Nivo launched Nivo Collect this summer as an off-the-shelf AI service designed to take the document gathering, checking and chasing involved in packaging mortgage and commercial finance cases off brokers’ desks.

Just weeks after launch, more than 15 brokerages are already using the service on live cases, with another 10 firms in onboarding. Mortgage Soup spoke to Matthew Elliott, co-founder and chief commercial officer at Nivo, about the thinking behind Collect, what Nivo is learning from its early adopters and what comes next.

Mortgage Soup (MS): You launched Nivo Collect over the summer. What was behind the launch?

Matthew Elliott (ME): There’s obviously a huge amount of interest in AI, but I think the conversation often starts in the wrong place. People ask, “Where can we use AI?” when the better question is, “Where are our people spending time on work they shouldn’t be doing?”

For brokers, collecting documents is a really obvious example. You ask for bank statements, ID, accounts or whatever else a case needs. Some of it comes back, some of it doesn’t. Something is missing a page, something doesn’t match what you were told, so you check it all and go back again. You can end up with 10 or 15 rounds of email on a case that was never particularly complicated.

Nobody became a broker because they wanted to spend their day chasing PDFs.

That’s what we wanted Collect to solve. A broker emails in what they know about the case and Collect works out what is required, requests it, checks what comes back, follows up for anything missing and returns a packaged case ready for submission.

The important thing was to make it something firms could actually use, rather than another tech project they had to build themselves. It works over email, there isn’t a major integration project and brokers don’t need to start writing prompts or designing workflows. The technology is doing the work in the background.

MS: What has the response been like since launch?

ME: Really positive, and faster than we expected.

Within the first seven weeks we had more than 15 brokerages live and another 10 going through onboarding. More importantly, these aren’t firms playing around with a sandbox or putting dummy cases through it. They’re using Collect on real mortgage and commercial finance cases.

That’s important because you very quickly get away from the perfect examples you can create in a product demo. Real customers forget things. Documents arrive in the wrong format. Cases change. Some clients respond immediately and others need chasing three times.

That’s the reality brokers deal with every day, so that’s the environment the service has to work in.

We knew the administrative problem was there. What has probably surprised us is how quickly firms have understood the proposition and said, “Yes, that’s a job I would happily stop doing.”

MS: Why do you think Collect has struck a chord with broker firms?

ME: Because it is quite easy to understand what you are buying.

The AI market has become incredibly noisy. There are a lot of products where you can spend 20 minutes on a website and still not be entirely sure what the thing actually does.

Collect is much simpler. You have a case that needs packaging. You send it in. The service gathers and checks what is required, chases what is missing and gives you the case back packaged.

That clarity is important because brokers aren’t really interested in buying “AI”. They want to know whether something is going to save them time, remove cost or create capacity in the business.

And I think capacity is really interesting for ambitious broker firms. If you take around five hours of administration out of every case, that doesn’t just make the existing process a bit faster. It means advisers and administrators have time to work with more customers and progress more business without necessarily adding another person every time volumes increase.

MS: Does working with live cases change the way Nivo is developing its proposition?

ME: Absolutely. Requirements written in a workshop are guesses. Once you have been running real cases for a few months, you start dealing in facts.

You see where customers get stuck, which information is routinely missing, where a broker needs to intervene and where the technology can comfortably handle something without them. Some of the most convincing moments are small ones. A pack comes back with a £95 a month commitment flagged that nobody mentioned at intake, with the evidence attached. That is usually the point where a firm stops thinking of it as a clever demo and starts treating it as a colleague.

That is one of the big reasons we wanted to get Collect into live use quickly rather than spend another year trying to design for every possible edge case before anybody touched it.

There is sometimes a tendency with technology projects to say, “We can’t launch until it does 100% of the job.” I think that’s the wrong test. If it reliably removes 80% of a repetitive task and hands the unusual 20% back to somebody who knows what they are doing, you’ve already changed the economics of that task.

We’ve been taking the feedback from the firms using Collect and feeding that straight back into the proposition. That’s a much better development process than us sitting in Manchester trying to imagine every type of case a broker might encounter. And, at the same time, our clients are seeing an immediate and positive impact on their business.

MS: What would you say to brokers who are still reticent about working with AI?

ME: I wouldn’t tell anybody they need an “AI strategy”.

I’d tell them to look at their business and find the task their team repeats every day that consumes time without adding much value.

Measure it. How long does it take? How many times do you chase a customer? How many hours are disappearing into administration every week?

Then look at whether technology can take some of that work away.

That’s a much more sensible starting point than deciding the business needs AI because everybody is talking about it.

And you don’t need to hand judgement over to a machine. I think that distinction is really important. The value a broker brings is understanding the customer, structuring the deal, knowing the market and making judgements about where that case belongs. We’re not trying to remove that.

Collect is aimed at the work around those decisions: asking for documents, checking them and chasing what hasn’t arrived.

Of course, firms should also be asking proper questions around data, security and control. Collect has been designed specifically for regulated financial services, with customer data kept within the EEA, no use of client data to train AI models and a full audit trail of AI interactions. We also hold ISO 27001 and ISO 42001 certifications, which cover information security and AI management.

You should absolutely understand what a piece of technology is doing before putting it into your business. But you can do that by starting with something narrow and measurable rather than commissioning some enormous transformation programme on day one.

MS: So, what comes next for Nivo Collect?

ME: The immediate job is to keep getting firms live and keep learning from the cases coming through.

Once that first group has enough completed cases behind it, we’ll also be able to start putting some proper numbers around the impact, particularly how much time is being taken out of packaging each case. That’s the measurement that matters to me.

We’re also working on making Collect much easier for brokers to experience for themselves.

We’ve just started opening up a test version where somebody can email a case straight into a generic address and interact with the service. So, rather than us standing at an exhibition or on a webinar telling somebody how clever it is, they can send something in and see what comes back.

I think that will be quite powerful because this stuff becomes much easier to understand when you actually use it.

Longer term, I think that’s where the AI conversation generally is heading. The novelty will disappear. People won’t particularly care which model sits underneath something or whether the word AI is in the product name.

They’ll ask: what job does it do, how much work does it take off my team, and can I trust it?

That’s probably a much healthier place for the market to get to.