I moved from Toronto to San Francisco with a relocation service myself. I knew exactly how many human touch points there were, and how many of them didn't need to be human at all. Jolie Ni, Founder of H
A business built on human labour, with all the fragility that brings.
Relocation is coordination-heavy by nature. Every move requires intake calls, vendor management, immigration logistics, short-term housing and shipping, all handled manually. That model works until it doesn't.
-
Too many touch points
Every move required a human coordinator at each step: intake, vendor coordination, immigration, housing, shipping. High-touch by design, expensive by default.
-
Volatile revenue, unstable headcount
Revenue fluctuated with economic conditions and the housing market, triggering endless cycles of hiring, training and right-sizing that drained resources.
-
Training costs that never ended
High turnover meant institutional knowledge walked out the door constantly. Training new coordinators consumed a significant share of revenue every single cycle.
-
Cases that fell through the cracks
With humans juggling dozens of concurrent moves across multiple vendors, things got missed. No system to catch what slipped. No guarantee of follow-through.
One dedicated agent per move case. Fully equipped. Always on.
We started by mapping the entire workflow and extracting every decision point. Then we built an operation agent that handles coordination end to end, with the right tools, the right permissions, and a human always in the loop for critical decisions.
-
Workflow mapping
Before writing a single line of code, we mapped every step and decision point in the relocation process to understand exactly where automation could replace manual work.
-
Internal knowledge access
Each agent is connected to the company's internal knowledge base: policies, vendor contacts, corner cases, and institutional know-how that would otherwise live only in people's heads.
-
Tool access
Agents can write emails, create support tickets, request signatures and access the company database, all with strict permission controls and human approval gates on write actions.
-
Dedicated agent per case
Each relocation gets its own agent instance with completely isolated memory. No data bleed between clients. No crossed wires between cases.
-
Human in the loop
Not a single critical decision goes out without human sign-off. A coordinator portal gives full visibility into every active case, pending approval and escalation in real time.
-
24/7 operations
The agent runs around the clock, responds in milliseconds, and never lets a case go cold, regardless of time zone, business hours or team capacity.
Designed for trust, not just efficiency.
Every design decision was made with one question in mind: would a human coordinator feel comfortable handing this case to the system?
Key design principles
- Each move case handled by a dedicated agent with isolated memory, so there is no context bleed between clients
- Least-privilege permissions: read-only by default, all write actions require human approval
- Coordinator portal with full visibility into every case, approval queue and escalation path
- Structured escalation: when the agent cannot handle something, it hands off with full context rather than a blank slate
- All outbound communications reviewed by a human before sending
Not everything deserves the same response.
The speed gain does not come from answering everything instantly. It comes from sorting by what a request actually needs, then spending human attention only where it changes the outcome.
Three tiers
- Basic inquiry. Answered by the agent in minutes, with no human in the path.
- Case-related. The agent drafts a response in minutes; a human reviewer takes around 30 seconds to check and send.
- Urgent or human-intensive. Escalated immediately, with full case context attached.
From days to minutes, and five times the caseload per operator.
The agent has been running for eight weeks. The headline change is not that work got faster, but that coordinator attention moved off routine coordination and onto the cases that genuinely need a person.
- Baseline
- The agency's own coordination workflow, measured before deployment
- Measurement window
- 8 weeks post-deployment
- Volume observed
- Average move spans roughly two months
| Metric | Before | After | Change |
|---|---|---|---|
| Response time | 3–4 days | Minutes | Tiered routing |
| Cases per operator | 10+ | 50+ | 5× |
| Coordinator hours freed | n/a | 6 h per case | 6 h |
| Labour cost per case | $600 | Not published | See below |
- Cost per case after deployment. The $600 baseline is real. We have not published an after-figure because we have not isolated the running cost of the agent itself, and a cost claim without that is not a measurement.
- Training cost savings. One of the original goals, but turnover across the window was too low to produce a meaningful figure either way.
- Customer satisfaction. Faster responses are not automatically better ones. No CSAT instrument was in place before deployment, so there is no honest before-and-after to show.