Sixty percent of Indian SMBs still process invoices by hand: reading a PDF, retyping the GST fields into a ledger, one invoice and one line at a time. That's not an edge case. That's the default for most small firms doing their own books.
A hardware distributor's accountant found a mismatched GSTIN three days before a monthly return was due, buried in a stack of forty vendor invoices that had already been retyped once each. Fixing it meant re-checking every entry against the original PDF, not because any software had failed, but because nobody had caught the typo the first time either. The return went in on time. The three days it took went nowhere near what they were supposed to.
That's the pattern underneath most back-office backlogs. Because errors surface at reconciliation instead of at entry, fixing them always costs more than the original mistake would have. Because that fixing eats the hours meant for chasing payment, the invoice pile and the collections pile grow together instead of separately. Nobody planned for either to happen; they're just what retyping produces often enough.
What manual back-office work actually costs
The instinct when a backlog like this shows up is to hire. More hands, more invoices cleared. It rarely works, because the bottleneck was never headcount, and a new hire trained to retype the same fields just adds one more person doing the same unpaid double-entry.
Where the time actually goes
Support follow-ups and scheduling have the same shape as invoicing, just with a different starting document. A customer asks about a delivery date over WhatsApp; the reply gets typed once in the chat and then typed again into a spreadsheet so someone remembers to check back Thursday. Automating that second step, not the conversation itself, is what gives the hours back.
None of this runs on its own. Someone still has to decide, once, what counts as an exception worth a human's attention: a vendor's first invoice, an amount over a set threshold, a delivery date that doesn't match the purchase order. Writing that list down is a half-day of work, not a project, and it's the part teams are most tempted to skip because it feels like the boring half of "getting AI." Get it wrong and the system just processes bad judgment faster than a person would have. Get it right, and every invoice or follow-up that doesn't match those rules is the only one anyone needs to look at by hand.
A small team testing this doesn't need to automate the whole back office at once. Pick the single task redone most often this week, whichever it is, and automate only that one first. Watch what happens to the backlog sitting next to it before touching anything bigger.