03 · The build
Six things that take
years to replicate.
A complete read is not a feature. It is an architecture — six layers, each of which represents work that cannot be shortcut by a model, an integration, or a connector. Together they describe why the read happens here, and not elsewhere.
The largest AI personal finance product launched in 2026 reaches its users via a bank-data connector with no GCC coverage. The structural gap remains open.
Moat 01
GCC-native statement parsing
Per-issuer parsers for the major GCC retail banks and card products — current accounts, credit cards, Diners, MasterCard, Shariah-compliant lines. Layout validation on import: the system refuses to ingest an unrecognised statement rather than guess at fields. No global tool has built this. Building it is twelve to eighteen months of data work that cannot be shortcut.
Why it holdsGCC retail banks do not expose open APIs at the depth required. PDFs are the source of truth and the parsing layer is the entry to the market.
Moat 02
Provenance-first data model
Every aggregated figure resolves to a row in the transaction register, and every row carries the source PDF hash, page, line index, raw amount, raw description, and import timestamp. Fabrication is architecturally impossible — a number that cannot resolve to a page does not appear.
Why it holdsMost AI-assisted financial tools cannot answer "where did this figure come from." Qleerly answers it for every figure, every time.
Moat 03
Clarification engine & benefit separation
An eight-question structured sequence — salary structure, employer loans, insurance, group policy, endorsements, contract and end-of-service, education reimbursement, annual bonus — field-validated on real expat compensation data. Separates employer-reimbursed items, investments, and insurance from real expense. This is the core IP.
Documented caseConverted an apparent salary deficit into a verified monthly surplus and an investment rate >15% — same data, structured reading.
Moat 04
Employer benefit configuration intelligence
A knowledge base of benefit structures across GCC industries — oil and gas, aviation, banking, construction — pre-loaded before a single employee logs in. School fee reimbursement caps, housing allowance bands, car benefit policies, end-of-service formulae. Each new client deepens it. Information not publicly documented; cannot be replicated without direct employer relationships.
Why it holdsConnecting a bank feed tells you what was spent. It cannot tell you what is reimbursable. Only employer-specific knowledge can.
Moat 05
Multicurrency & multi-account unification
QAR, GBP, USD, NGN, INR, PHP and others unified into one base-currency view with live conversion. Home-country accounts ingested alongside host-country accounts. Investment statements — endowments, offshore bonds, retirement plans — classified as wealth, not expense, so the contribution does not appear as a cost. Remittances tracked as structured cashflow.
Why it holdsNo global budgeting tool reads GCC-common investment plans. Without this layer, the financial picture is materially wrong.
Moat 06
Privacy architecture, not privacy policy
For the personal product: statements never leave the device; no telemetry; no cloud sync; AI categorisation receives redacted merchant strings only, with account numbers and names stripped by a unit-tested redaction layer. For the workplace product: individual employee financial data is technically inaccessible to employer administrators — a data architecture, not a policy promise. Employers see aggregate, anonymised signal only.
Why it holdsAs AI finance products face increasing privacy scrutiny, architectural separation is a trust differentiator that policy commitments cannot match.