Restaurant group
San Francisco, CA
The invoice is the delivery ticket, signed at the back door, with short shipments written on it in pencil.
The setup
The company and its vendors are made up. It still has its own QuickBooks company, its own local sales tax, and documents written the way this trade writes them.
Its own QuickBooks company
Nobody approved or corrected anything along the way. A document one of DocStreamAI's checks held for review stayed held, so every result is exactly what was read.
12 vendors in 10 trades, writing on 10 different layouts.
From
Broadline food distributors
To: Fennick & Thorlow Restaurant Group
From
Produce
To: Fennick & Thorlow Restaurant Group
From
Protein & seafood
To: Fennick & Thorlow Restaurant Group
From
Dairy
To: Fennick & Thorlow Restaurant Group
From
Beverage & alcohol
To: Fennick & Thorlow Restaurant Group
From
Paper & disposables
To: Fennick & Thorlow Restaurant Group
From
Linen service
To: Fennick & Thorlow Restaurant Group
From
Grease & hood cleaning
To: Fennick & Thorlow Restaurant Group
From
Pest control
To: Fennick & Thorlow Restaurant Group
From
POS software
To: Fennick & Thorlow Restaurant Group
The pile
Invoices
Receipts
Credit memos
Vendor statements
110 documents in 109 PDF files. The 5 vendor statements list invoices that were also sent, so the right number of new bills from them is zero.
43% arrived as paper
47 of 110 were scanned, photographed, faxed or photocopied. 63 were clean files.
The traps
The invoice is the delivery ticket: bill, packing slip and receiving record in one document, signed at the back door. Broadline distributors deliver several times a week, protein is billed at actual shipped weight instead of by the case, and drivers write short-ships onto the printed page in pencil.
The driver wrote on it
A case was short. The printed total no longer matches what is owed, and only the pencil says so.
Billed by the pound that shipped
Protein is invoiced at actual weight, so the case count and the price never multiply to the line.
The same invoice, twice
Paper from the driver at the door, then the same invoice again as a PDF the next morning.
A loyalty block in the money column
Year to date purchases sit beside the invoice total and are eighty times larger.
Excerpts drawn in the shape of this industry's documents, with invented names. The full set carries 18 kinds of difficulty across all five businesses. See every one
The results
Measured in two places. Inside DocStreamAI: what was read, what kind of document it is, whether it made it from the email, which vendor it belongs to and where it was categorized, each checked against the answer key written for every document. Then in QuickBooks: whether the record that landed matches.
54 text PDFs from Fennick & Thorlow Restaurant Group, the kind billing software makes. 53 reached extraction and were scored against their answer keys.
Extraction
Values read correctly off the page, every field pooled.
Every key field right*
Documents with every key field correct at once.
Detection
Invoice, receipt or credit memo, identified correctly.
Intake
Documents that made it from the email to a record DocStreamAI read.
Vendor statements kept out: 3 of 3
Vendor
Read after the vendor was in QuickBooks, and found it. How this is counted is on the summary page.
Categorization
Expense account matches the account the chart intends.
* Line descriptions, quantities and unit prices count as lower severity and are left out of this figure. QuickBooks records each line by its amount, so a wrong quantity or unit price does not change what lands in the books. Counting them too: 81.1% (43 of 53).
Extraction, field by field
Vendor name
100.0%
53 of 53 read correctly
Document number
100.0%
53 of 53 read correctly
Document date
100.0%
53 of 53 read correctly
Due date
98.1%
52 of 53 read correctly
Currency
100.0%
53 of 53 read correctly
Subtotal
96.2%
51 of 53 read correctly
Tax
96.2%
51 of 53 read correctly
Total
96.2%
51 of 53 read correctly
Number of lines
96.2%
51 of 53 read correctly
Line descriptionsLower severity
86.8%
46 of 53 read correctly
Line quantitiesLower severity
92.5%
49 of 53 read correctly
Unit pricesLower severity
94.3%
50 of 53 read correctly
Line amounts
96.2%
51 of 53 read correctly
Scanned PDFs
Scans, phone photos and faxes, on their own
Reported apart from text PDFs because one part of handling them is still being built.
43 scanned PDFs sent: office scans, phone photos, faxes and photocopies saved as PDF. 35 reached extraction and were scored.
Once a scan is read
Fields read correctly
Every key field right*
Type detected
Text PDFs, for comparison: 96.4% of fields read correctly, and every key field right* on 94.3% of documents.
Getting a scan to that point
The only attachment on its email
35 of 35
Picked up and read every time.
Sent with other attachments
0 of 8
Not picked up yet. The other attachments on the same email were read as normal.
Scanned vendor statements kept out
0 of 2
Not recognised as statements yet. 2 of the 2 were posted as bills.
Planned for a future update: picking up scanned PDFs that arrive alongside other attachments, and recognising a scanned vendor statement for what it is.
* Line descriptions, quantities and unit prices count as lower severity and are left out of this figure. QuickBooks records each line by its amount, so a wrong quantity or unit price does not change what lands in the books. Counting them too: 82.9% (29 of 35).
In QuickBooks
What landed in the books
The record in QuickBooks matches the document and what DocStreamAI read, per record type.
Created from invoices.
Not measured in this run.
Created from credit memos.
Not measured in this run.
Created from receipts.
Waiting on the card match. A receipt posts once its payment card is matched to an account in QuickBooks, and that is not set up in these companies yet.
We use analytics and advertising cookies to understand how visitors use the site and to measure our marketing. See our Privacy Policy.
