Vision & Document Understanding
Turn photos, PDFs, screenshots, and field notes into organized next steps.
How the files your business runs on become attached records, draft quotes, and clear next steps — with human review wherever it matters.
What does this layer do?
Deptly reads the photos, PDFs, screenshots, and field notes your business runs on — a customer's photos of a garage to clear out, an insurance form, a completion shot from the crew — extracts what matters, attaches each file to the right customer and job, and drafts the next step. Every extraction can be checked against its source, and anything unclear is flagged to a person instead of guessed at.
The paperwork that arrives as pictures
A large share of what service businesses call paperwork is not text at all. Customers send photos instead of descriptions — the garage to clear, the stain on the ceiling, the dent in the quarter panel. Insurers send PDFs. Crews send completion shots. Someone screenshots a confirmation instead of forwarding it.
Every one of those files carries information someone has to extract, act on, and file — which is why they pile up in a text thread until a person retypes them into the system. Vision and document understanding means AI does the reading and organizing, and a person reviews where it matters.
What it can read and extract
Photo intake for scoping: customer photos become a structured starting point — what is in the space, roughly how much of it, and anything that changes the job, like items your policies treat specially. Forms and PDFs: intake forms, insurance paperwork, estimates, and invoices are read for the fields that matter, so a document becomes data without retyping.
Evidence and completion proof: before-and-after shots from the field are attached to the right job as a record that the work happened, ready for the customer's receipt, a warranty file, or a claim file. Screenshots and forwarded odds and ends get the same treatment — identified, matched, filed.
The matching step leans on business memory: a photo is not just parsed, it is connected — this customer, this property or vehicle, this job. A file with nowhere to land is a question; a file that lands in the wrong place is worse. Matching is treated as carefully as reading.
Visual scope support, not visual verdicts
Photos can support an estimate; they do not replace judgment. Deptly drafts a quote from photos only where your pricing model makes that sound — volume-based junk removal, standard-size jobs — and the draft goes to you for approval before a customer sees it. Where photos cannot honestly support a number, the next step is a site visit, and Deptly schedules that instead.
The same restraint applies wherever a document could be mistaken for a decision. Photos can populate a claim file; carriers and adjusters decide coverage. A clinical image routes to a practitioner. A legal filing goes to the attorney. Reading a document is Deptly's job; ruling on one never is.
Confidence, review, and where files end up
Extraction is probabilistic, and the system is built to be honest about that. Every extracted detail stays linked to its source, so a person can verify in one glance. Clear, standard documents flow through with light review; bad handwriting, unclear photos, and partial documents are flagged for a person rather than guessed at.
Files end up attached to the right record in your systems — the job, the customer, the claim file — under your data settings, used only for your business's work. The output of a photo is never just a parsed blob; it is a filed record and a drafted next step you can see and check.
What it works from
- Photos from customers and your field team — spaces, vehicles, damage, finished work
- PDFs: estimates, invoices, insurance paperwork, intake forms, contracts
- Screenshots and forwarded documents from any channel you use
- Business memory, to attach each file to the right customer, job, property, or matter
- Your pricing and policy records, to turn extracted details into draft quotes and next steps
Everything above comes from your business — your records, your rules, your tools. Nothing is invented, and when information is missing, it becomes a question to you rather than a guess.
What this looks like in practice
Every proof item on this site is labeled for exactly what it is.
Photos become a quote-ready draft — junk removal
Homeowner
Can I get a quote to clear out this garage? Photos attached — pretty much everything can go.
8:03 AM, 4 photos
Deptly
Read the photos: roughly a half-truck load — furniture, boxes, a treadmill. Paint cans spotted and flagged under your hazardous-item policy: not quotable without your call. Draft quote built from your volume pricing.
8:05 AM
Deptly
Needs your OKDraft: $280 half-truck rate, paint cans excluded pending your decision. Photos and the extracted item list attached for a quick check. Approve, adjust, or switch to an in-person quote?
8:06 AM
Owner
Looks right. Send it, and note we can take the paint for a disposal fee.
8:11 AM
Deptly
Quote sent with the paint-disposal option included. When the crew finishes, their completion photos will be attached to the job and included with the customer's receipt.
8:12 AM
Where humans stay in the loop
This layer runs inside boundaries, not on its own. Here is who reviews what — and the limits it never crosses.
Human review
- Extracted details are shown next to their source photo or document, so a person can verify before anything depends on them.
- Low-confidence reads — bad handwriting, unclear photos, partial documents — are flagged for a person instead of guessed at.
- Quotes and estimates built from photos are drafts for your approval, never sent on the system's own judgment.
- Photos can support an insurance claim file, but coverage decisions belong to carriers and adjusters — and clinical images and legal filings route to qualified professionals, not to software.
Guardrails
- Documents are read to organize work, not to make determinations — a photo can populate a draft, never finalize a claim, diagnosis, or legal position.
- Items with safety, hazard, or compliance implications spotted in photos are flagged to a person rather than priced or dismissed automatically.
- Customer photos and documents are attached to the right job in your systems, retained under your data settings, and used only for your business's work.
- Completion-proof photos document that work happened — disputes about quality remain human conversations.
The full rulebook — control levels, approvals, and audit history — lives on trust and control.
Where this layer shows up
Technology is only interesting when it carries real work. These are the responsibilities this layer powers — and the layers around it.
Responsibilities it powers
The layers around this one
Questions owners ask
Photos, PDFs, screenshots, and scanned or photographed documents — the mix a service business actually receives: customer photos of a job, insurance paperwork, intake forms, invoices, completion shots from the crew. Each file is read for the details that matter, matched to the right customer and job, and filed in your systems.
Honest answer: extraction is probabilistic, and quality of input matters. The system is built around that — every extracted detail stays linked to its source for one-glance verification, and low-confidence reads are flagged for a person instead of guessed at. We will not quote an accuracy percentage, because the meaningful number depends on your documents, not a benchmark.
Only where your pricing model honestly supports it — volume-based work like junk removal, or standardized jobs with clear visual scope. Even then, the quote is a draft for your approval, never auto-sent. Where photos cannot support a responsible number, the drafted next step is a site visit, not a guess.
No, and it is built so it cannot. Photos and documents can be organized into a complete, well-labeled claim file, and Deptly can chase missing documentation — but what insurance covers is decided by carriers and adjusters. Deptly never tells a customer what their policy will pay.
Sensitive documents — patient records, legal filings, financial paperwork — carry stricter handling defaults: tighter access, approval-first processing, and routing to the qualified person rather than into an automated flow. During onboarding we walk through exactly how your document types are classified and handled, including any compliance requirements your industry carries.
Attached to the right record in your own systems — the job, the customer, the claim file — not marooned in a separate library you have to search. Files are retained under your data settings, used only for your business's work, and exportable or deletable on request.
Yes — that is one of its best uses. A crew lead texts completion photos and a quick note; Deptly attaches them to the job, updates the status, and stages whatever comes next, like the invoice or the customer's receipt. The crew's habit stays a text; the filing becomes automatic.
Under the hood, then in practice
See what Deptly could take off your plate
Answer a few questions about how your week actually runs and get an honest read on which responsibility this machinery should carry for you first.