By Sagar Shankaran, Founder of CallSphere
How Claude Cowork and ChatGPT Work turn a Tuesday survey into a signed binding estimate packet by 5pm - and what an owner must change about assigning work.
Key takeaways
Four-ten on a Tuesday afternoon in the third week of June. Your estimator finished a video survey of a four-bedroom colonial in Ridgewood at 3:15 and dropped the cube sheet into SmartMoving: 1,240 cubic feet, call it 8,700 pounds at seven pounds a foot, Ridgewood to a subdivision outside Charlotte, 620 miles on the mileage guide. There is a piano, and two crawlspace shelves the homeowner mentioned in passing. She said twice that she is talking to two other companies this week.
Your move coordinator has that survey in her queue behind four others, so the Hendersons get a written estimate Thursday afternoon. Both competitors will have quoted by Wednesday night. You know how this ends, and it has nothing to do with your crews.
Every mover I have visited in the last two years has the same bottleneck in the same place, and it is not sales and it is not the trucks. It is the seventy minutes between "we surveyed the house" and "the customer has a signed, priced, compliant packet in her inbox." In January that gap does not matter. Between Memorial Day and Labor Day, when roughly seven of ten moves happen and end-of-month weekends run three crews deep, that gap is the business.
The workaround everybody pretends is fine: the coordinator texts a ballpark. "Looks like eleven-five to twelve-two, formal estimate Thursday." The customer either takes that as a promise, which becomes a claim later, or as a stall, and signs with whoever sent real paper first.
Look at what actually has to exist before that Ridgewood job is booked. The written binding or non-binding estimate priced off Tariff 400-N with your agency's discount applied to linehaul, plus packing, plus valuation. The Order for Service. The Bill of Lading terms. The valuation election — Full Value Protection as the default, or the signed statement dropping to Released Value at sixty cents per pound per article. The High-Value Inventory form. And the two FMCSA booklets you are legally obligated to hand a household goods shipper: "Your Rights and Responsibilities When You Move" and "Ready to Move?"
In a nine-truck shop one person builds all of that, and that person also answers dispatch when a crew leader calls in with a shuttle problem. She rekeys the cube sheet into the tariff calculator, checks the mileage guide, adds long carry and stair carry lines, prices three packing scenarios because the customer asked, drafts the cover email, attaches the booklets. Seventy minutes is a good day. Ninety is normal in July.
Here is the plain definition worth keeping: an agentic assistant is software you hand an outcome to — "produce the Henderson binding estimate packet off our 400-N tariff at our agency discount, with three packing options" — which then works through your survey file, your tariff and your move management system on its own for as long as it takes, and hands back the finished packet for a human to check and sign.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent for home services in your browser — 60 seconds, no signup.
You have seen chat boxes. You typed a question, it typed back a paragraph, and you still had to do the work. That was the 2024 and 2025 version, and it was worth almost nothing to a moving company, because your problem was never that you needed a paragraph.
Anthropic shipped Claude Cowork on 12 January 2026, and OpenAI shipped ChatGPT Work on 9 July 2026 running on GPT-5.6. Both do the same new thing: you give a goal, they connect to your apps and your files, break the job into steps, work on their own for hours if the job takes hours, and hand back finished work — a completed spreadsheet, a priced quote packet, a filled set of forms. Both were aimed at office staff, not programmers. Cowork reached phones and the web in July, which matters when your coordinator is standing in the warehouse.
flowchart TD
A["Estimator finishes video survey, 3:15pm"] --> B["Coordinator states the goal, not the steps"]
B --> C["Assistant reads cube sheet + mileage + 400-N tariff"]
C --> D{"Binding or non-binding on this lane?"}
D -->|Binding| E["Prices packing, valuation, long carry, shuttle"]
D -->|Non-binding| E
E --> F["Builds estimate, Order for Service, valuation form, booklets"]
F --> G["Coordinator checks the number and signs"]
G --> H["Packet in customer inbox by 5:05pm"]
3:20 p.m. The coordinator opens the assistant and types one instruction: build the Henderson interstate packet from today's survey, price it binding off 400-N at our agency discount, three packing options — full pack, partial with kitchen and glassware, owner-packed cartons only — show Full Value Protection and the sixty-cent alternative side by side, flag anything in the survey notes that could turn into a shuttle or a long carry.
3:52 p.m. It comes back priced, with the mileage matched. It caught the crawlspace shelves in the surveyor's notes: "survey notes mention two crawlspace shelves not itemized on the cube sheet — confirm before binding." It flagged the destination subdivision as a possible 53-foot access problem and priced a shuttle as an optional accessorial rather than burying it. It drafted the cover email, attached both FMCSA booklets, and left the signature line blank.
