Case Study · Mortgage Lending
Every Lead Owned. Every Callback Kept.
How a regulated mortgage business made every lead visible, owned, and followed up, while keeping every lending decision with people.
This engagement was in mortgage lending. The method works on any workflow that has outgrown the way it’s run.
At a Glance
At a Glance
Industry
Mortgage lending, a regulated customer and communications environment
Business setting
Small office with multiple loan officers, reception support, and broker oversight
Lead sources
Inbound phone calls, website inquiries, and purchased leads
The problem
Follow-up depended on forwarded messages, inboxes, spreadsheets, personal notes, and individual memory
Who led it
Wen Giwa-Osagie led the work end to end, using the five-phase approach now formalized as the Giosena Method™
What was built
Eight connected workflows on GoHighLevel and Twilio, including a custom AI integration powered by Claude for internal next-step recommendations
Status
In daily production use since April 2026
Result
A 27% year-over-year increase in closed loans on the same marketing budget, Q3 2026 compared with Q3 2025
Core principle
AI handles internal coordination. Qualified people keep every customer conversation, mortgage judgment, and lending decision.
The Results
The Results
+27%
More closed loans, same marketing budget
Q3 2026 compared with Q3 2025
Measured over the first full quarter after launch, against the same quarter a year earlier. The marketing budget did not change.
Figure reported by the client. Loan volumes are kept confidential.
Every missed call now has an owner and a next action.
Six-month callbacks come back on time, not from memory.
Phone, website, and purchased leads all flow into one process.
In daily production use since April 2026.
Is a key workflow in your business running on memory and workarounds?
Start with one workflow.
The Challenge
The Challenge
Inquiries reached this mortgage business three ways: phone calls, website forms, and purchased-lead files. The team cared about every one of them. The problem was the process underneath.
A missed call became a receptionist’s message. A website form became an email in the broker’s inbox. A purchased-lead list became a spreadsheet that each loan officer worked their own way. And a customer who said “call me back in six months” depended entirely on one person remembering.
For a regulated mortgage business, speed alone wasn’t the answer. Any new system also had to protect human responsibility for customer conversations, mortgage guidance, mortgage eligibility review, lending decisions, consent, and communication preferences.
Hard to see and easy to lose
- Missed calls could sit without a clear owner.
- Future callbacks could disappear when a loan officer got busy or left the company.
- Purchased leads were handled inconsistently.
- Customer replies and communication preferences could be missed.
- Closed customers fell out of view once the loan closed.
The Design Question
How can every customer inquiry become a visible, owned, and reliably followed-up task, without letting AI or automation make mortgage decisions, give advice, or replace the human relationship?
How We Worked
The Giosena Method™
Wen Giwa-Osagie led the engagement from the first conversation through live operation, using the five-phase approach now formalized as the Giosena Method™.
Discover
Learn how the business actually operates
Diagnose
Determine why performance is constrained
Design
Redesign the operation, not just the software
Engineer
Build, integrate, test, and launch
Optimize
Stabilize and refine in live operation
What happened in this engagement
01 · Discover
We learned how leads, calls, messages, and follow-up actually moved through the office: who received them, where they went, and where they stalled.
02 · Diagnose
We found the root cause. The team wasn’t the problem; follow-up depended on people’s memory instead of a shared system, so opportunities leaked at every handoff.
03 · Design
We defined eight connected workflows and drew a clear line between what AI recommends, what automation handles, and what people decide.
04 · Engineer
We built, integrated, and tested the system across real paths and failure cases, launched it in April 2026, and documented the complete architecture.
05 · Optimize
We stayed hands-on after go-live, testing and refining the system in live operation until it held up reliably in the team’s daily work.
The architecture is fully documented, so the system doesn’t depend on any one person to understand it, and it can be maintained or extended without guessing.
The Operating Model
The Operating Model: Before and After
Before
Follow-Up Depended on People
Every path could end the same way. Whether a customer got a callback depended on who saw the message, who remembered the promise, and who had time that day.
After
One System, Every Lead Owned
Every inquiry, from any source, now becomes one CRM record with an owner, a priority, and a next action, and it stays visible from first contact through closing and beyond.
What Was Built
What Was Built
The system brings together five core capabilities, all live and in daily use:
01
A central system of record for every lead and customer, with visible ownership and full activity history.
02
Integrated phone and SMS communications, including an immediate text acknowledgment when a caller can’t reach someone.
03
A custom AI integration that recommends internal next steps, with a loan officer confirming before customer contact.
04
Communication controls that protect consent, customer preferences, and opt-outs.
05
Daily work queues and escalation so every loan officer knows what needs attention, and nothing overdue goes unnoticed.
The complete architecture is also documented, so the system can be operated, governed, and extended over time.
System Map
The Eight Connected Workflows
Every lead source flows into one record
The record drives every connected workflow
| # | Workflow | What It Does |
|---|---|---|
| 1 | Phone and missed-call handling | Turns every call into owned, visible human follow-up |
| 2 | Website inquiry intake | Creates or updates the record, assigns an owner, and starts follow-up |
| 3 | Purchased-lead intake | Assigns each lead and requires first human contact before automation |
| 4 | Assignment, routing, queues, and escalation | Gives every lead an accountable owner, a priority, and overdue-work visibility |
| 5 | AI-assisted next-step recommendation | Recommends a safe internal next action while people stay in control |
| 6 | Nurture and future follow-up | Preserves future opportunities and brings them back at the right time |
| 7 | Replies, preferences, opt-outs, and suppression | Routes replies to the right person and respects customer opt-outs |
| 8 | Post-close customer lifecycle | Keeps customers visible after closing and creates human-reviewed milestones |
All eight workflows are in production.
