How QuamIQ Makes Every Tally Answer Verifiable and Auditable

There is a quiet trap in every conversation about AI and financial data. Vendors reach for the language of perfection — flawless, always right, guaranteed — because that is what a nervous buyer wants to hear. But no finance leader who has actually closed a set of books believes in perfection, and claiming it is the fastest way to lose their trust. The number that survives an audit is not the one that was promised to be right. It is the one you can trace, reconcile, and defend.

So QuamIQ is built around a different standard: not "trust us," but "check us." Every answer QuamIQ gives you exposes the exact SQL it ran, reconciles to the paise against your own live Tally data, and can be verified by your CA without them needing to understand a single line of the AI. This post explains how that works — the mechanism, the IP layers behind it, and what auditability looks like in practice for an Indian SME.

Verifiable Beats Perfect

Consider two tools. The first tells you your net profit for the quarter is ₹63,73,628.33 and offers nothing else. The second tells you the same figure, then shows you the query it ran, the ledgers and voucher types it included, the date range it filtered on, and a line that reconciles that figure against what Tally itself reports — matched to the paise. Even if both produce the identical number, only the second one is usable in a board pack or an audit. The first asks for faith. The second offers proof.

This is the core of QuamIQ's positioning. We do not ask you to believe an answer is correct. We give you everything you need to confirm it — and to hand that confirmation to your auditor. Correctness you can prove is worth more than correctness you are asked to assume.

What "Reconciled to the Paise" Actually Means

Every answer QuamIQ returns is checked against the live Tally database it queried, before you see it. Where a corresponding Tally report exists — a P&L figure, a GST total, a ledger balance — QuamIQ places its own figure next to Tally's and confirms they agree. Here is what that reconciliation looks like on a real set of books:

The status column is the point. It does not report a percentage or a score. It reports whether QuamIQ's figure ties out against your own Tally — the same source your CA would open to check it. When a figure reconciles, you can quote it. When it does not, QuamIQ tells you rather than showing it, and reworks the query before returning an answer.

Why Tally Makes This Hard — And Why That Is the Whole Point

Getting a Tally figure to reconcile exactly is genuinely difficult, and understanding why explains what QuamIQ's IP is actually for. Tally is not a flat relational database you can point plain SQL at. It carries decades of Indian accounting convention baked into its structure, and each convention is a place where a naive query silently produces the wrong number.

Recursive group hierarchies

Tally organises accounts as nested groups. "Sundry Debtors" is a group that may hold sub-groups, which hold ledgers. A total-receivables figure must walk that entire tree recursively — miss one sub-group and you undercount. A tool that queries flat tables will look confident and be wrong.

Voucher-type logic

Every transaction is a voucher of some type — Sales, Purchase, Receipt, Payment, Journal, Contra. Many calculations depend on including or excluding the right ones. A cash-flow figure must exclude Contra vouchers, which merely move money between bank and cash; include them and you double-count. This is accounting knowledge, not SQL.

Dr/Cr conventions and opening balances

Tally stores balances with debit/credit sign conventions that differ from how a database column behaves, and it records opening balances in a structure distinct from regular transactions. Read a credit balance as a plain positive, or sum a date range without handling opening balances correctly, and every period-based figure drifts.

GST component decomposition

GST lives across many ledgers — CGST, SGST and IGST on both output and input sides, plus adjustment ledgers for reversals, RCM and ISD. A correct GST liability aggregates the right combination with the right signs. Miss one ledger or flip one sign and you have a wrong compliance number heading into GSTR-3B.

The IP That Makes Answers Trustworthy

QuamIQ handles all of the above not with one clever query but with a stack of validation layers, each with a specific job in making an answer verifiable rather than merely plausible.

The Tally Knowledge Graph

This layer encodes 18 accounting rules — the Dr/Cr conventions, the recursive group traversal, the voucher and opening-balance quirks above — so that queries respect how Tally actually models money. It also maps your specific setup: your ledger names, your groups, your cost centres, so "what is Apex Corp's outstanding?" resolves to the right ledger even though every Tally install names things differently.

Execution-Guided Selection

Before any query runs, this layer validates it against your live Tally schema — the real table and field names present in your database. A query that references something that does not exist is caught and corrected before execution, not after it returns a confident but wrong answer.

The Business Specialist Engine and Semantic Result Critic

Once a figure is computed, this layer checks whether it makes business sense before you see it. Net profit above gross revenue, GST output exceeding what the sales base could produce, a month 10x the trailing average — these are the shapes of a query gone wrong, and they trigger a rework rather than a display.

Correction Memory

When you correct an answer, QuamIQ remembers the exact business context — which ledger, which voucher type, which date logic. That correction is applied to similar future questions, and after three corrections on the same pattern it becomes a permanent template. The system gets more reliable with use, and it does so without a developer in the loop.

Auditability: Every Answer Shows Its Working

Reconciliation confirms the number. Auditability lets someone else confirm it independently. For every question answered, QuamIQ exposes:

• ●The exact SQL query executed against your Tally database
• ●The ledgers and voucher types included or excluded
• ●The date range and filters applied
• ●The reconciliation status against your live Tally figure

This is what turns QuamIQ from a black box into a glass box. Your finance team can re-run the logic themselves. Your CA can review the methodology without touching the AI at all — the SQL and the Tally reconciliation are the whole story. During due diligence or a statutory audit, "here is the query and here is where it ties to Tally" is a far stronger answer than "the software said so."

On-Premise: Freshness Is Part of Correctness

A figure can be computed perfectly and still be wrong if it is computed on stale data. Cloud BI tools that sync Tally on a schedule always work on a snapshot — hours or days old. QuamIQ runs on a machine on your own network and queries your live Tally directly, so every answer reflects the current state of your books, including a voucher entered a minute ago. And because it is on-premise, your numbers never leave your server — for a genuinely new question, only the question text and the database schema go to the cloud model, never the financial data itself.

"We do not ask you to trust the answer. We show you the SQL, reconcile it to your Tally to the paise, and let your CA check it. Verifiable beats perfect."

Conclusion

Auditability is not a feature bolted onto QuamIQ — it is the design principle underneath it. Validated across 500+ carefully designed questions and 458 tested across 12 real-world personas, QuamIQ was built so that no answer arrives as a claim you have to take on faith. Each one arrives with its working attached.

For a business making daily decisions on its own money, that is the standard that matters: not a promise of perfection, but a number you can trace, reconcile, and defend. If you want to see your own Tally figures answered that way — with the SQL and the reconciliation in view — book a demo at demo@quambase.com.