Electronic Proof of Delivery: Settle Disputes With Photos and AI

A delivery business settles most disputes by capturing four things at the door (a photo of the parcel in place, a timestamp, the location of the stop and a name or signature where needed) and filing them under the order. AI then assembles those records into a one-page dispute pack and drafts a neutral reply. A person still decides the outcome.

The weak link is almost never the software. It is the thirty seconds at the door: a driver who photographs a blurred porch, a stop closed with no note, or a photo that cannot be matched to an order. Fix the capture habit first, then use AI to save the office hour you spend assembling evidence once a customer says "it never arrived".

Follow me on Instagram@sagnikteaches

What a defensible delivery record contains

Think of each delivery as a small file. When a customer says a parcel never came, you want to show, without a phone call to the driver, that this item reached this place at this time. A complete record has six parts:

Connect on LinkedInSagnik Bhattacharya

1. Order or consignment number that matches the customer's own reference.
2. Timestamp generated by the app, not typed in by the driver.
3. Location stamp showing the stop was made near the delivery address, from the driver's device.
4. Photo of the parcel in place, with a house number, door or other landmark visible in the same frame.
5. Recipient name or signature, where your agreement or the item's value calls for one.
6. Driver note, for any exception: left with neighbour, safe place, no access, damaged on arrival.

Most proof-of-delivery apps, including the ones that come with route-planning and courier software, capture these automatically once you switch the settings on. What differs is whether the timestamp and location are locked in by the app or editable afterwards. Ask any vendor whether a driver or an office user can alter a record after the stop is closed, and whether changes are logged. A record that can be edited quietly is weak evidence.

Subscribe on YouTube@sagnikteaches

If you are just starting and have no app yet, a shared form that asks for the order number, a photo and a note, and stamps each submission with the time, can work for a small team. It is less tidy than a dedicated tool, but it beats a driver's text message.

Set the capture rules drivers actually follow

Write a one-page rule sheet and put it in every vehicle. Keep it to what changes the outcome of a dispute.

• Photograph the parcel where you left it, from about a metre away, with the door or house number in view.
• Take one wide photo and one close photo when leaving the parcel in a safe place. The wide shot proves the location; the close shot proves the item.
• Photograph damage at the door, before the recipient takes the parcel, whenever you notice it.
• Never include people, children, interiors or number plates in a frame. If someone is in view, retake it.
• Write the exception in the note before you leave the stop, not at the end of the shift.
• If you cannot deliver, photograph the attempt notice or the locked gate, and note the time.

Test the rules for a week with two drivers and review fifty photos together. You will find the same problems every time: photos taken from too far away, thumbs over the lens, wet lens in rain, flash glare on a glossy door. Five minutes of coaching on those fixes more than any software setting.

Build the dispute pack in minutes with AI

When a claim arrives, an office worker typically hunts through the app, the customer's email thread and the driver's messages. AI is good at the dull step in the middle: laying the facts out in order and spotting what is missing. Here is a repeatable method.

1. Export the record for that one delivery: the data fields and the photos. Do not export the whole day's route.
2. Paste the customer's claim and any messages exactly as they wrote them.
3. Ask the assistant to build a timeline, compare the address in the claim with the address in the record, and list what proof exists and what is missing.
4. Review the result yourself, including looking at the photos.
5. Send the pack to the customer or your own customer, or file it with a carrier or insurer if needed.

A prompt that works:

You are helping a small delivery firm prepare a dispute pack. I will paste one delivery record and the customer's complaint.

Return:
1. A timeline of events with times, in plain sentences.
2. A table: evidence item, what it shows, any weakness (for example "photo does not show house number").
3. Discrepancies between the complaint and the record (address, time, item, who received it).
4. What additional evidence I could still obtain (driver statement, neighbour, camera footage at a depot).
5. A neutral, polite draft reply that states what the record shows without accusing anyone. Use [first name] and [your name] as placeholders.

Do not decide who is right. Do not invent facts. If something is not in the record, say "not recorded".

