The people in this story are made up. The problems are the ones we hear most.
Three demos, three different products
Demo 1
- clean sample
- looks perfect
Demo 2
- knows invoices only
- no credit memos
Demo 3
- good extraction
- manual CSV upload
Elena runs operations for a regional food distributor. Her team handles supplier invoices, customer purchase orders, delivery receipts and credit memos, all arriving as PDFs and scans. She sat through three software demos in one week.
The first vendor's demo used a perfectly scanned sample invoice, and everything worked. The second tool only knew invoices and receipts, not credit memos. The third extracted beautifully, then handed her a spreadsheet to upload into her ERP by hand. By Friday she wasn't sure what she was comparing.
The demos weren't wrong. She just hadn't decided what questions mattered before watching them.
The 10 questions
Your documents
- Q1 own documents
- Q2 new types
Your team
- Q3 who changes workflows
Your systems
- Q4 sources in
- Q5 data out
Trust
- Q6 uncertain documents
- Q7 see the original
- Q8 accuracy
- Q9 data handling
Cost
- Q10 time to first workflow
1. Will it work on our documents, not your sample?
Bring your own worst examples: crooked scans, phone photos, the supplier with the strange layout. A demo on clean samples tells you little.
2. Can we train it on document types you don't already know?
Every business has forms no vendor has seen. Ask how you add a new type, and who does it: you, or their services team?
3. Who changes the workflow when our process changes?
If every change needs a developer or a support ticket, you'll stop changing it. A no-code canvas puts that in the hands of the people who know the paperwork.
4. Where do documents come in from?
Match the vendor's ingestion options to where your documents actually arrive today. Changing how customers send paperwork is a hidden cost.
5. Where does the data go afterwards?
Ask whether it can call your software's API or save to your database. "Export to CSV" means someone is still copying data by hand.
6. How does it handle documents it isn't sure about?
Every tool makes mistakes. What matters is whether uncertain or unrecognized documents go to a person instead of silently into your system.
7. Can we see the original next to the extracted values?
When a number looks wrong, someone needs to check the source quickly.
8. How do we measure accuracy on our own documents?
Ask for a side-by-side run on a batch of your real files before you commit, and decide in advance what accuracy you need for each field.
9. What happens to our data?
Where documents are stored, for how long, who can see them, and whether they're used to train anything beyond your own workflows. For medical, legal or financial documents, ask about the specific rules that apply to you.
10. What does the first workflow cost in time, not just money?
Ask how long it takes to go from sample documents to a working first workflow, and what you need to provide.
How to run the comparison
- Pick one document type
- Collect 20 to 50 real examples
- Same set to every vendor
- Score the 10 questions
- Choose on your documents
Pick one document type that causes the most retyping. Collect 20 to 50 real examples, including the messy ones. Ask each vendor to run the same set, and score the answers to the ten questions above in a simple table. The tool that wins on your documents, not on its demo, is the right one.
Where Document Processing Factory fits
- Q2 train your own types
- Q3 no-code canvas
- Q4 S3, SharePoint or email
- Q5 API call or database
We built Document Processing Factory around questions 2 to 5: you train the classifier on your own document types, build and change workflows on a no-code drag-and-drop canvas (or have us build them for you), bring documents in from Amazon S3, SharePoint sites or email, and send the extracted data to your software with an API call or to a database. Ask us the other questions too, and test it on your own files.