65 Warehouse Documents, One Spreadsheet
How to Batch-Process Handwritten Receiving Slips

The weekly pile: 50+ documents that all need typing

In a typical small-to-midsize warehouse with paper-based receiving, the document inflow follows a predictable rhythm. Monday morning: the weekend's deliveries, left at the dock on Saturday by carriers who don't wait for receiving inspection. Tuesday through Thursday: 6 to 8 deliveries per day, each with a delivery note, a packing slip, and sometimes a certificate of analysis or a batch traceability sheet. Friday: rush deliveries for weekend production orders, plus the weekly cycle count cards from the inventory team.

By Friday afternoon, the receiving desk holds somewhere between 50 and 80 individual documents. Each one needs to be entered into the inventory system — QuickBooks, Fishbowl, an Excel tracker, or the company's ERP. The data entry clerk, who might also handle purchase order matching and supplier inquiries, processes these one at a time: pick up the top document, squint at the handwriting, type the PO number, type the item codes, type the received quantities, check for discrepancies, move to the next document. Repeat 65 times.

The one-at-a-time approach doesn't just take time — it makes errors harder to catch. When you process 65 documents sequentially over 8 hours, your error rate increases with fatigue — the documents you type at 4:15 PM get a tired brain and tired eyes.. A 2026 Lido analysis on manual data entry tracked error frequency across a workday: 0.5% error rate at 9:00 AM, climbing above 3% by 4:00 PM. The documents you type at 8:30 AM get the best version of your attention. The documents you type at 4:15 PM, after processing 57 previous documents, get a tired brain and tired eyes. That's when the handwritten "80" becomes a typed "30."

Why one-at-a-time typing breaks at scale

The efficiency problem with sequential document processing isn't just the sum of individual typing times. It's the task-switching overhead between documents and the compounding fatigue effect. Here's the breakdown:

Context switching per document. Each new document requires the data entry clerk to orient themselves: what type of document is this (delivery note vs. packing slip vs. cycle count card)? Which supplier? What's the handwriting style? Where are the key fields on this particular layout? This orientation takes 15-30 seconds per document — essentially free for the first few documents, but cumulatively significant across 65. At 30 seconds per orientation, that's 32 minutes of pure context-switching per batch.

Error propagation without cross-checking. When you process documents one at a time, you have no reference for what a "normal" value looks like. If document #23 has a handwritten quantity that's ambiguous — could be 50 or could be 80 — you guess based on what looks right. If you were processing all 65 documents into a single table simultaneously, you'd see that every other receipt for that item shows quantities of 50, making your guess an educated one. Batch extraction does this automatically: the AI reads all documents with the same column definitions, producing a table where patterns and outliers are visible at a glance.

Tracking what's been done. Which of the 65 documents have been entered? Which are pending? Did document #41 get entered but the data entry clerk forgot to stamp it "ENTERED"? Paper-based tracking of data entry progress is itself a source of errors — Documents get entered twice (inflating inventory) or missed entirely (creating phantom shortages). (creating phantom shortages). A batch process solves this: upload all 65 documents, process them as one set, and the output is either complete or not — there's no ambiguity about which individual documents have been touched.

Batch strategy: how to organize a week's worth of warehouse paperwork

Before uploading, a few minutes of organization pays off in cleaner output and easier review. Here's the grouping strategy that works for most warehouse operations:

Group by document type first, not by date or supplier. Goods receipt notes, dispatch confirmations, and cycle count cards have fundamentally different data structures. A goods receipt needs columns for PO number, supplier, item, ordered quantity, received quantity, batch number, and condition notes. A dispatch confirmation needs columns for sales order number, customer, item, ordered quantity, picked quantity, carrier, and tracking number. A cycle count card needs columns for location, item, system quantity, counted quantity, variance, and counter ID. Processing these three document types in the same batch with the same column definitions produces rows where most columns are empty — the delivery note columns don't apply to the cycle count card. Process each document type as a separate batch with its own column definition.