DELIVERY RECORD:
...

COMPLAINT:
...

The wardrobe claim, laid out by the assistant

Say a small furniture delivery firm drops a flat-pack wardrobe, and the customer writes: "Nothing arrived on Thursday." Illustrative output from the prompt above might include this table:

You now see the case clearly, including its flaws. What would you fix? The assistant may state the box was "delivered to the correct address"; the record only supports "delivered near the correct address, in a porch, with a label matching the order". Edit that sentence before anything goes to the customer. A dispute reply that overstates the evidence can lose a dispute that a careful one would have won.

Write the reply the way a customer wants to read it

The tone of the first message often decides whether a customer escalates. Ask the assistant for a short, warm reply that does three things: acknowledges the problem, says what the record shows in plain terms, and offers a next step. Here is the shape, with placeholders:

Hello [first name],

I'm sorry the wardrobe hasn't reached you. I've checked the record for Thursday: our driver closed the stop at 14:12 and photographed the boxes in your porch beside a green door, with the order label visible. I've attached both photos.

Could you check around the porch, with any neighbour who may have taken it in, and let me know by [date]? If it still can't be found, I'll start a trace with the driver and come back to you the same day.

[your name]

Notice what the draft avoids: no blame, no "our records prove". For ideas on the wider handling of unhappy customers, this guide to AI and customer complaints sets out where assistants help and where they make things worse.

Four more situations and what the pack should contain

"Delivered to the wrong house"

Match the location stamp and wide photo against the address, then ask the assistant to compare the street number in the photo with the order. If the number is not visible, say so honestly and offer a driver statement. A straight "the number isn't in the photo, so I can't prove it either way; here is what I can do" keeps the conversation calm, where insisting on a delivery you cannot show tends to push the customer towards a chargeback or a public review.

"The parcel arrived damaged"

You need the packaging at pickup, any damage photo at the door and the recipient's own photo. Ask the assistant to list differences between the pickup condition and the claimed condition. Damage claims often belong with the sender or the carrier's insurance, so the pack must go in whatever format that party needs.

"Someone else signed for it"

Signatures are only useful if the name is printed alongside them, since a scrawl proves little. Where a name was captured, check it against the customer's household or business. If the signature app captured only a squiggle, add a rule that drivers type or write the printed name too.

A repeat claimant

When one address makes several claims in a month, the evidence pack matters, but so does a calm process. Ask the assistant to produce a simple table of past claims for the address, with dates, outcomes and photos available, and take it to a person for a decision. Never let the assistant label a customer as dishonest. Patterns are a prompt for a conversation, not a conclusion.

Claim handling before and after, in hours and dollars

Say a six-driver local delivery firm handles 220 deliveries a day, about 4,800 a month. Before it changed its process, 1.5 percent of deliveries (72) produced a "not received" claim. Each claim took an office worker about 25 minutes to chase, which is 30 hours a month. Roughly a third of claims ended with a refund or reshipment because the firm could not show proof, and the average cost was $35 per case.

After the capture rules and a dispute-pack prompt, the same 72 claims took about 8 minutes each: 9.6 hours. Because the records were stronger, the share ending in a refund fell to about one in five, roughly 14 cases at $35, which is $490 instead of about $840. The time saving is about 20 hours a month, and the cost saving about $350, before counting happier customers. These figures are illustrative; your own claim rate and handling time will differ, and you should measure for a month before and after. The method matters more than the totals: count claims, minutes per claim and outcomes, then compare.

Privacy of recipients: what you hold and for how long

Delivery photos show front doors, house numbers, sometimes people, children's toys, parked cars and the interiors seen through open doors. Add the recipient's name and address and you hold personal data. The aim is to collect what you need to settle a dispute and no more.

