Can AI read handwritten and PDF orders? The honest state

Can AI read handwritten and PDF orders?

Yes — AI that reads handwritten, photographed, and even voicemailed orders exists and is shipping today. Conexiom advertises extraction from "PDF, Excel, email, CSV, even handwritten notes"; Choco digitizes orders from "voicemail, WhatsApp, email, text, photos and more"; Pepper's Order Agent takes "email, text, photo, and PDF"; Cut+Dry's help docs list handwritten notes and images among what its AI Order Desk drafts from; Parseur's OCR covers 200+ languages including handwriting.

PDFs are the question's second half, and they split in two. A text-layer PDF — the digital-born kind that procurement software, Word, and Excel produce — is reliably readable, across essentially the whole category. A scanned or photographed PDF is an image, which needs OCR first — and OCR is the fork in the market: some tools include it, and others, PeasyOrders among them, deliberately don't.

This post is the honest map of that split: what's technically going on, who reads what today by their own pages, and why we chose the narrow side of the fork.

What's actually inside a PDF?

The word "PDF" hides two different objects. A text-layer PDF carries its text inside the file — every character is data, sitting at a position on the page. Software reading it isn't guessing what the page says; it's reading what the file contains. That's why text-layer extraction is reliable: wrong interpretations are possible, but wrong characters essentially aren't.

A scanned or photographed PDF contains no text at all — just pixels. Before anything can be extracted, OCR has to decide which pixels are which characters, and that decision can go wrong quietly: a smudged 8 becomes a 3, a handwritten 7 becomes a 1, and the output looks exactly as confident either way. The practical test takes two seconds: if you can select and copy text in your PDF viewer, there's a text layer; if your cursor just draws a box on the page, it's an image.

Handwriting is the hardest version of the image problem — strokes vary by writer, characters overlap, and there's no font to anchor on. It's genuinely impressive that shipping products handle it at all. It's also exactly where a misread hides best, because the error arrives formatted like a success.

A text-layer PDF arrives

The text is in the file

digital-born: exported from procurement software, Word, or Excel

Parsed directly

read by meaning — customer, items, quantities — no guessing at characters

Matched, priced, reviewed

a draft a person confirms

Reliably automatable — across the category, PeasyOrders included

A scanned or photographed PDF arrives

No text in the file — only pixels

nothing to read until OCR decides what the image says

Some tools OCR it

Conexiom-class tools and Parseur do — misreads can arrive looking like successes, which is why review matters

In PeasyOrders: kept, not parsed

the attachment stays on the order — worked in a side-by-side view, typed once next to the source

Either way, a person closes it — by reviewing the OCR or by entering it side by side

Who reads what today?

By each vendor's own pages — capabilities conceded plainly, because they're real:

Tool	What its own pages say it reads
Conexiom	"PDF, Excel, email, CSV, even handwritten notes" — enterprise order automation with exception review
Choco	Orders from "voicemail, WhatsApp, email, text, photos and more"
Pepper	"Email, text, photo, and PDF" — help docs add the email body, PDF order forms, spreadsheets, photos, and voicemails
Cut+Dry	"Emails, texts, PDFs, and voicemails" — help docs add handwritten notes and images
Parseur	Emails, PDFs, and documents, with OCR in 200+ languages including handwriting
PeasyOrders	The email body, the PDF attached (text layer), and the spreadsheet — no OCR; anything unparseable is kept and worked side by side

Two honest notes on that table. First, reading a format isn't the whole job — extraction still has to become a matched, priced, reviewed order, and every tool above handles that differently. Second, the broad readers keep humans close: Pepper's captured orders become drafts a rep reviews and confirms into the ERP, Cut+Dry's docs say it "won't create the order for you — you need to review and submit it," Conexiom routes orders needing attention to exception review, and Parseur documents an optional review step. The wider the input funnel, the more the review step matters.

Why doesn't PeasyOrders read handwriting or scans?

Because for the orders it's built around — orders arriving by email at small and mid-sized US wholesale distributors on QuickBooks Online — the text layer carries most of the volume we see, and reading only the text layer keeps OCR's silent-misread class of errors out of the pipeline entirely. That's the design reasoning, stated as a trade-off rather than a virtue: an OCR pipeline reads more and has to be double-checked more; a text-layer pipeline reads less and never hands you a confidently wrong character.

And for what the text layer doesn't cover, the answer isn't rejection. An attachment PeasyOrders can't parse — a scan, a photo, a fax — stays on the order, and your team works it in a side-by-side view: the original document on one side, the order editor on the other, the lines typed once next to the source. Same per-customer pricing, same review, same export to QuickBooks Online as an Estimate. Nothing is bounced for its format, and nothing is guessed silently — a person reviews everything before it touches QuickBooks either way.

The plain concession that goes with this: if handwritten faxes, photographed order sheets, or voicemails carry a real share of your volume, the OCR-capable tools in the table above are where to look — that's their territory, honestly earned. And one line on the roadmap, undated because we don't date what hasn't shipped — photos & scans: coming soon.

How do you choose?

Count a week of your actual inbound orders and sort by what they physically are:

• Mostly emails with text-layer PDFs and spreadsheets — the whole category reads these; the narrow tools do it without a platform attached. This is PeasyOrders' home ground.
• A real share of photos, scans, or handwriting — you want OCR in the tool: Conexiom at enterprise scale, Choco, Pepper, or Cut+Dry in food distribution, Parseur if you're building the pipeline yourself.
• Voicemail as a load-bearing channel — Choco, Pepper, and Cut+Dry all name it. PeasyOrders doesn't capture calls in any form; a phoned order is added manually in one click into the same reviewed queue.
• A mixed bag with an occasional scan — the side-by-side lane exists for exactly that occasional one: kept, worked once next to the source, never lost.

The bottom line

Can AI read handwritten and PDF orders? Handwriting: yes, in shipping products — Conexiom, Cut+Dry, and Parseur document it — with review steps standing behind the OCR for good reason. PDFs: reliably when there's a text layer, only via OCR when there isn't. PeasyOrders sits on the deliberate side of that fork: text layer read completely, images never guessed at, unparseable attachments kept and worked side by side, and a person reviewing everything before it touches QuickBooks. If your orders are emails into QuickBooks Online, that narrow answer is the dependable one.

PeasyOrders starts at $99 a month with a 30-day money-back guarantee — see pricing.