Your board wants to know what you’re doing about AI. Employees want better tools and some reassurance about their jobs. Vendors promise to transform the business. Somewhere in all that noise, you’re supposed to make an investment decision and explain when it will pay off.
Consider how a rollout can unfold. The company buys licenses, runs training and invites a few enthusiastic employees to demonstrate what they’ve built. Several weeks later, the adoption dashboard is green. It’s too early to demand the full financial return. But ask how those activities will change cost per loan, turnaround time or conversion, and the answer may still be vague. The rollout has a plan. The return doesn’t.
An October 24 chart from Apollo’s Torsten Slok raises that question on a larger scale. Profit margins and revenue per employee have risen for the Magnificent Seven: Apple, Microsoft, Alphabet (Google), Amazon, Nvidia, Meta (Facebook and Instagram) and Tesla. Across the other 493 companies in the S&P 500, the picture is much flatter. That doesn’t isolate AI’s contribution or prove that companies outside those seven aren’t benefiting, but what it does is give us a reason to examine the gap between adopting a technology and earning a return.
Putting intelligence to work
Companies have always been limited by how much thinking their people could do: if a problem needed more brainpower, you hired, trained or waited. That’s changing. Intelligence is now an economic input you can buy and point at a problem. But we don’t know how to allocate it well yet.
A company-wide AI license or platform subscription buys access. It doesn’t tell you which customer problem to solve, how to connect your systems or how to make the result useful. Someone still has to figure that out. Usage is easy to count, and I understand why leadership teams end up reporting “all green” on the AI adoption slide. But that doesn’t tell us whether we’ve put the intelligence to work on a problem worth solving.
When we skip those decisions, we end up looking at AI first and the business second. A tool can summarize documents, so we find documents to summarize. It can draft emails, so we teach everyone to draft better emails. We haven’t yet asked why the work takes that form, or whether the same investment could let us offer customers something we can’t economically deliver today.
Imagine a processor using AI to write follow-up messages faster. The messages exist because nobody has a reliable view of the file’s status. The underwriter still waits for a complete package, and the loan officer still calls for updates. The processor produces more polished reminders, and everyone still has to read and act on them. The tool works, but nobody has removed the reason for sending the messages.
This is how AI can amplify a broken process. Some of what looks like institutional knowledge is the accumulated scar tissue of systems that never worked properly. I want to preserve people’s hard-earned judgment. I’m less enthusiastic about paying to make every workaround permanent. AI gives us a chance to reconsider how the business operates. If we ignore this opportunity, we risk spending the budget making our existing complications run faster.
Mapping the process is a useful start. You can see delays, remove duplicate steps and make the handoffs cleaner. But a cleaner map doesn’t necessarily change what people are expected to do. Suppose information now moves automatically from intake to underwriting. The processor may still check every transfer, and the underwriter may repeat the check because they’re accountable if it’s wrong. If nobody changes those expectations, automation becomes another thing to supervise alongside the original job. The old checks still consume time, and now the company is paying for the tool too.
Redesigning the work and the role
That’s an adoption problem a training session won’t fix. Someone who remains accountable for a mistake needs to know what they can trust, what they must verify and what they’re authorized to stop doing. A good process redesign should address those questions. In practice, it’s easy to stop at the flow of the file without working through what changes for each person handling it.
When Sarah Gonzalez joined me on my podcast, The Signal, she described starting with the customer or employee experience she wanted to create and working backwards from there. Sarah is Chief Innovation Officer at Logan Finance, a lender focused on non-QM loans. Her nearly 30 years in mortgage include COO and president roles; her responsibilities at Logan include helping people adapt to new ways of working as the technology changes.
“I don’t start with the org chart,” she said. She works backward from the experience to what can be automated, what requires judgment and what employees need to learn. Logan’s AI-powered eligibility tool illustrates that distinction. It checks loans against rules, reducing the need for underwriters to search through matrix after matrix. Sarah reported a significant lift in underwriter production. The tool handles rules-based eligibility; the underwriter owns the credit decision.
