
Getting tax documents in the door is only the first step. A W-2, 1099, or K-1 can be sitting in the right engagement and still require someone to identify the source, pull the right fields, and get that information into a form the preparer can review and use.
Tax document extraction software for CPAs is designed to handle part of that work by turning information inside source documents into structured tax data. Soraban uses that process to reduce repetitive reading and re-keying while keeping review, corrections, and professional judgment with the firm.
The distinction matters because OCR (optical character recognition), document recognition, extraction, validation, workpaper preparation, and tax software data entry solve different jobs. Understanding where each one starts and stops makes it easier to see what actually changes in the firm's process.
Extraction gives the information inside a source file structure. A wage amount, withholding figure, payer name, or other value stops being text buried in a PDF and becomes a defined field with meaning.
Before that can happen, the system has to read the page and identify what kind of tax form it's processing.
OCR reads letters, numbers, and other characters on a page. It turns text inside a scan or image into machine-readable information.
Document recognition answers a different question: What am I looking at?
If the system recognizes a file as a W-2, 1099, K-1, or another supported form, it knows which fields to expect. OCR identifies the text on the page. Recognition identifies the type of source that text came from.
Once the source type is known, the system can identify specific fields and associate values with defined labels.
A useful shorthand is text → document type → field → value.
Reading “42,000” from a page is only part of the work. The number becomes useful when the system knows whether it represents wages, withholding, interest income, or another field. That context is what makes the value usable downstream.
Before field-level processing begins, the file has to be tied to the right client and engagement, identified correctly, and organized for the work that follows.
Accounting firms already receive files through workable channels. A client may use a portal, send a scan, take a phone photo, or provide several files at once. The intake method isn't the issue. What matters is what still has to happen after the file arrives.
During busy season, partial and imperfect submissions are normal. Even a clean electronic upload may require someone to confirm the client, year, source type, and place in the engagement before a preparer can use it. A late file may also need to be matched to work that has already started rather than treated like a brand-new submission.
Soraban Collect handles that upstream work. It manages intake, tracks missing items, and handles file classification, naming, matching, and organization. That keeps document readiness separate from reading and structuring the tax data inside the source material.
Input quality varies too. A scan may be skewed, a page may be cropped, several forms may arrive inside one PDF, or pages may be missing. Catching those issues first keeps basic file problems from becoming field-level problems later.
Once the file is ready, the next question is straightforward: Does the system know what it's looking at?
Tools built for tax work may recognize common tax forms such as W-2s, 1099s, and K-1s. Recognition gives the system the context it needs to locate the fields associated with that form.
Coverage still has limits. Less-common forms, unfamiliar layouts, or files the system cannot confidently classify need a clear exception path instead of being treated like ordinary inputs.
This is a document-level issue. The concern isn't yet whether an individual value is correct. It is whether the file itself can enter the standard processing path.
That makes supported-form lists only part of the evaluation. Firms should also ask what happens when recognition fails, how uncertain files are surfaced, and what staff need to do before the file can continue. A long coverage list matters less if unrecognized files disappear into a queue that no one knows how to work.
After recognition, the next job is giving individual values the right context.
A W-2 can contain several amounts that look similar on the page but mean different things. The system has to distinguish wages from Social Security wages, federal withholding, and other fields. The same applies to payer names, interest amounts, dividend figures, and supported K-1 values.
A value becomes usable when it is tied to the right field and given a defined destination. That is what separates raw text capture from usable tax data. The firm can then review the information and decide what's ready for the next stage.
Identifying a field doesn't automatically make the value ready to proceed. The information still has to be checked so the team can separate routine data from items that need attention.
The sequence is extraction → validation → exception → correction → review → approval.
Validation can check whether taxpayer information, tax year, source context, and expected fields line up. If a value is missing, conflicts with other information, or can't be read with enough confidence, it should enter an exception path instead of moving forward as if nothing is wrong.
Accuracy matters, but a headline percentage doesn't explain what happens in the difficult cases. Firms also need to know how uncertain values are surfaced, who can correct them, and what has to happen before the information is approved.
