Claude Skills for Accountants: Isaac Perdomo on Automating Tax Return Reviews

Artificial intelligence is changing how accounting firms handle repetitive work. But one of the most interesting developments is not another standalone accounting application. It is the ability to turn an existing review checklist into a repeatable AI-assisted process.

Consider tax return review.

A preparer finishes a return and sends it to a senior accountant. The reviewer checks identifying information, compares figures, looks for inconsistencies, and determines whether the return is ready for the next stage.

What if an AI assistant could perform an initial review, flag potential errors, and show the accountant exactly where each concern appears?

That is one of the practical applications Isaac Perdomo demonstrated during a recent episode of Growing Your Firm, hosted by David Cristello, founder and CEO of Jetpack Workflow.

Isaac, founder of Opzer, works with accounting firms to identify valuable automation opportunities and implement AI safely.

During the conversation, he demonstrated how Claude Skills can help accountants review tax returns, identify inconsistencies, and organize findings for human review.

The discussion also covered AI browser automation, bookkeeping cleanup, security risks, and why accounting firms should prioritize high-volume processes before investing in new technology.

For accounting firm owners, the takeaway is practical: AI can support quality control, but it works best when firms already have clear procedures, reliable information, and appropriate human oversight.

What Are Claude Skills for Accountants?

Claude Skills are reusable sets of instructions and supporting resources that guide Claude through a particular task.

Instead of explaining the same process every time an employee opens a new conversation, a firm can define how a task should be completed and reuse those instructions.

For accounting firms, this could involve a standardized review checklist, a bookkeeping procedure, or a set of quality-control rules.

During the podcast, Isaac described a Skill as something similar to a software feature. It gives Claude a predefined process rather than leaving every decision to an open-ended prompt.

For example, a tax return review Skill might instruct Claude to:

• Inspect the supplied return for inconsistencies.
• Compare related fields and figures.
• Flag potentially incorrect identifying information.
• Identify amounts that do not appear to reconcile.
• Organize findings by their potential significance.
• Show the relevant information so an accountant can investigate.

The important distinction is that the Skill defines a repeatable review procedure. It does not make Claude a licensed tax professional or guarantee that the return is correct.

A firm still needs qualified accountants to assess findings, verify calculations, apply current tax rules, and approve the completed work.

How Isaac Perdomo Demonstrated AI Tax Return Review

One of the most useful parts of the interview was Isaac’s live demonstration.

He showed David how Claude could review an uploaded tax return using a predefined Skill.

The Skill contained instructions about what to look for in the return. It also directed Claude to create an interactive visual display, called an Artifact, so a human reviewer could inspect the findings.

Instead of receiving only a long text response, the accountant could see flagged issues and the information that led to them.

1. Upload the Tax Return

Isaac began with a tax return document.

In the demonstration, the document was provided to Claude for review.

This illustrates a straightforward workflow: an accountant supplies a document and asks Claude to evaluate it using a defined set of review instructions.

For real client work, however, firms should not upload sensitive tax documents into an unapproved AI environment. Client confidentiality, data retention, vendor agreements, access permissions, and applicable professional obligations must be addressed first.

2. Apply a Predefined Review Skill

Rather than asking Claude to check the return without guidance, Isaac used a Skill containing instructions about what to examine.

This is where existing accounting procedures become particularly valuable.

Many firms already have tax return review checklists covering identifying information, schedules, calculations, and supporting documents.

Isaac recommended adapting those checklists to the firm’s own review requirements.

That means a firm can start with procedures its accountants already understand, then use AI to assist with the initial checks.

3. Identify Potential Errors and Inconsistencies

During the demonstration, Claude identified several issues for further investigation.

One example involved a mismatch between the state code and ZIP code.

The return contained a Georgia state code alongside a Massachusetts ZIP code.

Claude flagged that inconsistency as something the reviewer should investigate.

Another example involved a question with potentially conflicting answers.

Isaac also described the system’s ability to highlight more significant issues, such as figures that do not reconcile or amounts that may be overstated or understated.

These examples show how AI can support document-level quality control.