• Collect less. Photograph the parcel and the door number; avoid faces and interiors. If drivers use their own phones, make sure photos are taken inside the delivery app rather than the phone's camera, so copies do not sit in personal galleries or personal cloud backups.
• Control who sees it. Office staff who handle claims need access; the whole company does not. Use separate logins, not a shared password.
• Set a retention period that matches your claim window, write it down, and delete automatically when it ends. Longer-than-needed storage is a risk with no benefit.
• Tell recipients. A short line in your delivery confirmation ("We take a photo of your delivered parcel as proof of delivery and keep it for 90 days") is honest and often reduces complaints. Pick your own period; 90 days is only an example.
• Honour requests. If someone asks to see or delete a photo, you need a way to find it by order number.

Data-protection duties depend on where your customers live; if you have customers in the EU, take advice on lawful basis, retention and any need for a privacy notice. This tutorial gives practical habits, not legal advice.

Using AI tools without leaking those photos

Treat an assistant like any outside processor. Upload only the one delivery in dispute, crop out anything unrelated, and blur faces if any slipped in. On consumer plans, check the setting that controls whether chats help train the model. OpenAI's help centre says it does not use content from ChatGPT Business or Enterprise to improve its models by default, and Anthropic's privacy centre says the same of its commercial plans, such as Team and Enterprise. Settings and terms change, so read the vendor's current privacy page before uploading. Never upload a full route or a customer list to get a summary. If you need AI across the whole day's records, the tutorial on reading delivery notes with AI covers how to handle documents in bulk with fewer personal details.

Choosing the capture tool: ten questions to ask a vendor

If you are picking or reviewing a proof-of-delivery app, the features list matters less than how the records behave afterwards. Ask these, and write down the answers in a one-page comparison.

1. Are timestamp and location generated by the device and locked once the stop is closed?
2. Can an office user or driver edit or delete a record afterwards, and is every change logged with a name and time?
3. Can the photo be required before a stop can be completed, or can a driver skip it?
4. Does it work with no signal, storing the record on the phone and sending it later with the original timestamp?
5. Can I export one delivery as a single PDF with photos and data, so a dispute pack takes seconds?
6. Can I search by order number, address and date?
7. Can I set an automatic retention period, and does deletion remove photos from backups?
8. Where are the photos stored, and who at the vendor can see them?
9. Can I give a customer a link to their own proof without them logging in?
10. What does it cost per driver or per stop, and is there a limit on stored photos?

Questions 1, 2 and 5 do the most work. A record that cannot be altered and exports cleanly makes every later step easier, including the AI step. Prices and plans change often, so confirm current costs on the vendor's own page rather than relying on an article.

Before and after: one driver note rewritten

Driver notes are where records go soft. Here is a typical before and after, which you can use in your driver guide.

You can ask an assistant to rewrite a week's worth of vague notes into this format as training material, with names and addresses removed. It is a quick way to show drivers what you mean without a lecture.

When the dispute is with a carrier or a customer's insurer

Many small firms do not only receive claims; they raise them. If you hand a parcel on to a larger carrier and it goes missing, or a business customer claims against your cover, the pack has to meet someone else's format. Ask for their claim form first, then ask the assistant to map your record onto each field and list anything the form wants that you do not have. Typical gaps are item value, packaging description and the weight at pickup, which is why a pickup photo and a weight note belong in the capture rules too. Keep claim deadlines in a shared calendar: many carriers and insurers set short windows, so check the time limit in your own contract.

The same records feed invoicing. A trucking firm that cannot show a signed delivery record waits longer for payment; this tutorial on chasing late invoices and missing PODs shows how to use the proof you already collect to speed up payment as well as settle disputes.

Check the process is working

After a month, review four numbers and one sample.

1. Claims per 1,000 deliveries.
2. Average minutes to resolve a claim.
3. Share of claims where you could produce a photo and timestamp.
4. Share resolved without refund or reshipment.
5. The sample: pull twenty random delivery records and score each photo (parcel visible, number visible, no people, in focus).

If the third number is under 90 percent, the problem is capture, not office work. If photos score well but the fourth number stays low, your reply wording or policy may be the issue. For the front-end of the same problem, sending proactive delivery updates with AI prevents many "where is it?" claims from ever becoming disputes, and the courier workflow tutorial places proof in the whole job from quote to invoice.