Good exception handling keeps those fields visible so staff can focus on what actually requires attention without treating every captured value as equally uncertain.
Corrections should remain clear after they're made. If someone changes a captured amount, the next person should be able to see the corrected value in context, understand what changed, and know whether another review step is still required.
Review gets harder when a number is separated from the source that produced it. Source traceability lets a preparer or reviewer connect a value back to the supporting file and see the context around it.
That matters when a field looks unusual, needs correction, or simply needs confirmation. Instead of searching through folders or reopening several files, the reviewer can return to the source and understand where the value came from.
Traceability also creates a clearer record of questions and changes made during prep and review. Soraban Prepare keeps that source support tied to the work the preparer and reviewer are inspecting, so source lookup doesn't become another separate task.
That context becomes especially useful when work changes hands. The reviewer should be able to follow the support without asking the preparer to reconstruct how the number got there.
Structured tax data isn't the same as a completed workpaper. A reviewed value may be ready to move downstream while the preparer still needs an organized record of the work behind the return.
That distinction separates Soraban Connect from Prepare. Connect handles form recognition, extraction, mapping, validation, review, and movement into tax software. Prepare is Soraban's Workpaper Management System, where the engagement is organized for prep and review.
A binder is the engagement file that brings supporting material together. A leadsheet is the summary or control layer for a section. A workpaper provides the detailed support behind a number, treatment, or conclusion.
Prepare builds and populates those materials, drafts reviewer notes, keeps source support available, and preserves accountant annotations and edits. Structured data can feed that work, but it doesn't replace the file structure that preparers and reviewers rely on.
Keeping those jobs separate also makes software evaluation easier. A tool that captures fields well may still leave workpaper setup untouched, while a strong workpaper system may not remove tax-software data entry.
Once a value has been recognized, structured, and reviewed, it still needs the right destination. Field mapping connects that value to the appropriate field in the firm's tax software.
Mapping tells the system where information belongs. It doesn't mean the value has been approved. The firm still needs a chance to inspect questionable information and decide what's ready for export.
Soraban Connect handles that handoff. It supports reviewed tax data moving into UltraTax, CCH Axcess, Lacerte, and Drake, reducing repetitive keystroking when staff would otherwise re-enter values already present in source files.
The value of that step is straightforward: staff spend less time typing information the firm already has, while approval remains part of the process before the data reaches the return.
Automation works well for repeatable tasks with known rules and destinations. Recognizing supported forms, identifying fields, mapping values, surfacing differences, and reducing repetitive entry all fit that category.
Professional judgment does not.
Preparers and reviewers still need to resolve ambiguous facts, assess exceptions, decide how unusual items should be treated, and determine when the work can continue. A system can surface a difference or uncertainty, but it can't take responsibility for the tax judgment behind the final return.
That boundary matters during busy season. Automation can take on the repetitive handling around professional decisions, not the decisions themselves.
For a firm evaluating this category, the better question isn't how much the technology can do alone. It is how much predictable work it can remove while keeping the right decisions with preparers and reviewers.
Taxpayer information is sensitive, so security belongs beside accuracy, reviewer control, and process fit in the buying decision.
Firms should understand who can access client data, how authentication works, how information is encrypted in transit and at rest, what's logged, how long data is retained, and how third-party AI providers may use customer information. Those controls can contribute to compliance practices, but they don't make the firm automatically compliant.
Soraban's relevant controls include SOC 2 Type II, MFA, encryption in transit and at rest, logging, audit trails, and data-retention controls. Client data isn't used to train third-party commercial AI models.
The useful test is specificity. A vendor should be able to explain what happens to sensitive information while it's processed, reviewed, stored, retained, and removed under the firm's data-handling requirements.
Accuracy alone doesn't tell you if a tool will fit. A product can capture fields well and still create extra work if staff have to fix unclear mappings, search for source files, re-key information, or manage exceptions outside the normal process.
A useful pilot should look like a real busy season, not a demo folder filled with clean examples. Test common forms, less-common files, poor-quality inputs, corrections, and expected exceptions with the people who would actually handle them. Use enough variety to expose the handoffs that a polished demo can hide.