They do not establish that Claude can reliably identify every tax error or determine the correct tax treatment in every situation.

4. Present Findings in a Claude Artifact

The demonstration went beyond a simple list of warnings.

Isaac used a Claude Artifact to present the review findings in a more accessible format.

David described seeing the tax return alongside Claude’s comments, with highlighted areas drawing attention to potential issues.

This matters because a useful review process should not simply tell an accountant that something may be wrong.

It should help the reviewer locate the relevant information and determine what needs to be checked.

A visual review interface can make the initial quality-control process easier to navigate, particularly when multiple findings require attention.

5. Send the Return for Human Review

The purpose of this process is not to eliminate the senior accountant.

It is to help the preparer or junior reviewer catch potential problems before the return reaches the next stage.

Isaac described AI-assisted review as a way to extend review capacity rather than necessarily reduce review time in every case.

That distinction is important.

A senior reviewer may still need to verify the same critical areas, assess tax positions, and determine whether the return is ready for filing.

The potential benefit is that avoidable inconsistencies can be identified earlier, allowing the senior accountant to concentrate on more complex questions.

AI-Assisted Tax Return Review: A Practical Workflow

The following framework translates Isaac’s demonstration into a possible workflow for an accounting firm. It is an illustrative process, not a verified Opzer implementation checklist.

This approach treats AI as an additional quality-control step, not a replacement for professional judgment.

Why Claude Skills Could Help Junior Accountants

Tax return review is often one of the more demanding parts of a firm’s production process.

Junior accountants may know how to prepare returns but still need guidance identifying inconsistencies, understanding unusual situations, or recognizing when something requires escalation.

Senior accountants provide that guidance, but their time is limited.

Isaac’s demonstration suggests a potential middle step.

Before sending a return to a senior reviewer, a junior accountant could use an approved Claude Skill to perform a preliminary check.

The Skill could identify potential problems and direct the preparer to the relevant fields.

The preparer would then investigate those findings, make appropriate corrections, and document unresolved questions.

This could help firms establish more consistent preparation and review habits.

It could also make review criteria more visible to newer employees.

However, AI-generated findings must be treated as suggestions for investigation. A model can miss errors, misread documents, or incorrectly flag valid information.

The firm should continue to apply its existing review standards.

How to Build a Claude Skill for Tax Return Review

Accounting firms interested in this approach do not need to begin by designing an entirely new review methodology.

A more practical starting point is an existing checklist.

1. Document Your Current Review Process

Start by identifying what your team already checks.

For example:

• Taxpayer and entity identifying information
• Names, addresses, and tax identification numbers
• Consistency across relevant forms and schedules
• Mathematical and financial reconciliations
• Missing or conflicting responses
• Supporting documentation
• Items requiring senior review

These are illustrative review categories. Each firm should use checklists appropriate to the return type, jurisdiction, tax year, and engagement.

2. Define What the AI Should Flag

A useful Skill needs clear instructions.

Rather than telling Claude to find all errors, specify which checks to perform and what evidence should accompany each finding.

For example, the firm might require each potential issue to include the field involved, the observed inconsistency, and the reason human review is needed.

It should also distinguish between a confirmed document mismatch and a possible issue requiring additional context.

3. Require Evidence for Every Finding

AI output becomes more useful when reviewers can trace a finding back to the source document.

A review process should require references to relevant fields, pages, or schedules wherever possible.

The accountant must then verify those references against the original return and supporting records.

This reduces the risk of accepting an unsupported AI conclusion.

4. Test the Skill With Approved Examples

Before introducing AI-assisted review into production, firms should evaluate it using appropriately authorized, synthetic, or de-identified examples.

Include both correctly prepared returns and returns containing known inconsistencies.

Measure whether the Skill identifies expected issues, produces false positives, or misses important problems.

Isaac’s podcast demonstration shows one working example. It does not provide a measured accuracy rate or independent validation across different return types.

5. Establish Human Approval and Security Controls

Tax return review involves confidential financial and personal information.

Before using any AI platform, the firm should assess its data handling practices, access controls, retention settings, contractual protections, and professional responsibilities.

Human reviewers should remain responsible for resolving findings and approving completed returns.

