Man and Van Business: Quote From Photos and Plan the Day With AI

Ask every customer for three photos (the items, the way out, the parking spot) and five access answers. Have ChatGPT or Claude list the items and estimate the load as a fraction of your van, price it from a rate card, then give the confirmed jobs' time windows and Maps drive times to AI to check the running order.

Treat the estimate as a band with a confidence note, never a measurement. OpenAI's and Anthropic's own documentation both say counts and spatial judgements from images are approximate, so every quote carries a line saying the price is subject to confirmation on arrival. AI does the listing, the arithmetic and the rule-checking; your rate card and Maps supply the prices and the drive times.

Follow me on Instagram@sagnikteaches

Most man and van enquiries arrive as a text saying "how much for a sofa, a wardrobe and some boxes?", and each takes ten minutes of back-and-forth before there is a price. The routine below aims for a first reply inside a few minutes and a day plan that survives the job that overruns. All prices and minutes are illustrative, so replace them with your own after the first 20 jobs.

Connect on LinkedInSagnik Bhattacharya

The intake message: three photos, five access questions, one job type

Quoting from photos starts before the AI sees anything, because the estimate is only as good as what customers send. Save the request below as a reply template so every enquiry gets the same message, and ask once, politely. Do not quote from a single photo.

Subscribe on YouTube@sagnikteaches

Thanks for getting in touch. To give you a price without a visit, please send:
1. A photo of the whole room or space from the doorway, with everything to be moved in view.
2. A photo of the way out: the stairs, hallway or lift, and the front door.
3. A photo of where a van could stop, as close to the door as possible.

And please answer five questions:
a) Which floor is it, and is there a working lift?
b) How many flights of stairs are there, and are any narrow or twisting?
c) How far is the nearest place I can legally park from the door, and does it need a permit or a booking?
d) Is anything very heavy, very large or needing taking apart (wardrobe, piano, safe, trampoline)?
e) Which days and times suit you, and is there a hard deadline (key handover, lift booking)?

Also tell me what is staying, what is going to a new address and what is going to the waste site. I will reply with a price that is confirmed when I arrive. I delete photos once the job is done.

Each request earns its place. Photo one gives the AI the items. Photo two shows the carry, which is where most surprises hide: a spiral stair or a low door can double the time on site without adding a single box. Photo three tells you whether you can park on the doorstep or will be shuttling items along a street. Asking for something of known size in shot, such as a standard door or a one-litre bottle, helps the AI judge scale, and asking customers to keep faces out of the frame keeps your phone cleaner.

The five questions exist because photos cannot answer them. Floors, lifts and parking limits decide the extras on your rate card, and a hard deadline decides which time windows you can offer. For a whole-house move the better tool is a video walk-through, covered in our tutorial on turning customer videos into removal quotes. Three photos suit the smaller jobs a van does all day: a sofa and a bed, a flat clearance, a bike and a few boxes.

Turning photos into a load estimate in van fractions

Describe the load as a share of your own van, where a full load means the van filled to the roof behind the bulkhead. Customers understand "half a van", while cubic measures mean little to them, and an AI's volume figure from a photo is far shakier than its item list. So give it a calibration sheet. Load your van once with a known set of furniture, photograph it, and write down the share each piece took. Fifteen minutes with a sofa, a wardrobe, a mattress and ten boxes gives you a table the AI can work from.

The prompt below carries the sheet inside it. The shares are illustrative for a long-wheelbase panel van; yours will differ.

You are estimating a van load for a man and van business. Use only the photos attached and the calibration sheet below. One full load = the van filled to the roof behind the bulkhead.

Calibration sheet (share of one full load): sofa 3-seat 0.12; sofa 2-seat 0.08; armchair 0.04; double mattress 0.05; double wardrobe taken apart 0.08; chest of drawers 0.04; dining table with 4 chairs 0.10; washing machine 0.03; standard box 0.01; bicycle 0.03.

Return a table with: item, count, share each, confidence (high, medium or low) and a note. Then give the total share, a range (low to high) and the band (quarter, half, three-quarter or full) that contains the middle of the range.

Rules: list only what you can see. If items are stacked or hidden, say so and add a line called "unseen items" with a low-confidence allowance. Do not guess weights or dimensions. Finish with the questions I should ask the customer before I quote.

Here is the kind of answer to expect for a one-bedroom flat clearance (an illustrative output, not a recorded one).