Responsibility Model
How AI Fits Safely
The AI runs in production as an internal workflow recommendation layer. It is not a customer-facing chatbot, and it is not a lending tool. The production workflow uses Claude, Anthropic’s AI model, to interpret approved CRM context and recommend an internal next step, while qualified staff retain responsibility for customer conversations and lending decisions.
Coordinates
CRM and Automation
- Creates and updates records
- Assigns owners
- Creates tasks and reminders
- Escalates overdue work
Recommends
AI-Assisted Logic
- Reads the workflow context
- Recommends an internal next step
Decides and acts
Qualified Human
- Conducts customer conversations
- Confirms customer-specific outreach
- Provides mortgage guidance
- Reviews mortgage eligibility
- Makes lending decisions
Control boundary: The AI can recommend an internal next step. It cannot give individualized mortgage advice, determine mortgage eligibility, make a lending decision, send a customer message, override an opt-out, or enroll a lead in a campaign on its own.
In Practice
A Real Example
1 · Customer replies
2 · AI recommends
3 · Loan officer acts
When a customer replied, “I’m at work; call after 5:30,” the AI read the context and recommended a scheduled human follow-up, with a reminder set for the requested time.
It didn’t send a message, and it didn’t replace the loan officer. The loan officer made the call and documented the outcome.
That’s the role AI was built to play: remove the coordination burden around human judgment, without touching the judgment itself.
Communication Controls
Communication Safety by Design
In a regulated business, a message that fires at the wrong time, to the wrong person, is a real risk. So every automated email and SMS passes through a control gate before it can be sent.
Checkpoint
Communication control gate
If a message isn’t permitted, it doesn’t go out, and the human follow-up task stays visible so the customer still gets a response.
Testing
How It Was Tested
A workflow that runs once in a demo isn’t the same as one a business can trust every day. Before launch, we tested the real customer paths, the handoffs between people and systems, and the ways things can go wrong: missed calls, website inquiries, purchased leads and first-contact rules, nurture and future follow-up, opt-outs and unsubscribes, AI recommendations, AI failures and invalid responses, conflicting customer information, and duplicate events.
Outcomes
What Changed
Since going live in April 2026, the system has changed how the business runs day to day:
Missed calls became owned work. Every one now has an owner and a next action, instead of sitting in voicemail or an informal message.
Future follow-up stopped depending on memory. A requested six-month callback is scheduled, assigned, and brought back on time.
Every lead source flows into one process. Website inquiries and purchased leads no longer live in inboxes and spreadsheets.
Loan officers start each day with a clear queue of tasks, priorities, customer history, and due dates.
Prospects who weren’t ready stayed in view through structured follow-up and nurture.
Relationships continue after closing, instead of ending operationally when the loan closes.
Old opportunities came back. Prospects reengaged later, and some referred others.
Note on outcomes
The closed-loan figure was reported by the client. The marketing budget was the same in both quarters. Loan volume can still be affected by factors outside any workflow, such as interest rates and market demand, so the increase shows the business’s results with the system in place, not an audited attribution to the system alone. Operational observation and CRM activity also showed an improved ability to retain, reconnect with, prioritize, and work leads that might previously have been lost.
Why It Matters
Why It Matters
The result wasn’t “AI replacing staff.” It was a more reliable operating system: one that makes work visible, preserves customer history, protects communication controls, keeps people accountable, and keeps future opportunities with the business instead of in one person’s memory.
Core principle
AI creates the most value when it removes the coordination burden around human judgment, not when it replaces the human relationship.
For regulated and high-trust businesses, the approach is the lesson: understand the real process first, decide where people must stay accountable, use automation for coordination, and use AI only inside clear boundaries.
Who Designed This
Wen Giwa-Osagie
Founder, Giosena · Product Leader & AI Systems Architect
I bring two disciplines to every engagement. As a product manager, I find the real constraint and decide what’s worth fixing first. As an AI systems architect, I design how the solution works and see it through to a working system: where AI belongs, what stays with people, and how it gets tested. When one person owns both, the business problem and the technical design stay connected from the first conversation to launch.
I’ve spent more than 17 years doing this work across fintech, mortgage technology, e-commerce, and SaaS. Highlights include:
Beyond mortgage
Is This Your Business?
You may be a fit if an important workflow in your business is:
Time and cost
- Taking too long
- Costing too much to run
- Requiring too much manual effort
Quality and flow
- Creating errors, rework, or inconsistent results
- Delayed by handoffs, approvals, or waiting
- Limited by a bottleneck
Control and growth
- Dependent on one person’s knowledge or memory
- Difficult to manage, measure, or scale
- Overdue for change, and you’re not sure whether AI is part of the answer
The workflow could involve:
The problem is not always a bottleneck.
It may be repetitive manual work, disconnected systems, poor handoffs, unnecessary approvals, unclear ownership, avoidable errors, or cost that has become normal.
We don’t start with AI tools. We start with the work that is creating friction in the business. Then we design the right solution.
Start with one workflow
Find what is making the work harder than it needs to be.
In a 15-minute scoping call, we look at one workflow that is taking too long, costing too much, creating errors, requiring too much manual effort, or limiting growth.
We identify the issue and determine the right response: process redesign, automation, AI assistance, clearer human ownership, or a combination.
15 minutes. No obligation. Start with one workflow and expand as the results come in.
Engineer a smarter business.