a term, usually annual.
Can a vendor licence cover all our customers?+
Yes. A benchmark licence covers your product across your customers' deployments, with refreshed versions after rule changes and the right to cite results under agreed terms. Three ways to use a benchmark.
Do you ever advise against testing?+
Yes. If the scoping call shows a test is not worth doing yet, we say so.