Refinancing home loan · a prototype direction

Your exact refinancing savings, computed in the chat

The live journey does three things today, all visible on public pages (July 2026): it advertises a personalised number ("You could save S$4,668* per year") yet delivers it through a three-step lead form and a specialist callback; it offloads the savings math to seven separate calculators; and it starts every specialist conversation from a cold form fill.

This prototype answers in the conversation instead. A deterministic engine computes every figure (amortization month by month, legal and valuation fees, the lock-in penalty); the model handles the questions and the plain-English explanations, and it cannot calculate. If the numbers say stay put, the report says stay put. The specialist still gets the case, now with a pre-qualified summary attached. See the eval results.

Tips. Click a persona below the chat to pre-fill a starter message; edit or send as-is. The trace panel on the right shows every tool call, every deterministic-wrapper firing, and any server guard that overrides the model. Reset starts a new conversation.

view evals →
Hi — I can work out whether refinancing your home loan actually saves you money, using deterministic math rather than a callback. I need five things: your outstanding loan amount, current interest rate, remaining tenure, whether you're still in lock-in, and HDB or private. Share what you know and I'll ask for the rest.
Try as:
What a win looks like

Ship this as an A/B against the current three-step form. The metrics are the share of refinancing visitors who reach a specialist contact, and the specialist's close rate on engine-summarised leads versus cold form fills. The read happens two weeks after launch; if neither number moves, kill it.

The direction I'd test against it

Direction B drops the conversation entirely: the existing rate table, with each row expanded to show that user's computed monthly delta and break-even from the same engine, plus one "check my numbers" field set. It is cheaper to build with less to go wrong, and the A/B decides between them.