A photographed receipt becomes a posted expense in SAP Business One — that is what this demo shows. TEKAI reads the receipt image, extracts the vendor, date, amount and tax, and drafts the expense entry for approval. The shoebox of paper stops being a data-entry backlog.
Field teams collect receipts on the road; accounts keys them in weeks later; by then half the context is gone — which trip, which client, which project the taxi ride belonged to. A fifty-rupee parking slip costs the same keying effort as a fifty-thousand-rupee invoice, and there are far more of the slips. Expense capture is the classic low-value, high-friction task: nobody’s judgement is exercised by typing a vendor name and an amount off a crumpled thermal print, yet someone does it hundreds of times a month. That is exactly the shape of work AI should absorb — repetitive, well-defined, and currently done late and grudgingly. And the cost shows up twice: once in the hours spent keying, and again in the accounts that are wrong at month-end because the backlog was cleared in a hurry — expenses landing in whichever account was quickest rather than the one that was right.
On a demonstration database: a receipt photo goes in, and the AI reads it — vendor, date, line amount, GST where the receipt shows it — and drafts the expense posting with the right account coding proposed. The approver sees the receipt image and the draft entry side by side; one approval posts it into SAP Business One with the image attached as proof. Illegible or ambiguous receipts do not post wrong — they flag for a human to handle, which is the difference between automation you can trust and automation you have to audit. Because posting happens close to the event rather than weeks after it, expenses hit the right period and the right accounts while the person who spent the money still remembers why — and field teams and travelling staff submit from wherever they are, so the paper never has to survive the journey back to the office. Every figure in the demo comes from a demonstration database, not customer data; the flow is the same for your own receipts.
Plain OCR has existed for decades, and it gives you exactly what it promises: text. “SHARMA FUELS 09-12 DIESEL 2,340.00” as characters is not an expense entry. The hard part is meaning — which vendor master that string corresponds to, which expense account diesel for a delivery vehicle belongs in, which tax treatment applies to the GST line. A language model grounded in your chart of accounts and vendor masters closes that gap: it reads the text the way OCR does, then interprets it the way your accountant does. That is the difference between “scanned” and “posted” — and it is why receipt scanning apps produced folders of images while this produces accounting entries. The grounding also keeps the proposals consistent: the same vendor string resolves to the same vendor master every time, instead of whichever near-match a tired keyer picked at month-end. The same capture-and-post pattern extends naturally to purchase bills, delivery notes and weighbridge slips — capture once, post instantly.
SAP Business One (SQL or HANA) · TEKAI at ₹5,000/company/month India, $150 global · your own AI subscription (Claude, ChatGPT or Perplexity) · approval stays human on every posting · permission-scoped access, nothing trained on your data (/tekai/security/).
Ordinary phone photos — genuinely unreadable receipts flag for manual handling rather than guessing.
Yes — tax components are extracted where the receipt shows them.
Submission flows are configurable; the constant is human approval before posting.
Drafts land against the vendor and date, so duplicates surface at approval.
It proposes coding from your chart of accounts; the approver confirms. To see the flow on your own setup, start with the five-minute Fit Test.
New to the category? Start with the overview: AI for SAP Business One, explained.