Its questions would be: does the wardrobe come apart, are there more boxes behind the door, and is the sofa going to the waste site or a new address? Those become your follow-up message, and the answers move the estimate more than any second photo.

What I would fix in that output

• Hidden stacks. A pile photographed from the front shows one row. Keep the "unseen items" rule, and for stacked loads ask for a second angle or a 20-second video.
• Merged items. "Table and chairs" counted as one thing hides four chairs. Add a rule to list every piece separately.
• Invented precision. Left alone, models volunteer weights and dimensions from a photo. Weight matters for lifting, but a guess is worse than a question, so the prompt forbids it.
• Mirrors and reflections. A mirrored wardrobe door can be counted as a second wardrobe. Check the item list against the photo before you price.

Media limits are rarely a problem for photo quotes, but know them. Anthropic's documentation gives 20 images per message on claude.ai at 10 MB each, and OpenAI's image FAQ accepts PNG, JPEG and non-animated GIF at 20 MB each, resizing images before analysis. Anthropic also warns of mistakes with low-quality, rotated or very small images, which is why the intake message asks for whole-room shots rather than thumbnails. For a deeper look at what these tools read well and badly, see our tutorial on what ChatGPT can read in PDFs, spreadsheets and photos.

A rate card the AI cannot misread

Price from a card, not from the AI's sense of what a job is worth. Bands remove false precision: the estimate says "half a load", and the card says what half a load costs. Everything below is illustrative USD for two people and a long-wheelbase van, with the van within about 20 metres (65 feet) of the door and no stairs.

One selection rule keeps the AI honest: price the band that holds the middle of its range, and when the range crosses a band edge, price the higher band and tell the customer the price drops a band if everything fits. The same card also works for the reply you send; our tutorial on courier quote replies from a rate card shows how to let an assistant draft the message without letting it invent prices.

One flat clearance quoted from four photos, start to finish

Take the flat in the table above. The customer sent three photos plus a fourth showing a second flight of stairs, and answered the five questions: second floor, no lift, two flights, van within 15 metres, wardrobe still assembled, and everything going to the waste site because the flat is being emptied.

1. Band. Total 0.47, range 0.38 to 0.55, so half a load at $120.
2. Stairs. Two flights at $15 is $30.
3. Taking apart. The wardrobe is assembled, so add $15.
4. Waste. The site's fee for this load is $27 on its price list (illustrative), plus your $15 handling: $42.
5. Total. $120 + $30 + $15 + $42 = $207, subject to confirmation on arrival.

Now the check most owners skip. Two people for about two and a quarter hours at $20 an hour each is $90 of labour. Van and fuel for the job come to roughly $30, and the $27 of site fees is passed through. That is $147 of cost against a $207 price: $60 of profit, or 29 per cent. If your target is 35 per cent, the price needed is $147 divided by 0.65, which is $226. The fix belongs in the rate card (say, half load at $140, which brings this job to $227), not in a discount or a surprise on the doorstep. The margin arithmetic here is the same used in our tutorial on markup versus margin.

The message that goes back to the customer states what you assumed, so they can correct it:

Hi [first name], thanks for the photos. I have priced: a 3-seat sofa, a double wardrobe (I will take it apart, $15), a double mattress, about 10 boxes, a chest of drawers and a bike. Half a van load, second floor with two flights and no lift, van parking within about 15 metres. Everything goes to the waste site, with the site's fees at cost.
Total: $207. This is confirmed when I arrive: if the load is clearly smaller or larger I will tell you before I start and we agree the new price first.
Does Thursday between 10 and 12 suit? Reply YES and I will hold it.

How long a job takes decides which windows you can offer

A price is only half the quote; the other half is how long you will occupy the day. Keep a timing table next to the rate card. These minutes are illustrative for two people, include the walk-through and wrap-up, and exclude taking things apart and the waste run.

Add 10 minutes for each large item taken apart and the real length of your waste run: in the example, 25 minutes of driving plus 15 at the site, so 40. The flat above is 65 + 10 + 40 = 115 minutes before you drive to the next stop. Offer windows two to four hours wide so the plan has slack, and only sell a one-hour slot for a small job and at a premium.

Route order: let Maps measure, let AI check the windows

Split the work by what each tool is good at. Maps knows the roads. AI knows your rules. A language model has no live traffic and is unreliable at geography, so never ask it to guess drive times or to "optimise" distance. Paste in the drive times from Maps for the legs that matter, give it the windows and durations, and ask it to build a timeline and flag every rule that breaks.

