From you. Store the active rate sheet as a WordPress custom post type or option array: 30-year fixed, 15-year fixed, FHA, VA, jumbo, ARM tiers. Update daily or as often as your secondary desk does. The bot reads the latest values per conversation. Some brokers also pull from a pricing-engine API into a transient that refreshes hourly. Either way, the bot is never quoting a rate older than your rate sheet.

The bot does not do the arithmetic in natural language. SleekAI calls a registered tool function (a PHP function on your site) that computes amortisation, PMI thresholds, DTI, and cash-to-close exactly. The model decides what to compute and how to explain the result, the function does the math. This eliminates the LLM arithmetic-drift class of errors that makes generic chatbots dangerous in this category.

An affordability calculator is not personalised financial advice and is generally outside the licensed-advisor rules in the US (NMLS) and UK (FCA). The bot is explicit that it is an estimate and not a commitment, recommends a licensed loan officer for an actual loan estimate, and does not collect SSN or full credit-pull authorisation in the conversation. Your compliance team should review the system prompt for your specific licensing jurisdiction; the prompt is one editable file.

It can pre-screen, not pre-qualify. Pre-screen means estimating affordability from self-reported income and assets. Pre-qualify is a regulated term with specific definitions that vary by jurisdiction. Configure the system prompt to use 'estimate' or 'pre-screen' language exclusively, and reserve 'pre-qualify' for the formal loan officer step that includes a credit pull and verification.

Yes for both, with prompt branches. ARMs need extra fields (initial fixed period, index, margin, caps) and the math function computes the initial payment plus a worst-case adjusted payment. Refinances need the existing loan balance, current rate, and remaining term to compute break-even months on a rate-and-term refi. Both flows are added by extending the system prompt and the tool function; the user-facing conversation stays in the same widget.

Stored as ZIP-code-level averages in a WordPress post or option array, or pulled from a real-estate-data API per address. The bot defaults to an average if the user gives no specifics, but offers to refine if they have the actual tax bill. Insurance is similar, with a flag for flood-prone areas that triggers an additional disclosure and a higher default rate. Both add to the all-in monthly payment.

Yes. At any point the bot can offer a 'save this scenario' action that captures the inputs and the calculated outputs to a custom post type or Forminator entry, optionally emailing the visitor a PDF summary. A loan officer can open the saved scenario, see exactly what numbers the visitor was looking at, and follow up with a real loan estimate that matches. This is the highest-converting flow in the category.

Yes if you supply local rules. UK-style mortgages have product fees and 2- or 5-year fix periods. Canadian mortgages have 25-year amortisation defaults and stress-test rates. EU markets vary widely. The math function is generic; the program rules and stress tests live in your WordPress posts. Run separate bots per market via multibot if you serve more than one, with distinct rate sheets and compliance text per region.