One question is especially revealing: Who owns each type of exception? An admin may handle a file issue, while a preparer or reviewer handles a question about the captured information. That ownership should be clear before a broader rollout.
Also test tax software compatibility, source traceability, correction handling, reviewer controls, workpaper implications, security, and manual handoffs around the tool. Pay attention to where staff leaves the normal process to resolve an issue, because that work is easy to miss in a feature comparison.
The pilot should help the firm learn where work actually disappears, where it moves, and where staff still have to step in. That tells you more than a handful of ideal files ever will.
Data capture works best when the work around it stays connected. Soraban is the execution layer for that flow. Practice management organizes work. Tax software calculates returns. Soraban moves work through the firm.
The four products have distinct roles:
Connect is the primary product for data capture, but the surrounding work still matters. Organized source material gives it a cleaner input, while review and source context help determine what's ready for tax software entry.
That places extraction inside the larger flow across intake, prep, data entry, and delivery rather than treating it as an isolated OCR function.
After a pilot or implementation, the question changes. Instead of asking what the tool says it can automate, look at the work your team still performs.
Are staff still re-keying the same information? How often do captured values need correction? Are exceptions easy to identify and resolve? Can a reviewer return to the source without searching through folders? Does approved information reach the next stage without another manual handoff?
Accuracy is one signal, but it belongs beside those operational measures. A strong percentage means less if preparers still rebuild information or reviewers still have to hunt for context.
Start with a simple baseline from the current process. Note where staff re-enter information, how corrections are handled, where source lookup slows review, and which handoffs regularly require another person to step in. Then compare the same points after the new process is in use.
The comparison should show whether the work actually changed, not just whether the software produced an output. If one manual step disappears but two new checks show up elsewhere, that tradeoff belongs in the evaluation too.
If preparers can keep moving with less rework, reviewers can resolve questions with the source close at hand, and fewer handoffs depend on repetitive entry, the process is doing more of the work it was meant to do.
Software can identify fields such as taxpayer and payer information, wages, withholding, interest, dividends, and supported K-1 values. Coverage varies by form and layout, so firms should test the software against the source documents they actually receive.
Accuracy should be considered alongside exception handling, corrections, reviewer controls, and source visibility. A percentage alone doesn't show how the software handles uncertain or incorrect fields, which is often where the firm still has work to do.
A difficult file should be surfaced for attention rather than treated like standard input. Poor image quality, missing pages, unusual layouts, or unclear text may require staff to correct the issue before processing can continue.
Yes. Staff should be able to review and correct questionable information before it moves into tax software. Firms should also understand how corrections remain visible and what approval happens before the data moves forward.
Unsupported files should follow a defined exception or manual path. They should remain visible to the team rather than being skipped or pushed through a process that was not designed for that source type.
Yes, if the product supports the firm's tax software and the relevant fields can be mapped. Staff should still be able to review and approve the information before it moves into the return.
Yes. Practice management and data extraction solve different parts of the workflow. Firms can keep the systems that organize engagements while adding technology that handles tax data between source documents and downstream tax software.
Ask whether staff can connect a captured value back to its supporting file and see the surrounding context during prep and review. The reviewer shouldn't have to search through separate folders just to understand where a number came from.
Ask about access controls, authentication, encryption, logging, data retention, and how client information is used by AI providers. Those controls can contribute to security and compliance practices, but they don't make the firm automatically compliant.
CPAs should test the software with the kinds of files their teams actually handle. Include common forms, less-common documents, poor-quality scans, corrections, exceptions, source tracing, reviewer controls, and downstream data movement rathe
Getting text off a tax document is only the beginning. Before that information is useful, the system has to recognize the source, give individual values structure, surface exceptions, preserve context for review, and move approved data to the right place.
Those stages solve different problems. Technology can take repetitive handling out of the process, but preparers and reviewers still decide what's correct, what requires judgment, and what belongs in the return.
Soraban connects that work across Collect, Prepare, Connect, and Deliver while working alongside the systems your firm already uses.
Want to know what that looks like in your firm's tax workflow? Request a demo to see how Soraban works.
Under 30 days to your first live season. No migration. No commitment until you see it working.