Auto-sort expenses into categories

A contractor paying the same electrician, the same cement supplier, and the same site labour contractor every month used to pick a category from a dropdown for every single payment, even though the answer was almost always identical to the last one made. A year of that adds up to hours spent re-deciding something that was already decided eleven times before, without changing.

Expense categorisation learns from how the owner, or their CA, has categorised similar vendors and descriptions in the past, and suggests the same category the next time a matching transaction appears. The owner accepts or overrides each suggestion, and a correction is remembered for next time — but the suggestion is a time-saving draft, never a tax determination. Whether an expense is actually deductible, or how it should be treated for tax, stays entirely the CA's call to make.

Category suggestions exist to save the owner a repetitive decision, not to make a tax decision on their behalf. Khata never decides whether an expense is deductible or how it should be tax-treated — a category label like "office supplies" or "site materials" is a bookkeeping bucket the CA later reviews and, where the tax treatment requires it, reclassifies or adjusts as needed. The suggestion engine's only job is pattern-matching against how this owner and this CA have categorised similar things before; it has no independent view of tax law.

The learning mechanism carries a specific, quiet risk worth understanding clearly. A single wrong correction can miscategorise a whole vendor's future expenses until someone notices the pattern. If an owner mistakenly recategorises one supplier's payment under the wrong bucket, every subsequent payment inherits the same wrong suggestion, compounding the error rather than catching it. A periodic CA review is meant to surface exactly this drift — not because the system is unreliable, but because a learned rule reflects what was clicked, not necessarily what was correct.

The more common and less dramatic version of this same risk is simple drift between what an owner picks day to day and what a CA would actually prefer for that category. An owner categorising quickly between customers optimises for speed; a CA reviewing months later optimises for correctness against the tax treatment the category implies. Neither is wrong exactly, but they can diverge, which is why the suggestion is framed as a draft awaiting the CA's review, not a settled classification once accepted.