A Skill should not independently submit returns, change client records, or perform other consequential actions without appropriate authorization and controls.

6. Incorporate the Review Into the Firm’s Workflow

Once a Skill has been tested and approved, the firm needs to decide where it fits in the production process.

For example, the workflow might include preparation, preliminary AI-assisted review, issue resolution, senior review, and final approval.

Each stage should have a clear owner.

This is where workflow management for accounting firms matters.

The AI tool can assist with a specific review activity, while a workflow management system can help the firm track assignments, deadlines, review status, and outstanding work.

Where Jetpack Workflow Fits Into AI-Assisted Accounting Processes

Isaac’s discussion highlighted the continuing importance of practice management and workflow systems.

Even as AI becomes capable of completing more individual tasks, accounting firms still need a dependable way to organize client work.

A tax return may pass through several employees before it is ready for delivery or filing.

The firm needs to know who is preparing it, whether preliminary review is complete, which issues remain unresolved, and who is responsible for final approval.

AI can support individual activities within that process, but it does not automatically provide firm-wide operational visibility.

The distinction is between automating a task and managing the complete client workflow.

Both can be valuable, but they solve different operational problems.

Beyond Tax Reviews: Other AI Use Cases Isaac Highlighted

Although the live demonstration focused on tax return review, the interview covered several additional applications.

AI Bookkeeping Cleanup and Transaction Categorization

Isaac described bookkeeping cleanup as an area where his team had seen significant improvements.

He discussed using AI to categorize transactions based on historical statements, available financial data, and a chart of accounts.

According to Isaac, some cleanup projects that previously required extensive manual work could now receive an initial categorization pass much faster.

He described examples where work that might previously have taken months could be approached in an afternoon or less.

That is his reported experience, not a benchmark established by the interview.

The actual time required will depend on document quality, transaction volume, complexity, and the amount of reconciliation and human review needed.

AI-generated categorization should also be verified before entries are posted to a client’s books.

AI Browser Automation for Legacy Accounting Systems

Another area Isaac discussed was browser automation.

Many accounting firms rely on bank portals, payroll platforms, and other legacy systems that do not provide convenient API integrations.

These systems can require employees to sign in manually, navigate pages, and download statements or reports.

Isaac believes improving AI browser capabilities could make some of those workflows easier to automate.

For example, an authorized agent might navigate a portal and retrieve a statement for review.

However, he emphasized that financial systems present significant security risks.

If an AI agent has permission to transfer money, a compromised or misdirected agent could potentially perform actions beyond the intended task.

Isaac recommended restricting permissions and maintaining human supervision when financial transactions are involved.

For banking and payroll workflows, firms should prioritize read-only access where supported, strong authentication, audit logs, and appropriate approval controls.

AI-Assisted Video Editing and Marketing

Isaac also shared an example outside accounting production.

He described giving an approximately 30-minute client interview recording to Claude and requesting shorter versions for marketing.

According to his account, the tool produced one-minute, three-minute, and ten-minute clips with cuts, captions, and branding elements.

He estimated that the process took roughly 60 to 75 minutes.

This example illustrates how AI could support administrative and marketing tasks in addition to accounting work.

It should not be interpreted as a guaranteed processing time or output quality for every recording.

What Accounting Firms Get Wrong About AI Automation

When David asked about common mistakes, Isaac identified a problem that was more operational than technical.

Some firms do not know where to begin.

Others have so many automation ideas that they struggle to choose which ones deserve attention.

His advice was to focus on processes that create the most value.

Rather than starting with whichever AI tool is attracting the most attention, firms should identify the work consuming the most staff time.

That means looking at recurring processes, understanding how employees currently complete them, and identifying opportunities to reduce manual effort or improve quality.

For example, a firm might have opportunities in tax return review, bookkeeping cleanup, document collection, or client follow-up.

The best starting point depends on its workload, risks, and existing procedures.

A Practical AI Prioritization Framework

The following framework builds on Isaac’s recommendation to start with high-volume processes.

For a firm just beginning with AI, a repeatable task with limited consequences and easily verifiable results may be a more suitable pilot than a high-risk process involving unrestricted access to client financial accounts.

