Voice AI for mortgage leads

Firms that buy leads have a particular arithmetic problem. You pay per lead whether or not you reach the person, contact rates on purchased data are unforgiving, and the value of a lead decays by the hour. Anything that increases the proportion you actually speak to changes the economics directly.

That is the honest commercial case for putting an automated voice system on lead handling, and it is a narrower case than the marketing around these tools suggests.

Know what your leads are before you automate them

The right answer depends entirely on where the leads come from, and firms often lump together sources that behave completely differently.

Exclusive leads from your own website or a directory listing, where the person deliberately approached your firm and expects a call. High intent, high tolerance, and a poor candidate for automation on the first contact. These people want an adviser and you should give them one.

Shared or aggregated leads, sold to several brokers at once. Whoever calls first has an enormous advantage, contact rates are low, and the person may not remember filling anything in. Volume is high and each individual lead is worth little. This is the strongest case for automated first contact, because the alternative is that half of them are never called at all.

Aged data and re-marketed lists. Low intent, high annoyance risk, and the sort of contact most likely to generate a complaint. Automating this is not a productivity decision, it is a decision about what kind of firm you are.

Referrals from an estate agent or accountant. Never automate. The relationship is the asset and the introducer will hear about it.

What the technology can realistically do here

Reach people quickly and in parallel, which is the whole point. A system can attempt fifty numbers in the time an adviser attempts five.

Establish the basics: whether the person still wants advice, roughly what for, whether they are buying or remortgaging, when they are looking to proceed, and when they can talk.

Put an appointment into a real diary and confirm it.

What it cannot do is qualify in any meaningful sense. A caller saying they are self-employed tells you very little. Whether their accounts support the borrowing is not something a voice agent can establish, and asking it to try produces a confident answer you cannot rely on. Keep the questions factual and few.

It also cannot rescue a bad lead source. If the data is poor, faster contact just means you discover that sooner, which is worth something but is not the transformation being sold.

The consent question, which is your problem not the vendor's

Calling purchased data brings obligations that sit with the firm making the call.

You need to know how consent was obtained for each source and be able to evidence it. Ask the supplier for the wording the individual actually saw, the date, and the source URL. A vague assurance that the data is opted in is not evidence, and PECR rules on unsolicited marketing calls apply regardless of what a supplier told you.

Screen against the Telephone Preference Service unless you have a specific, recorded consent that permits the call.

Automated calling has additional exposure. Systems that dial and play a recorded message without a live person are subject to stricter rules than person-to-person calls, and the distinction between an interactive agent and an automated recorded call is not always as clean as a vendor claims. Take advice on your specific setup rather than assuming.

Tell people it is an automated system when they answer, promptly. Beyond fairness obligations, a firm that is coy about this while calling purchased data is stacking two reputational risks on top of each other.

Measuring it honestly

Vendors will offer you a contact rate. It is the least useful number available.

The metrics that mean something run further down the funnel. How many attempted calls resulted in a conversation with the intended person. How many of those resulted in a booked appointment. How many of those appointments were attended rather than no-shows. How many attended appointments became an application. And what the cost per application was, all in, including the lead cost and the software.

That last figure is the only one that tells you whether this was worth doing, and it is the one that takes three months to get. Run the comparison against a control group if you can: a portion of the same lead source handled the way you handle it now. Without a control you are comparing this quarter's market to last quarter's and calling the difference a result.

Watch the no-show rate especially closely. An appointment booked by a machine with someone who was mildly agreeable on the phone is not the same asset as one booked by an adviser who built two minutes of rapport, and a rise in booked appointments that is entirely offset by a rise in empty diary slots is not progress.

Where it damages you

Three failure modes are worth naming.

Calling too soon and too often. A lead contacted six times in a day by an automated system will remember your firm's name, unhelpfully.

Sounding evasive. If the person asks whether they are speaking to a real adviser and gets a non-answer, the call is over and so is the relationship.

Reaching someone in difficulty. A proportion of mortgage enquiries come from people under financial strain. Automated qualification handles that badly, and Consumer Duty expects better of you.

A sensible position

Use it as a triage layer on high-volume, low-intent sources, to get to a human conversation faster than you otherwise would. Keep it away from referrals and from anyone who approached your firm by name. Be plain about what it is, know where your data came from, and judge it on applications rather than on contact rates.

If your lead volume is modest, the honest answer is that the money is better spent on being able to answer the phone.