Know what Maps will and will not do. Google's help page for computers says you can add up to 9 stops including the final destination, describes dragging stops into order yourself rather than automatic reordering, and says the leave-at or arrive-by setting only works for a route with one destination. So Maps can time the legs but cannot hold windows. The phone app may behave differently, so test yours.

Here is an illustrative day: five jobs, one van, two people. The waste site takes a last entry at 16:30, and each unload there takes 15 minutes.

Build a timeline for one van and two people, starting at base. Use only the drive times, windows and durations below.
Rules: arrive inside each window; never start before it opens (show the wait); a job with waste on board must be followed by a waste site run before any delivery; the waste site's last entry is 16:30 and each unload takes 15 minutes; do not order stops by distance alone.
Show a table: stop, arrive, start, finish, leave, wait, rule check. Show every addition so I can check it. If no order works, say which window to change and why.

Jobs and windows: J1 30 min 08:00-10:00; J2 75 min 09:00-13:00 (waste on board after); J3 35 min 12:00-14:00; J4 30 min 13:00-17:00; J5 45 min 14:00-18:00 (waste on board after).
Drive times in minutes: base-J1 12; base-J2 18; base-J3 25; base-J4 20; base-J5 15; J1-J2 15; J1-J3 30; J1-J4 28; J1-J5 22; J2-J3 18; J2-J4 14; J2-J5 20; J3-J4 22; J3-J5 35; J4-J5 25; J2-waste 25; J5-waste 12; waste-J3 25; waste-J4 22; waste-base 18.

The first draft of that kind of answer often goes wrong in a recognisable way (illustrative). Suppose it orders the stops by closeness to base, J1, J5, J2, J4, J3, and puts the waste run last. Worked through with your numbers, that plan reaches J5 at 08:52 for a window that opens at 14:00, J2 at 15:05 against a window that closed at 13:00, J3 at 17:26 against 14:00, and the waste site at 18:26, almost two hours after its last entry. A fluent summary can hide all four problems, which is why the prompt demands a rule check column and every addition.

The corrected plan, with the van leaving base at 08:03:

Check two lines by hand before you trust the rest: 10:15 plus 25 minutes is 10:40, and 10:55 plus 25 is 11:20. If the AI's arithmetic slips, ask it to show each addition, or paste the table into a spreadsheet. Run the same prompt again at lunchtime with the real finish times, because the most useful moment for it is the morning that went wrong. For bigger rounds, our tutorial on route planning for small fleets covers when optimisation software takes over from this method.

What to pay for, and when a route tool earns its fee

Start with what you already have. A seven-stop day, five jobs plus two waste runs, sits inside Google Maps' nine stops on a computer, so Maps plus a chat assistant covers it for one van.

Five jobs a day on 22 working days is 110 stops before waste runs, just over Routific's free 100 orders, so the question is whether a tool saves more than it costs. At an assumed $60 an hour for van and crew, $150 a month is two and a half hours, which is about seven minutes per working day. Routific's free plan keeps every feature except SMS notifications, so you can test it on your next 100 jobs before paying: if it saves more than seven minutes a day, it earns its fee. If you run two vans, the maths changes quickly in its favour.

Messages that go out the evening before

Most of the day's lost time comes from customers who are not ready or not there. Four short messages prevent most of it. Ask the assistant to draft them once, in your voice, then save them as templates with the placeholders intact.

Evening before: Hi [first name], [your name] from [business] confirming tomorrow's job. Arrival window [10:00 to 12:00]. I will text when I am about 20 minutes away. Please keep the parking space clear and have everything by the door. Reply if anything has changed.

On the way: On my way, about 20 minutes from you.

Running late: Sorry, an earlier job is running about 30 minutes over. I now expect to arrive around [time], which is still inside your window. Reply if that does not work and I will re-plan.

Different on arrival: Having seen it in person, this is a three-quarter load rather than half, so the price is $165 plus the extras we agreed. Happy to go ahead? If not, I will leave without charging for the visit.

The last template deserves a decision before you need it: do you charge for a wasted visit, and do you honour the original price when the customer's photos hid something? Write the answer into your terms once. Get customers' agreement to receive texts when they book, and check the messaging rules where you trade; a message about a booked job is not the same thing as marketing.

Checking the estimates against real jobs for a month

You do not know how good the photo method is until you log it. A sheet with one row per job is enough, and an assistant can tidy and total it at the end of each month.