Within a document type, batch by processing week or day. Monday's receiving documents plus Tuesday's plus Wednesday's — batch them together if they share the same column structure. The output Excel will have a column for receipt date, so you can filter or pivot by day after extraction. There's no advantage to processing each day's documents separately; a larger batch amortizes the column definition and upload overhead across more documents.

Flag damaged or questionable documents before upload. If a delivery note has water damage that makes half the fields illegible, set it aside. Process the clean documents through AI extraction, then decide whether the damaged document is worth the extraction attempt or faster to type manually. The AI will flag the illegible fields, but if the document is 50%+ damaged, the review time may exceed the manual typing time.

Step-by-step: from a stack of mixed documents to one combined Excel

Step 1: Scan or photograph the batch

For a batch of 30 goods receipt notes, a document scanner with ADF (automatic document feeder) processes the stack in 2-3 minutes. Remove staples, flatten folded documents, and feed them through. The scanner produces individual PDF or JPG files — one per document.

If you're collecting documents from multiple docks or locations, Collection Link removes the physical collection step. Generate a shareable upload link, send it to each receiving dock, and receivers submit their completed forms directly to the batch processing queue. Documents arrive digitally — no scanning required, no waiting for inter-office mail to deliver the paper from Dock C to the data entry desk. Each uploaded file appears in the same processing queue, ready for batch extraction.

Step 2: Define your extraction columns once

The column definition is the template that gets applied to every document in the batch. Define it once, save it as a preset for future weeks, and reuse it. For a goods receipt batch, the column set typically includes:

The "Condition" column can use inferred extraction: define it as "Condition (options: Good, Damaged, Short, Over, Wrong Item)," and the AI reads the receiver's handwritten notes and classifies the condition automatically. The "Notes" column captures the original comment text — "3 boxes crushed, corner of pallet" — so you have both the structured classification and the raw observation.

Step 3: Process the batch and review flagged fields

Upload all documents. The AI processes each one, extracting values for every defined column and populating one row per document in the output table. Fields with ambiguous handwriting, poor image quality, or missing data get flagged.

The review pass is where batch processing shows its advantage over one-at-a-time typing. Instead of squinting at 30 individual forms, you're looking at a single table where flagged fields are highlighted. You can scan down the "Received Quantity" column and spot-check flagged values against the original forms — or, if you've organized the batch well, against the physical goods on the dock. A value of "197" when every other receipt for that item shows "200" is probably correct — the variance is real. A value of "30" when every other receipt shows "80" and the handwriting on the form shows a closed "8" — that's a misread. The table view makes these patterns visible. Single-document processing hides them.

Step 4: Export and feed into your inventory system

Export as Excel (XLSX). The output is one table where every row is a document and every column is one of the fields you defined. From here, the data feeds into your existing inventory workflow:

QuickBooks / Fishbowl. Map the Excel columns to the QuickBooks or Fishbowl import template. If your column names match the import fields, it's a direct upload. Most warehouse-focused accounting platforms accept CSV or Excel imports for receiving transactions.

Excel-based inventory tracker. If you maintain inventory in Excel, the extracted data appends directly to your master inventory sheet. A VLOOKUP on the item code matches received quantities to the inventory ledger. The batch output structure — one row per receipt — is already in the format most Excel inventory trackers expect.

ERP import. SAP, Oracle NetSuite, and Microsoft Dynamics 365 all support data import via CSV or Excel. The batch output's column headers map to the ERP's receiving transaction fields. The only manual step is adding any internal fields (company code, plant, storage location) that aren't on the physical delivery note — and this can be done as a bulk column fill in Excel before import.

Handling mixed document types in one pass

While processing each document type separately gives the cleanest output, there are situations where mixing types makes sense — for example, a small warehouse that receives 5 delivery notes and 3 dispatch confirmations per day, and wants to process the day's documents as one batch.

The approach: define columns that are broad enough to cover both document types, accepting that some columns will be empty for some documents. A "universal warehouse document" column set might include: Document Type (inferred from the form structure), Reference Number (PO, SO, or DN), Supplier/Customer Name, Date, Item Code, Item Description, Expected Quantity, Actual Quantity, Variance, Notes, and Receiver/Picker ID.

The "Document Type" column uses inferred extraction: "Document Type (options: Goods Receipt, Dispatch, Cycle Count, Return)." The AI reads the form structure and classifies it — a form with "Purchase Order" and "Delivery Note" headers gets classified as "Goods Receipt," while a form with "Sales Order" and "Carrier" fields gets classified as "Dispatch." This classification becomes a filter column in the output Excel, allowing you to separate document types after extraction.

The tradeoff: mixed-type batches produce output where some rows have more empty cells than a single-type batch would. The advantage: you process everything at once rather than setting up three separate batches. For operations processing fewer than 20 documents at a time, the setup overhead of separate batches outweighs the cleaner output. For operations processing 50+ documents of each type, separate batches are worth the overhead.

What to do when handwriting gets messy

Not every document in the batch will extract cleanly. Here's the practical triage for problem documents:

Lightly messy: the standard case. Most warehouse handwriting falls into this category — legible to a human but with enough variability that you have to pause and think. The AI handles these with variable accuracy: simple fields (dates, PO numbers, item codes) extract well; complex fields (handwritten quantities written over printed numbers, cursive condition notes) produce more flags. Review the flagged fields. Move on.

Moderately messy: the speed-writing receiver. One receiver on your team writes fast and sloppy. You know the handwriting. The AI doesn't. Expect 30-40% of fields from this receiver's documents to be flagged. The review pass takes longer, but it's still faster than typing the entire form from scratch — because the AI got 60-70% of fields right and you're only correcting the flagged ones.

Severely messy: the oil-stained carbon copy. A third-copy NCR form, handled with oily gloves, with handwriting that's barely visible. The AI will flag most fields. At this point, the review time may approach or exceed the manual typing time. The pragmatic call: if the document contains critical data (the only record of a high-value receipt), review it in the batch interface. If it's a duplicate or contains data available elsewhere, skip it and handle it manually. The AI's value isn't in solving impossible input — it's in processing the 80% of documents that are legible, so you have time to handle the 20% that need full human attention.

FAQ

How many documents can I process in one batch?

There's no hard limit. Batches of 30-80 documents are practical for most warehouse scenarios. Larger batches (100+) process correctly but take proportionally longer — a batch of 100 documents might take 5-8 minutes to process. The review time is the real limiting factor: reviewing 100 rows of flagged fields takes longer than reviewing 30. For weekly processing, batching by day (Monday's documents, Tuesday's documents, etc.) keeps review batches manageable.

Can I save my column definitions for reuse next week?

Yes. Once you define a column set for goods receipt notes, you can reuse it every time you process that document type. The column definitions are tied to your account, not to a specific batch. Next week's receiving batch uses the same column template — upload the new documents, select the saved column set, and process.

What if the documents in my batch are from different suppliers with different layouts?

The AI reads each document independently, finding the data for each column by understanding what it means rather than where it sits on the page. Supplier A's delivery note with the PO number in the top-right corner and Supplier B's delivery note with the PO number in a table header both produce the correct PO number in the output column. No per-supplier configuration is needed.

Can I batch-process photos taken on different phones by different receivers?

Yes. The AI processes each image independently — different cameras, different lighting, different angles all work within the same batch. The key requirement is that each image captures the full document. Partial images where fields are cut off will produce extraction gaps for those fields.

How do I track which physical documents have been processed?

Stamp each physical document "ENTERED" after uploading, or include a batch ID in your file naming convention. The output Excel includes a row for each uploaded file, so you can cross-reference the Excel against your physical stack and confirm that every document produced a row. Missing rows mean the document wasn't included in the upload batch.