Skip to content
(214) 444-7884
ITECS

Custom AI Agents for Accounting and CPA Firms

Accounting AI from the ITECS Dallas team

ITECS builds governed AI for regional and local CPA firms of 50–1,000 people. A tax return readiness agent checks each client's documents against last year's return, tracks expected K-1s, drafts missing-items requests for staff approval, and hands preparers a source-linked packet, while people keep each determination, signature, and filing.

  • Capacity pressure
  • Readiness pressure
  • Trust pressure

Firms already pay for extraction inside their tax software. What busy season lacks is a governed layer across the portal, document system, workflow tool, and tax package that knows which returns are truly ready, which are blocked and why, and who needs to hear from the client.

Accounting signal map

Tax + audit + trust

Tax

Intake, readiness, and K-1 tracking

Audit

PBC requests and evidence matching

Trust

Section 7216, WISP, and reviewer sign-off

Busy-season IT behind accounting AI

Accounting AI from a team that already plans Dallas CPA firms' busy season.

ITECS supports Dallas accounting firms with tax-season readiness, Microsoft 365 governance, secure client document workflows, and backup planning for partners, staff, and seasonal users. ITECS AI keeps the same boundary: it does not replace your tax software publisher, your compliance counsel, or the professionals who sign. Read about ITECS IT services for accounting firms.

  • June 24, 2026

    The IRS Office of Professional Responsibility issued its first guidelines for responsible AI use in tax practice, tied to Circular 230 duties.

    IRS OPR Issue 2026-19
  • 65%

    Share of AICPA members who prepare returns that use AI for tax research; 32% use it for client communication.

    The Tax Adviser, Aug. 2026
  • Aug. 2026

    The Texas board told CPAs that professional responsibility cannot be delegated to software and that AI-generated analyses are not audit evidence.

    Texas State Board Report

Operating pressure

Accounting AI Starts With Capacity, Readiness, and Trust

The first step is not picking a model. It is finding where preparers wait on incomplete returns, where managers chase status by email, and where client data needs firmer boundaries before AI touches it.

Capacity pressure

A thinner talent pipeline and steady deadline volume leave less room for preparers to sort documents and chase clients.

  • Preparers starting returns that turn out to be incomplete
  • Status chased through email and spreadsheets
  • Seasonal staff onboarded during the busiest weeks
  • Reviewers buried in rework from missing items

Readiness pressure

K-1s arrive late or corrected, brokerage statements are reissued, and organizers come back half answered.

  • Expected K-1s tracked from memory
  • Corrected 1099s that arrive after preparation starts
  • Generic reminder blasts instead of specific requests
  • Extension decisions made in the final days

Trust pressure

AI is only usable when return information stays inside Section 7216, the firm's written security plan, and professional review.

  • Consumer AI accounts outside the firm's written security plan
  • Third-party AI vendors not yet vetted or contracted
  • Client consent and engagement-letter disclosure questions
  • AI output that must be verified before it reaches a client or the IRS

Example busy-season signal board

The Metrics an Accounting AI System Should Make Visible

A tax readiness program should show the managing partner which returns are ready, which are blocked and why, and how many extensions are coming. These figures are illustrative for a fictional 140-person CPA firm on March 30 of busy season; discovery replaces them with the firm's own data.

Readiness bridge

Returns blocked by cause

RiskRecoverable
Expected K-1 not received214 returns
Brokerage 1099 pending or corrected131 returns
Organizer unanswered or incomplete102 returns
Prior-year source document missing89 returns
Unreadable or duplicate uploads52 returns

Returns ready for preparation

1,512

63% of 2,400 individual engagements, packets built

Returns blocked on client items

588

24.5% of engagements, broken down by cause

Client follow-ups awaiting approval

146

Drafted by the agent, not yet sent

Projected extensions

410

Mostly late K-1s and corrected brokerage statements

What leadership sees

  • Ties each blocked return to its cause and the client request that clears it
  • Separates returns likely to clear before the deadline from likely extensions
  • Routes client messages and readiness releases for staff approval

Accounting AI use cases

Where AI Creates Measurable Value for CPA Firms

Start where the documents already arrive and a professional already approves the result: tax readiness, audit requests, close work, and client communication.

CAS Month-End Close Assistant

Surface uncategorized transactions, unreconciled items, and unusual variances, and draft client questions and proposed entries for accountant approval.

A standard close with exceptions queued

Bookkeeping & Bank Reconciliation

Categorize bank and card activity under approved rules, enter vendors and bills, flag likely owner draws, and reconcile on the client's schedule, with open items routed by email.

Books current between closes

K-1 Tracking & Extension Forecasting

Build the expected K-1 list from last year, log arrivals and corrected copies, and project which returns will need extensions.

Extension calls made weeks before the deadline

Audit PBC & Evidence Requests

Roll forward last year's PBC list, track open requests, and match received documents to selections while auditors perform and conclude the procedures.

Less request administration for seniors

IRS Notice Triage & Response Drafts

Classify the notice and its deadline, pull the related return and workpapers, and draft a response for an EA or CPA to review.

