ZynoviqWhen your AI is unsure, a qualified professional decides.
Testing tells you how often your AI is wrong. Some decisions are too costly to leave to "usually right": a large reverse-charge entry, an unusual credit claim, a case the AI itself is unsure about. Route those low-confidence or high-value decisions to a queue reviewed by qualified professionals. You get a written answer and the rule behind it, next business day.
How the queue works
- Your system flags a decision
By a rule you set, such as a value threshold or a risky category, or because the AI's own confidence is low.
- The case arrives in the queue
With the facts and the AI's proposed answer, under your data processing agreement, visible only to named reviewers.
- A qualified professional reviews it
And returns a verdict: agree, disagree with the correct treatment, or escalate. The professional is named as the giver of the opinion on that case.
- You receive the written answer
With the rule behind it, within the next-business-day target stated in your contract.
- Your team decides
You stay in control of the final entry. The reviewed case can join your regression set, so the AI is tested on it after every change.
What you get back for each case
| Field | What it contains |
|---|---|
| Verdict | Agree · Disagree · Escalate |
| Correct treatment | The answer a qualified professional gives on these facts, in the structured fields your system uses |
| Rule and authority | The section, notification, circular or written policy behind it, with its valid-as-of date |
| Reviewing professional | Named as the giver of the opinion on each case, with credential and role |
| Time taken | Against the next-business-day target in your contract |
| Regression candidate | Whether the case should join your regression set, with your agreement |
The points that matter
- Next business day, as a target stated in the contract.
- A named professional is the giver of the opinion on each case.
- Each reviewed case can join your regression set, so a decision the AI could not make once becomes a test it must pass later. AI regression.
- Not real-time. Real-time review is planned, and only after demand.
- Not tax advice to your end customers unless separately agreed.
- Priced per review or as a monthly allowance of reviews, quoted. How we price.
How this differs from testing
| Testing | Expert review queue | |
|---|---|---|
| Cases | Our synthetic cases | Your real decisions |
| Timing | Periodic | As decisions arise, next business day |
| Purpose | Measure how good the AI is | Decide the case the AI could not |
| Data | None of yours | Your case facts, under a data processing agreement |
Scope and limits
- For flagged cases only. Not a team reviewing all your volume, and not staff placed in your office.
- Real case data requires a data processing agreement under India's DPDP Act, access limited to named reviewers and agreed retention. Security and data handling.
- Review is professional judgment on a specific case, not an opinion on your wider position, and not a certification of your AI.
- Domains: GST and Indian indirect tax first; payroll and company law on request, where we have qualified professionals for the workflow. Domains.
Expert review queue — FAQ
Which decisions should go to the queue?+
Those above a value you set, in categories you consider risky, or where the AI's own confidence is low. The routing rules are yours; we help you draft them from the failures found in your evaluation.
Is this real-time?+
No. The queue is reviewed by people, with a next-business-day target stated in your contract. Real-time review is planned, and only once there is demand for it. For decisions that cannot wait, your system should hold or escalate internally.
Does the queue use our real data?+
Yes, unlike testing. That is why it needs a data processing agreement under India's DPDP Act, access limited to named reviewers and agreed retention. Data processing.
Is a reviewed case tax advice to our customers?+
No, unless separately agreed. The opinion is given to you on a specific case. It is not tax advice to your end customers and not an opinion on your wider position.