4:05 p.m. The coordinator calls the customer about the crawlspace, adds forty cubic feet, has it re-price, reads the total out loud once to herself, and sends it. The Hendersons have your packet before either competitor has called them back. That is the entire change. Not smarter pricing — earlier paper.
Assume 620 interstate jobs a year, one coordinator building packets, 70 minutes a packet, and 60% of surveys falling between 15 May and 15 September. Assume your close rate is 38% on packets sent the same day and 26% on packets sent 48 hours later — use your own numbers, every shop has them in the CRM. Average interstate revenue per booked job, say $5,900.
| Assumption | Today | Goal-assigned |
|---|---|---|
| Surveys converted to packets, peak season | 372 | 372 |
| Coordinator minutes per packet | 70 | 18 (check and sign) |
| Peak-season coordinator hours on packets | 434 | 112 |
| Share of packets out same day | 45% | 90% |
| Booked jobs from those packets | 115 | 132 |
| Revenue difference at $5,900 a job | — | +$100,300 |
Halve that number if you are feeling pessimistic and it still pays for itself many times over. And notice the 322 coordinator hours that came back — in peak season those are not idle time, they are the follow-up calls on last week's quotes that nobody was making.
This is the part owners skip. If you hand the assistant a task list, you get a slightly faster task list. The gain comes from handing over an outcome and moving your people to the review gate.
Still reading? Stop comparing — try CallSphere live.
See the home services AI agent handle a real call — complete, industry-specific, and live in your browser. No signup.
Three changes. First, the standing instruction has to contain your actual rules — your discount off 400-N, when you go binding versus non-binding, which ZIP codes always need a shuttle, your minimum weight. Write it once, the way you would write it for a new hire. Second, the coordinator's job description changes from "builds estimates" to "approves estimates and calls the customer about the exceptions." Frame that as the promotion it is, or she will assume you are replacing her. Third, nothing goes out unsigned. The packet is a legal document under 49 CFR Part 375 and the person who signs it owns the number.
The survey is still the survey. If your estimator misses the attic and the garage loft, the assistant will price a beautifully formatted wrong number, and on a binding estimate you eat the difference. Garbage in, faster garbage out.
Judgment stays yours. Whether to bind a job for a customer who has already changed the date twice. Whether the piano is an upright or a baby grand, which is a phone call, not a document. Whether to take a job into Manhattan at all in the last week of the month. And the 110% rule on non-binding estimates — deciding what you can actually collect at delivery — is a business call with a claims history behind it.
It will also get a tariff item wrong now and then, usually an unusual accessorial like an extra pickup or a bulky article charge. That is why the check-and-sign step exists, and why eighteen minutes rather than zero is the honest number in the table above.
No, and if you try it that way you will be sorry in September when claims and delivery spreads pile up. It replaces the seventy minutes of document assembly. The coordinator still calls the shipper about the crawlspace, still fights the destination agent over a delivery window, and now has time to chase last Thursday's quotes.
You give it your tariff, your discount and your accessorial sheet, and you instruct it to price only from those and to say so when something is not covered. Then you spot-check twenty packets against hand-built ones before you trust it unattended. Any assistant that cannot show you which line of which document a charge came from is not one you should be signing behind.
Generally no. These assistants work across the files and applications you already have — the survey video and cube sheet, the tariff workbook, the estimate template, your email. Whether it can write back into your move management system depends on that vendor. The honest starting point is read-and-produce: build the packet outside the system, sign it, and key the booked job the way you do now.
One lane. Pick your most common interstate corridor, take five surveys you already quoted last month, and have it rebuild those five packets. Compare them line by line to what you actually sent. You will learn more in ninety minutes about where it helps and where it drifts than from any demo.
One last note from us. The other half of that June afternoon is the phone: the shipper calling at 7:40 p.m. to ask whether the estimate covers the piano, the referral who calls while all three coordinators are on other lines. CallSphere builds AI voice and chat agents that answer the line and the website chat around the clock, book the survey, and capture the lead with the details attached. It does not price your tariff. It keeps the call that starts a job out of voicemail during the four months when every call counts.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
The monthly IEEE 1366 reliability close takes 64 hours across three people. What goal-driven agents change, the arithmetic, and what stays with the engineer.
Engine fault reports plus driver inspection photos now catch liftgate and emissions failures nine days early. The arithmetic on one stranded loaded van.
How pest control service managers hand the monthly food-account trend packet to a 2026 work agent as a goal - and what has to change about assigning work.
The phased plan, insurance estimate, predetermination narrative and financing page, finished before the patient leaves. What the owner has to change to get it.
Why co-pack quotes take six days, and how 2026 agents that return finished work rebuild the packet — costed formula, freight, spec sheet — in two hours.
A 1/1 commercial submission packet costs an account manager nine hours, eight of them gathering. In 2026 you hand over the goal and review the finished packet.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI