Tax data entry automation can take over the typing of printed boxes on standard forms: W-2 wages and withholding, 1099-INT interest, 1099-DIV dividends, 1098 mortgage interest. What stays with the preparer is judgment: K-1 statements, basis, state allocations and corrected forms. Software reads the page; a person decides what the numbers mean for the return.
Which tax data entry fields software reads reliably
The fields software reads most reliably are printed numbers in fixed, labelled boxes on standard IRS forms. A W-2's Box 1 wages or a 1098's Box 1 mortgage interest sits in the same place on every copy, carries one amount and means one thing. Those fields are the safe part of tax data entry to hand off.
Three things make a field a good candidate. It has a fixed label the IRS defines, so there is no doubt about what the number is. It holds a single value, not a list or a code that changes the meaning. And it can be checked against something else, either another box on the same form or the same box on last year's form.
The W-2 is the clearest case. The IRS General Instructions for Forms W-2 and W-3 (2026) define Box 1 as wages, tips and other compensation, Box 2 as federal income tax withheld, Box 3 as social security wages and Box 4 as social security tax withheld. Each box has one number. Box 4 can even be tested against Box 3, which makes a misread easy to spot.
Interest and dividend forms behave the same way. The IRS instructions for Forms 1099-INT and 1099-OID (01/2024) label Box 1 as interest income and Box 4 as federal income tax withheld. The IRS instructions for Form 1099-DIV (01/2024) label Box 1a total ordinary dividends and Box 1b qualified dividends, and say that Box 1a includes the amount in Box 1b. That relationship gives a built-in check: 1b should never exceed 1a.
How far a firm can trust what the software returns is a separate question from which fields suit it. The post on how accurate OCR is on tax forms explains field-level accuracy and what a confidence score does and does not tell you.
Common forms and what suits automation
The table below lists the fields most firms capture from each form. The "suits automation" column describes how the field is laid out, not a measured rate.
| Form | Fields to capture | How well it suits automation | What the reviewer checks |
|---|---|---|---|
| W-2 | Employer EIN, Boxes 1 to 6, Box 12 codes, Boxes 15 to 17 | High for Boxes 1 to 6; medium for Box 12 code letters | Box 4 against 6.2% of Box 3, Box 6 against 1.45% of Box 5, count of W-2s per person |
| 1099-INT | Payer name and TIN, Boxes 1, 3, 4, 8 | High | Total interest against last year, one form per account |
| 1099-DIV | Boxes 1a, 1b, 2a, 4, 7 | High | 1b not above 1a, foreign tax in Box 7 noted |
| 1099-B | Boxes 1a to 1g, 2, 12 per transaction | Medium; many rows per statement | Row count, totals against the summary page, basis-reported flag |
| 1098 | Boxes 1, 2, 5, 6 | High | Interest against last year, one form per mortgage |
| Schedule K-1 (Form 1065) | Boxes 1, 2, 5, 6a, 19, coded items in 11, 13, 20 | Low for coded items and attached statements | Every code read, every statement page present |
What tax prep data entry still needs a human?
Tax prep data entry still needs a human wherever a number on the page does not translate straight to a line on the return. That covers K-1 codes and attached statements, basis, state allocations, and corrected forms. Software can lift the figures from those pages. It cannot decide what the figures mean for this client.
K-1s are the hardest. The IRS Partner's Instructions for Schedule K-1 (Form 1065) (2025) show that Boxes 11, 13 and 20 each use lettered codes, and that some codes point to a statement attached to the K-1. A reader can capture "20, code A" and an amount. Deciding where that item belongs on the partner's return is preparer work.
Basis is the same story. The same IRS instructions say the capital account in item L comes from the partnership's books and cannot be used to figure the partner's adjusted basis, and that tracking basis is the partner's responsibility. No reading of a single page produces a basis figure. It comes from the firm's own records across years.
State allocations need context the form does not hold. A W-2 can carry two state lines in Boxes 15 to 17, and the W-2 instructions say a second W-2 is needed for more than two states. The wages are printed. Whether they were earned in one state while the employee lived in another is a question for the preparer and the client.
Corrected forms need a decision about which copy wins. A W-2c, titled Corrected Wage and Tax Statements by the IRS, fixes errors on a W-2 already filed. A 1099 with the CORRECTED box checked does the same for that form. Software can flag that a corrected copy exists. A person confirms which figures go on the return.
The tax data entry automation path from upload to export
Tax data entry automation follows the same path for every document: the client uploads a file, software splits it into individual documents, identifies each one, reads the fields, runs checks, and exports the result for the preparer. Each step can fail on its own, so it helps to know what happens at each one.
Upload
The client sends a file through an upload link, an email attachment or a portal download. Quality is decided here. A crooked phone photo or a faint photocopy makes every later step harder. The post on tax source document scanning covers settings and photo habits that keep a file readable.
Split and identify
One upload often holds several documents. A 30-page PDF might contain a W-2, two 1099-INTs and a brokerage statement. Software separates them and labels each by form type, issuer and tax year. A mislabelled page is worse than an unread one, because it lands in the wrong row.
Read and check
Each identified form is read against a schema for that form type: a list of the boxes it should have. Checks run as the fields are filled. Does Box 4 fit Box 3? Is the employer's EIN nine digits in the 00-0000000 format the 2026 W-2 instructions call for? Is this the same W-2 that arrived last week?
Export
The output goes to the preparer as a file. A product can be built to push data straight into a tax program, or to export a spreadsheet the preparer keys from or imports. Schema CPA takes the second route: it reads W-2s, 1099s, K-1s and 1098s, flags duplicates and missing forms, and exports Excel or CSV, without importing into any tax program. The document categories on the homepage list every form type it reads.
The different export paths, and what to test before buying a product that offers each, are covered in the post on scan and populate tax software. For a full buyer's list, see the scan-and-populate evaluation checklist.
How should a reviewer check automated tax data entry?
A reviewer should check automated tax data entry by tying out totals, not by re-reading every box. Compare form counts per type against what was expected, test the arithmetic between related boxes, and compare each figure with last year's. A review that retypes the data has thrown away the time the automation saved.
Count the forms first
Start with counts. If last year's return had two W-2s, three 1099-INTs and a 1098, the export should show the same set or a reason for the difference. A missing form is easier to catch by count than by reading. A doubled form shows up as an extra row from the same issuer.
Test the box arithmetic
Next, run the arithmetic checks that the forms themselves allow. The W-2 instructions give the 2026 social security wage base as $184,500 and say Box 4 should not exceed $11,439, which is that base times 6.2%. Box 4 should be 6.2% of Box 3, and Box 6 should be 1.45% of Box 5. On a 1099-DIV, Box 1b should not be larger than Box 1a.
Here is an example with a fictional client, Dana Okafor, and a fictional employer:
| Field | Read value | Check | Result |
|---|---|---|---|
| Box 3 social security wages | 68,400.00 | ||
| Box 4 social security tax withheld | 4,240.80 | 68,400.00 x 6.2% = 4,240.80 | Pass |
| Box 5 Medicare wages | 68,400.00 | ||
| Box 6 Medicare tax withheld | 991.80 | 68,400.00 x 1.45% = 991.80 | Pass |
If Box 4 had come back as 4,248.80, the check would fail by eight dollars. That points to a single misread digit, and the reviewer opens that one page instead of the whole file.
Compare with last year
Finally, compare each figure with last year's. A 1098 showing mortgage interest twice last year's amount may be right, after a refinance, or may be a misread. Either way it earns a look. Large swings and missing issuers are where review time pays off.
Where to spend the saved time
The time a firm saves on keying is better spent on the items from the previous section: K-1 codes, basis and state lines. The post on the cost of manual data entry in an accounting firm works through how many hours a season's typing takes, which is the budget a review process draws from.
When tax data entry automation is not worth setting up
Tax data entry automation is not worth setting up when a firm's documents do not suit it: very low volume, mostly handwritten forms, or mostly one-off documents with no standard layout. In those cases the setup, testing and review of exports can cost more time than typing the figures by hand.
Volume is the first test. A firm with a few dozen simple returns and one or two W-2s per client may key everything in less time than it takes to upload the files and check the export. The arithmetic changes as the count of forms per return rises, which is why the cost of manual data entry is worth measuring before deciding.
Handwriting is the second. Some forms still arrive filled in by hand, such as W-2s from household employers or 1099s from small payers. Printed boxes read far better than handwritten ones. If most of a firm's pages are handwritten, expect more reads that need a person to confirm them.
One-off documents are the third. A client's closing statement, a foreign pension letter or a custom brokerage summary has no fixed layout. A reader can pull text from it, but each one needs a person to decide which numbers matter. If a firm's clients send more of these than standard forms, the gain is small.
A quick test before committing
Pick ten returns from last season. Count the standard forms, the handwritten pages and the one-off documents. If standard forms make up most of the pile, automation has room to help. If one-off and handwritten pages dominate, keep keying by hand and revisit the question when the client mix changes.
Mistakes that erase the benefit
Two mistakes undo most of the gain. The first is accepting exports without counts or arithmetic checks, so a doubled W-2 slips into the return. The second is reviewing so heavily that every box is re-typed anyway. Set the review to the checks above, and keep judgment items with the preparer from the start.