Responses
Sent to nine institutions on 4 September. Replies arrive as email, portal notification or a PDF letter — the AI reads whichever form it takes and normalises it into the same six fields, so they can be compared at all.
Every institution, every reply
Normalised by the AIApprove and decline is not the real taxonomy
Most systems only handle yes and no. In practice you get five outcomes, and three of them need a person: approved in full, approved at a lower amount, approved with conditions, referred for more information, and declined. Capitec came back at R1 600 000 against R1 755 000 requested — that is not an approval the buyer can use without finding another R155 000, and it must never be presented as a straight yes.
No reply is also an outcome. Discovery is two days past its own turnaround and nobody has noticed. The clock is the only thing that catches that.
Compare and rank
Internal view. Ranked on whichever measure you choose, because there is no single measure that is right for every buyer.
Look at what changes when you switch the sort
Capitec shows the lowest monthly payment on the table — because they offered 30 years, not 20. Over the full term that is R1 181 830 more than Nedbank, and they would only lend R1 600 000, so the buyer needs R445 236 in cash instead of R292 236. Rank by monthly and Capitec is first; rank by total cost and it is last. A buyer looking only at the monthly figure would take by far the most expensive loan here.
No single ranking is honest for every buyer. That is exactly why the buyer's own screen sorts but does not recommend.
Total cost over the full term
Cash needed on the day
Deposit plus costsWhat the buyer sees
The same facts, alphabetical, with sorting they control and no recommendation from us.
Why this screen is deliberately different from the last one
Your consultants need a ranked list to do their job — that is operations, and it is fine. The moment a ranked list is put in front of the buyer with a best option marked, it stops being information and becomes advice, which needs an FSP licence and a record of advice.
So the buyer's screen lists alphabetically, lets them sort for themselves, and marks nothing as best. Same data, different stance. Keep this split in the software, not in a training manual — people forget training.
Every offer we received is shown. We do not rank them and we do not recommend one over another. If you would like advice on which suits your circumstances, tell us and we will refer you to a licensed financial adviser.
Declines
Captured, coded and counted. Decline reasons are the most valuable data this business generates — they tell you who not to send to next time.
This application
Across the last 12 months
242 submissionsThis is the pre-flight engine, built from your own history
Suburb concentration is the top decline reason at 31%. Once you know Standard Bank keeps declining in a suburb, you stop sending there and tell the buyer up front — instead of costing them three days to learn it. The bank-rules file writes itself out of this table, which is something no new originator can buy and you accumulate from day one.
Lender panel
Every institution in South Africa that writes home loans, not just the big four. You tick who you are accredited with, and the system only sends where you can actually be paid.
Verify this list before you rely on it
The first ten come straight off BetterBond's own form, which names the credit providers they submit to. The rest are from general knowledge of the market and I am not certain all of them currently write home loans or take originator business — treat the unconfirmed ones as a list to phone, not a list to code.
Prototype. The applicant, the responses, the rates and the decline history are all simulated. Instalments are calculated at a prime rate of 10.75% over 20 years and exclude credit life, initiation and monthly service fees. Lender panel membership, commission terms, turnaround times and lending criteria must be confirmed with each institution.