What a forward-deployed team finds inside a mortgage company

The forward-deployed engineer is having a moment. Palantir made the title famous, and the AI labs have picked it up: put your own people inside the customer, wire the product into how they actually work, and stay until it runs. It is an old model. IBM did it with mainframes. What is new is who else is doing it.

Moburst is a marketing agency. It started in 2013 doing app store optimization and has since become a full-service digital agency, mostly through acquisitions. Earlier this year its co-founder and COO, Lior Eldan, sent a team on-site to a mortgage company for several days with a brief that had nothing to do with marketing: walk through what every team does, and find where AI could take work off people's plates.

On Between Two COOs he explained why an agency would do that, what the team found, and the operating rule that came out of it.

Why a marketing agency went on-site

Moburst had decided AI was existential and gone all in internally: an AI champion in every department, one-day hackathons where whole teams build an MVP, a central team building products. Then it brought that energy to clients.

The reaction, in Eldan's words, was “what's AI have to do with us?”

That was the gap. His clients knew AI was a thing and assumed it was good. They could not see how it applied to anything outside marketing, or to their day-to-day business at all. Nobody was going to close that gap from a slide deck. So Moburst did a few on-site days with one client, a mortgage company, and walked the floor.

What the team found

Processes that had existed for decades. That was the first thing.

Then the specifics. Team members whose job was to look through forms to make sure they had been filled in properly. Account executives who take customer calls on weekends, from home, with no context in front of them about who is calling or where the file stands. Customers who sign a document, photograph it on their phone, and send back something blurry, which starts a fifteen-minute back-and-forth with the account executive to get a usable copy.

None of that is exotic. It is the ordinary residue of a business that works, has always worked, and has never had a reason to look at itself. And all of it sat inside a regulated environment where everything has to run on-prem, which rules out the easy answer of pointing a cloud model at the problem.

The first fixes were not AI

This is the part worth sitting with. The most obvious win was a small tool that tells the customer, at the moment they scan, that the image is blurry and how to fix it. That is a form. It is a piece of digital product work Moburst has done a thousand times for marketing clients, applied to an operations problem.

Eldan's framing was that the agency got there by “coming from a digital perspective on how to do this right and layering AI on top of it.” The AI is the second layer. The first layer is noticing that a human being is squinting at a phone photo, and that the customer was never told what a good scan looks like.

The same patterns showed up at other clients. Call centers. Intake processes. What happens when someone calls after hours, and what experience they get. Moburst is now standing this up as a proper AI transformation service, and Eldan is clear that they are not alone. The AI companies themselves are launching forward-deployed offers, because the barrier to using AI is getting it into the hands of the people who do the work, in a form they can use on a Tuesday.

The rule underneath it

Asked what a COO should take from all this, Eldan did not talk about models. He talked about process.

“People run to automate processes which aren't fully fleshed out, and then it doesn't work, and then they blame the AI.”

His rule: make the process work manually first. The right people, the right hand-offs, the outcome you actually want. Only then automate it. And when you do automate, context is everything. An agency deliverable moves through strategy, creative, media, and digital before it ships, so the shared context layer matters as much for the AI as it does for the humans.

Michael's version, from the top of the episode: if it is running well, you don't think about it. When it is not running well, you do. A forward-deployed team is a way of paying attention to the parts of a business nobody has thought about in years.

If you run a services business

A few things follow, none of them about hiring engineers.

Start by walking the process, in person, with the people who do it. The forms, the phone photos, and the weekend calls do not show up in a discovery call.

Expect the first deliverable to be embarrassingly simple. A better scan prompt. A context screen for whoever picks up the phone. The AI work comes after, and it is better for having something clean to sit on.

Assume the platforms will do the mechanical parts. Eldan expects agencies to run smaller teams. What does not shrink is the judgment about what to build and in what order, which is exactly what the on-site days are for.

And do not automate anything you have not first made work by hand. It is the least glamorous rule in the episode, and the one most likely to save you a quarter.

FAQ

What is a forward-deployed engineer? An engineer, or increasingly a small team, that embeds with a customer to integrate a product into the customer's actual operations and stays until it works. IBM used the model to install mainframes and Palantir made the title famous. In Moburst's version the embedded team is marketers and builders rather than software engineers, and the product is whatever the client's process turns out to need.

Why did Moburst, a marketing agency, start embedding with clients? Because its clients could not see how AI applied to anything outside marketing. Lior Eldan says Moburst had gone all in on AI internally and brought that to clients, whose response was to ask what AI had to do with them. Closing that gap meant going on-site and walking through what each team actually does.

What did the team find inside the mortgage company? Processes that had existed for decades, staff checking forms by hand, account executives taking weekend calls with no context, and customers sending blurry phone photos of signed documents that took fifteen minutes to sort out. All of it inside a regulated environment where everything has to run on-prem.

Why were the first fixes not AI? Because the biggest problems were digital product problems. The first tool told customers a scan was blurry and how to retake it. Eldan describes the approach as coming from a digital perspective on how to do it right and layering AI on top. The AI is the second layer.

What is Lior Eldan's rule for automating a process? Make it work manually first, with the right team members and the outcome you want, and only then automate it. His warning is that people rush to automate processes that are not fully fleshed out, the automation fails, and they blame the AI.