The garage row is the instructive one: boxes stacked two deep, photographed from the front. Suppose that after 30 jobs 21 landed in the AI's band, 6 came in a band higher and 3 a band lower. The bands on the card are $45 apart, so under-called jobs repriced on arrival recover $270 but at the cost of six awkward doorstep conversations, while the three over-called jobs give back $135. A cheap fix is a standing rule: when two or more items are marked low confidence, or the photo shows stacks, price the upper band and offer to drop it if the load fits. Review the log monthly and change the calibration sheet, not the customer.

Insurance, waste paperwork and customers' photos

Three checks sit outside the AI. If you carry other people's goods, confirm your insurance covers goods in transit and public liability for work inside homes. If you carry waste for customers, check which carrier registration and which paperwork apply where you trade before you advertise clearances, and keep the waste site's receipts with the job record. And treat the photos as personal data: keep them only until the job is done, tell customers so in the intake message, and look at your chat assistant's data controls before uploading anything of a customer's home. For a business that fields a lot of enquiries, our tutorial on AI answering for junk removal shows how photo requests fit into phone and message handling.

Photo quotes and route planning for a one-van business

Can ChatGPT really estimate a van load from photos?

Roughly. OpenAI's and Anthropic's own documentation both say counts and spatial judgements from images are approximate, so treat the output as a band with a confidence note and ask the customer about anything hidden. Pricing in bands (quarter, half, three-quarter, full) absorbs most of the error, and a month of logged jobs shows how much margin you need on top.

Do I need route-planning software for five jobs a day?

Usually not. Google Maps handles a start point plus up to nine stops on a computer, and AI can check the order against time windows if you paste in the drive times. Software starts to earn its fee with several vans or more than about a hundred stops a month: Routific's free plan stops at 100 orders a month, and paid route tools start at roughly $125 to $150 a month.

Should I send customers the AI's item list?

Send a short confirmation of what you have priced, meaning the items and the access you assumed, not the raw AI output. Customers can then correct it ('the wardrobe is already in two pieces') before the day, which is the cheapest way to avoid an awkward re-quote on the doorstep.

How long should I keep customers' photos?

Only as long as the job needs, then delete them from your phone, your AI chat history and any shared folder, and say so in your quote message. Photos of a home show belongings and sometimes people. Check your AI plan's data controls too, so chats are not used to improve models.

Further reads

• AI Video Survey Tools for Movers Compared: Cost and Accuracy (2026) — When a house move needs a video survey, not three photos.
• Junk Removal Leads: Follow Up Bought Leads Before a Rival Does — Follow up quote requests before a rival answers them.
• Can a Small Courier Firm Use AI to Plan Delivery Routes? — Route planning for repeat delivery rounds rather than one-off jobs.
• AI Dispatcher for a Towing Company: Calls, Locations and Quotes — Taking location and price from callers who are stuck.
• How Can AI Help Courier Services? Quotes, Dispatch and Proof — Quotes, dispatch and proof of delivery for courier work.
• How Painters and Decorators Can Quote From Photos With AI — Another trade that quotes from customer photos.
• Cutting Parcel-Tracking Calls With AI Delivery Updates — Cut the 'where are you?' calls with proactive updates.
• AI in Logistics for Small Delivery Firms: Routes, ETAs and Messages — A three-van worked example covering route planning, delivery windows, delay messages and proof-of-delivery replies, with named tools and monthly costs.
• Mobile Hairdressing Business: Diary, Travel Time and Rebooking With AI — Plan a mobile hairdresser's diary around drive time: patch days, same-street batching and rebooking texts, with two Tuesdays compared in minutes and income.
• Junk Hauling Software: Dispatch Trucks and Time Windows With AI — Write the dispatch rules first (windows, durations, capacity, dump runs, inserts), then use software and an AI assistant to build and check each day's schedule.
• Solar Panel Cleaning Business: Quote and Route Jobs With AI — A worked pricing sum by panel count and access class, plus where AI helps with quotes, route clustering and recurring-plan reminders.
• Can AI Help a Junk Removal Company? Where to Start — Rank the AI jobs in a junk removal business by hours saved and risk, then start with the one that pays back fastest.
• Can AI Quote Junk Removal From Photos? Truck-Load Pricing Explained — Test AI truck-load estimates against your past jobs, set price ranges by fraction, and word the quote as subject to an on-site check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI and Anthropic image-understanding documentation, Google Maps help (multiple stops), Routific and Spoke Dispatch pricing pages, read in early November 2026.