Only what needs a human - with the evidence already gathered.
Everything the engine can't prove becomes a typed exception with side-by-side evidence, an AI hypothesis, and one-tap resolutions. Resolve it once, and it learns your rule.
A good reconciliation product is measured by what it refuses to guess. RakamHQ pushes everything uncertain into an exception queue - never into your books - so the ledger stays trustworthy and a human's attention goes only where it's needed.
Each exception arrives with the work already done: the order row, the provider row and the bank row side by side, a one-line hypothesis with a confidence score, and the resolutions you're most likely to pick.
What you get
- Typed exceptions - unmatched, mismatch, duplicate, stale, residual, and more
- Side-by-side evidence panes for every subject
- AI hypothesis with a confidence score - labeled, never auto-applied
- One-tap resolutions: merge, mark cash/COD, accept rounding, void duplicate
- Age badges so nothing rots - amber at 24h, red at 48h
- Every resolution writes an audit entry and can teach a rule
The exception queue, in detail.
Everything is typed
Whatever the engine can't prove is sorted into a type: unmatched, mismatch, duplicate, stale, residual, decomposition, and more. The type decides how the exception is presented and which resolutions are offered, so you never stare at a raw error. A mismatch shows the amounts that disagree; a duplicate shows the two payments; a residual shows the exact leftover.
The evidence is already gathered
Each exception arrives with the work done. The order row, the provider row, and the bank row sit side by side in evidence panes, every figure linked back to the source file line it came from. You judge the same facts the engine had, in one view, without opening four systems to piece the story together yourself.
A hypothesis, with its confidence
On top of the evidence sits a one-line hypothesis with a confidence score - 'likely the same payment recorded twice, 0.86'. It is labeled as a suggestion and never auto-applied, so the confidence guides your attention without ever writing to the ledger. You confirm or reject; the model only ever proposes.
The queue celebrates empty
The goal isn't a big queue - it's an empty one. When nothing needs you, the screen says so: sabai milyo. When something does, it's ranked by what matters, with an age badge - amber at 24 hours, red at 48 - so a real question about money never quietly ages out.
Resolve once, learn forever
A resolution is usually one tap - merge, mark as cash or COD, accept fee rounding, void a duplicate - and writes an audit entry as it goes. Tick 'remember this' and it becomes a rule scoped to your organization, so the same shape auto-resolves next time and the queue grows quieter as the engine learns your business.
FAQ
The exception queue,
explained
Common questions about the exception queue. Want it run on your own data first? The concierge reconciliation is free.
No. Anything below the confidence bar becomes an exception, never a silent entry. The ledger only ever contains proven or human-confirmed matches.
Related products
Get your free reconciliation.
Send us last month's eSewa report, Khalti export, bank statement and orders sheet. Within 48 hours we send back a one-page close - every payment matched, every fee computed, every settlement decomposed.
Files only · private upload link · nothing to install