Sarah made the leadership decision explicit: “When AI removes 20% or 30% of the tasks from someone’s day, leadership has another decision to make: what are we going to do with that capacity? Does that person handle more volume? Spend more time with clients? Focus on exceptions and quality? Move into a different role? Or allow us to grow without adding an incremental headcount? If we don’t answer that question, I don’t think we have actually redesigned the business—we have just automated pieces of the old one.”
This is where role redesign can turn a broad ambition into an investment decision. Start with the business outcome, then ask where people’s attention could improve it. For an underwriter, that could mean more time for complex credit decisions that determine loan quality. For a processor, it could mean resolving problems that threaten a closing. A loan officer might need more time for useful conversations with borrowers. Then take the job apart with the people doing it, including the work that never appears in the job description.
Group the tasks: gathering information, checking requirements, coordinating with others, making decisions and advising customers. Record how often each task happens, how much time it takes and why it sometimes needs to be done again. The box labeled “underwriting” becomes a concrete list of tasks you can examine: time spent finding the latest document, repeated guideline lookups, a condition that comes back twice because the request wasn’t clear. Beside those tasks, identify the decisions that need an experienced underwriter’s judgment. You can see both where effort is going and where you’d rather have it go.
For each task, decide what stays human, what AI could help prepare, what could run automatically within defined limits and what could disappear if an underlying problem were fixed. Include the new work of checking outputs and handling exceptions. A tedious check may still protect loan quality; reducing its effort has to preserve its purpose. Then make the management decisions: who owns the result, what can they stop doing, and how will their performance be assessed? Until those decisions are made, the task map is an inventory of possibilities rather than a redesigned role.
An AI roadmap starts to emerge from those decisions. You know where the intelligence you’re buying could make a difference. Preparing a complete, organized file might be a better first investment than trying to automate underwriting. Reconciling status across systems might remove more effort than another document summary. Some fixes won’t need AI at all. You can assess a vendor using actual cases from your business, including the investor-specific requirement and the exception resolved by a phone call. A promise to automate fulfillment now has to answer a specific question: how much work will our people still have to do?
Don’t stop at cheaper and faster
The point of this exercise is bigger than finding tasks to automate. An efficient mortgage assembly line is valuable, but it can also set a ceiling on our ambition. If we only ask how to move today’s file faster and at a lower cost, we may never ask what else the people and expertise inside the company could make possible.
Sarah offered a glimpse of a wider ambition that goes beyond optimizing the loan manufacturing process. In our August conversation, she described plans for a client-facing version of Logan’s eligibility tool later in 2026 or early 2027, allowing clients to check eligibility and request an exception before submitting a loan. That would move useful information earlier in the relationship. A tool built to help underwriters could also give clients answers earlier.
Consider a possibility beyond Logan’s example: personalized mortgage-readiness support for people months away from buying. With the customer’s permission, a lender could help them maintain a preparation plan as their timing or circumstances change, identify missing information and bring in a qualified person when a question needs judgment. That kind of attention isn’t new. The opportunity would be to provide that sustained, individual attention to more people at a cost the business can support, including people who never close. The test is whether the additional profitable business justifies the full cost of that service.
A loan officer freed from administrative work could spend more time on those relationships, but only if the role, workload and incentives make room for it. Measuring that person solely on this month’s funded loans would work against the service you’re trying to create. A new customer promise needs someone with the time, skills and responsibility to deliver it.
The same discipline applies to every proposed saving. Ten minutes removed from a task doesn’t automatically remove ten minutes of payroll expense, and capacity to handle more loans doesn’t create demand for them. Decide how the time will generate additional profit or an actual expense reduction, then deduct the costs of implementation, integration, training and oversight. Include the period when you’re paying to run old and new arrangements together. Loan quality and customer experience need to hold up as the economics improve.
You don’t have to rebuild the enterprise all at once. Choose one role, work with the people doing it and agree on the business outcome you want to improve. Follow the work across departmental boundaries rather than making one person’s day easier at someone else’s expense. Before the next purchase, you should be able to say what people will stop doing, what they’ll do with that time and what the business will be able to offer customers as a result. Then show how you’ll know it paid off.
That’s the answer I’d want to take back to the board.
Reuven Gorsht, Co-Founder and CEO, Deeded