When proof is not the problem

Sometimes the record is perfect and the customer still complains. The parcel was delivered but taken from the porch; the customer expected a signature you never promised; the package was left in the "safe place" the customer chose but then forgot. These are policy questions: what you promised, who bears the risk once the parcel is in a safe place, and what you will do as goodwill. Write those rules once, state them in your terms and delivery messages, and keep the dispute pack to facts. When a case needs a person to decide on goodwill, give them the pack and the cost of each option, not an AI recommendation.

If you would like to map the capture rules, tools and dispute steps for your own routes, that is the sort of job an AI implementation consultation covers.

Proof of delivery questions from small delivery firms

Is a photo enough, or do I also need a signature?

A photo of the parcel in place, with a timestamp and the delivery address visible, settles most 'not received' claims. A signature adds weight for high-value or age-restricted items and where a contract requires it. Use the strongest proof your customer agreements ask for.

Can AI decide who is right in a dispute?

No. It can lay out the evidence, spot gaps such as a missing timestamp or mismatched address, and draft a neutral reply. A person decides the outcome, and a customer's right to challenge it stays with them.

How long should I keep delivery photos?

Keep them for the period in which a claim can realistically arrive, often your customer's complaint window plus a margin, then delete them. Choose a fixed period, write it down and automate deletion. Ask an adviser if you have customers in the EU, where data-protection rules apply.

Should drivers photograph people at the door?

No. Instruct drivers to photograph the parcel and the door or house number, not the person, and to avoid children, interiors and number plates. Keep the instruction in the driver guide and check a sample of photos each month.

Can I upload delivery photos to ChatGPT or Claude?

You can, but photos show addresses and sometimes people, so use a business plan with training off by default, upload only the disputed delivery, and never paste whole route files. Consumer plans differ; check the privacy setting before uploading anything with personal details.

Further reads

• AI Courier Quotes: Reply to Delivery Enquiries From Your Rate Card — Quote replies that set expectations before delivery.
• Can a Small Courier Firm Use AI to Plan Delivery Routes? — Plan routes so late deliveries drop.
• AI Route Planning for Small Delivery Fleets and Mobile Teams — Route planning for larger vans and mobile teams.
• How to Automate Returns and Refunds With Clear AI Rules — Rules for refunds once a dispute is settled.
• Can ChatGPT Read PDFs, Spreadsheets and Photos? What Breaks — What assistants read well from photos.
• AI in Delivery and Courier Services: Handling Damage Claims and Disputes — Sort damaged-parcel and missed-delivery claims from photos and delivery records, then draft careful replies that a named person approves.
• Gutter Cleaning Invoices With Photo Proof and AI Job Summaries — A shot list, a voice-note-to-summary prompt, invoice wording and a dispute log: photo proof for gutter jobs that settles complaints early.
• Process Server Software: Dispatch, Proof of Service and Where AI Fits — Nine jobs process server software must do, four tools checked against them, a 140-job cost sum and where AI can help with intake and drafts.
• AI Dispatch Automation for Small Delivery Firms: Where to Begin — A small delivery firm's first dispatch pilot: capture jobs, suggest assignments, handle exceptions and measure results.
• Carrier Vetting Checklist: AI Checks With a Human Sign-Off — A grouped carrier vetting checklist with red, amber and green rules, double-brokering signals and a sign-off block a named person completes.
• Chargeback Response Template for Online Stores, Drafted With AI — Reason-code mapping, an evidence checklist and a response template you can fill with AI, plus the rule that nothing gets invented.
• PayPal Dispute Response for Sellers: Evidence Pack Drafted With AI — A seller-side PayPal dispute response: evidence by dispute reason, a reply template and AI prompts that keep the pack honest.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI help article on how data is used to improve model performance; Anthropic privacy centre article on model training for commercial products; general proof-of-delivery practice.