Claude Skills vs. Traditional Accounting Automation

Traditional accounting automation often relies on predefined triggers, integrations, and structured rules.

For example, a workflow may create a task when a new client is added or send a reminder when a deadline approaches.

Claude Skills work differently.

They provide reusable instructions that guide an AI model through tasks involving documents, interpretation, or other less structured information.

These approaches are not necessarily competing alternatives.

For accounting firms, the opportunity is to combine clear operational processes with appropriately controlled AI assistance.

A review Skill may help identify issues in a document, while the firm’s workflow system keeps the engagement moving through preparation, review, correction, and completion.

How to Introduce Claude Skills Without Disrupting Your Firm

Accounting firms do not need to redesign every process to begin experimenting with AI.

A controlled pilot is usually a more practical approach.

This is an editorial pilot checklist developed from the operational principles discussed in the interview, not an official Opzer implementation procedure.

A successful pilot should demonstrate not just that the AI can produce an impressive result, but that it improves the firm’s process without introducing unacceptable risks.

Frequently Asked Questions About Claude Skills for Accountants

What are Claude Skills for accountants?

Claude Skills are reusable instructions and resources that guide Claude through defined tasks. Accounting firms can use them to support activities such as preliminary document review, transaction analysis, and other repeatable processes.

Can Claude review tax returns?

Claude can assist with reviewing tax return documents and identifying potential inconsistencies. In the podcast, Isaac demonstrated a Skill that flagged possible errors and presented findings for human inspection. It does not replace a qualified tax professional’s review.

Can Claude Skills identify tax return errors automatically?

They can flag potential issues based on the supplied documents and instructions. However, results may contain false positives or miss important errors. Findings require verification against source documents and applicable tax rules.

Do accounting firms need coding experience to use Claude Skills?

Not every Skill requires an accountant to write software. However, building, testing, and maintaining more complex workflows may require technical assistance. Firms should assess their internal capabilities and security requirements before implementation.

Are Claude Skills safe for confidential accounting information?

Safety depends on the specific AI environment, its configuration, vendor terms, access controls, and the firm’s policies. Accountants should not assume that uploading client tax information to a consumer AI tool is permitted or appropriate.

What other accounting tasks can AI help automate?

Isaac discussed transaction categorization, bookkeeping cleanup, document review, and browser-based workflows. The appropriate level of automation depends on the task’s complexity, data quality, and potential consequences.

Should AI replace senior accountants in the review process?

No. The demonstrated approach uses AI to support preliminary review and identify issues earlier. Qualified accountants remain responsible for professional judgment, corrections, and final approval.

Conclusion: Better AI Starts With Better Accounting Processes

Isaac Perdomo’s demonstration shows why Claude Skills are worth exploring for accounting firms.

The interesting development is not simply that an AI model can read a tax return.

It is that firms can package review instructions into a repeatable process, have AI flag potential inconsistencies, and present findings in a format that accountants can investigate.

That creates an opportunity to support junior staff, introduce additional quality-control checks, and make existing review procedures more consistent.

But the technology does not eliminate the need for experienced accountants, reliable data, or secure processes.

Isaac’s broader advice is particularly relevant: identify your firm’s highest-volume work, understand how it is currently completed, and prioritize the opportunities that offer meaningful business value.

For firms exploring AI-assisted tax review, the best place to begin may be a checklist they already use.

About the guest: Isaac Perdomo is the founder of Opzer, an automation agency helping accounting firms identify and implement AI opportunities. During the episode, he mentioned that Opzer offers a collection of introductory Claude Skills, including a tax return review example.

Last Updated: November 2026

Related Articles

• How to Build AI Agents for Accounting Firms: Isaac Perdomo’s Three-Step Framework
• Claude Cowork for Accountants: Toni Witt on Building Real AI Systems
• AI Automation for Accounting Firms: Preston Cardwell on What to Automate First
• AI Use Cases for Accounting Firms: Jan Haugo on What Firms Should Automate First
• Bookkeeping Workflows: Nancy McClelland on Creating Tax-Ready Books