Each notice logged with its deadline on day one

Engagement Letters & Consent Forms

Draft engagement letters from firm templates with the firm's third-party disclosure language and any consent form counsel has approved.

Consistent letters with required language intact

Tax Return Readiness & Intake

Classify portal and email uploads, compare them with last year's return and the organizer, and mark each return ready, blocked with a reason, or not started.

Preparers start only complete returns

Client Missing-Items Requests

Draft specific, plain-language requests that list exactly what is missing, held for staff approval before anything goes out.

Fewer generic reminders, one auditable record

Tax Research With Citation Checks

Summarize authorities and draft memos as desktop agents, with each citation checked against the primary source before use.

Faster first drafts, mandatory cite checks

Firm Knowledge Agent

Answer staff questions from the firm's procedures manual, review standards, written security plan, and software how-tos, with links to the source.

Consistent answers for seasonal staff

Flagship enterprise agent

Tax Return Readiness Agent: Each Return Marked Ready or Blocked

An illustrative solution design, not a reported client result. It checks each client's documents against last year's return and the organizer, tracks expected K-1s, drafts the missing-items request for staff approval, and builds the source-linked packet a preparer starts from. People decide, prepare, review, sign, and file.

Reads portal uploads, last year's return and organizer, the workflow tool, and existing extraction output. Never files or signs a return, writes into the tax software on its own, or makes a tax determination.

See the full tax return readiness design

Example agents

Example agents for Accounting and CPA Firms

Each is a starting design. An AI Blueprint adapts it to your goals and the systems it will connect to.

  • Tax Return Readiness Agent

    Checks each client's uploads against last year's return and the organizer, tracks expected K-1s, and drafts the missing-items request for staff approval before a preparer starts.

    • Accounting
    • Front office
    • Hosted AI Agents
    • Private AI Servers
  • CAS Month-End Close Agent

    Works the firm's close checklist for each client accounting services client, surfaces uncategorized transactions, unreconciled items, and unusual variances, and drafts client questions and proposed entries for the accountant to approve.

    • Accounting
    • Finance
    • Hosted AI Agents
    • Local AI Agents
  • Engagement Letter & Proposal Agent

    Drafts the engagement letter and proposal from firm templates, carrying the firm's third-party disclosure language and any consent form counsel has approved, for the engagement partner to finish and sign.

    • Accounting
    • Sales
    • Local AI Agents
    • Hosted AI Agents
  • Bookkeeping & Reconciliation Agent

    Keeps the books current between closes. Categorizes bank and card activity under approved rules, enters vendors, drafts bills from vendor invoices, flags likely owner draws, and reconciles each account on the schedule the client sets.

    • Accounting
    • Finance
    • Hosted AI Agents
    • Local AI Agents
See all 8 designs for Accounting and CPA Firms

Deployment patterns

Three ways to run an agent

Each pattern keeps your data in a different place and suits different work.

Local AI Agents

Where it runs
The employee's own Claude, Codex, or Copilot seat. No server.
Who evaluates it
Department heads and the employees who will build
What it connects to
Files and systems that employee can already reach
Typical first engagement
A Guided Build Sprint or a Local Agent Sprint

Hosted AI Agents

Where it runs
A governed Linux environment ITECS operates, on frontier models
Who evaluates it
COO, CFO, and operations leadership, with IT reviewing
What it connects to
ERP, CRM, finance, and document systems through controlled integrations
Typical first engagement
An AI Blueprint, credited toward the build

Private AI Servers

Where it runs
A server with data-center class GPUs inside your network, on local models
Who evaluates it
CFO, CIO, IT director, and regulated buyers
What it connects to
The same systems, with the data staying on premises
Typical first engagement
An AI Blueprint and a sizing call for the server

For firms, the tax readiness design runs hosted, under the US-only processing, no-training terms, and written Section 7216 confidentiality terms its controls require, and can move to a private server when counsel or the firm's written security plan keeps return information on premises. Close and engagement-letter work can start on one person's own AI seat as a local agent, then move hosted when it needs multi-user access or an audit trail.

Systems accounting agents read

  • UltraTax CS
  • CCH Axcess
  • CCH ProSystem fx
  • Lacerte
  • SurePrep
  • SmartVault
  • ShareFile
  • Karbon
  • XCM
  • Suralink
  • Caseware
  • QuickBooks Online
  • Xero
  • Sage
  • Microsoft 365

Technologies ITECS evaluates and integrates through supported import, export, or API methods confirmed in discovery; listing a product does not imply a formal partnership or a native connector.

Training

Training that fits

Train the people who will run it: leadership first, then the teams and employees who build.

  1. LeadershipExecutive AI Literacy BriefingHalf-day. A private, tailored session for your board, leadership team, or department heads. We translate AI from buzzword to business reality, calibrated to your industry.$4,500 / session
  2. TeamsTeam AI Workshop2 hours · up to 12 employees. A hands-on, role-based group session that helps a team use approved AI tools safely and productively in its real work.$2,900 / session
  3. BuildersGuided Builds90-minute co-build sessions. ITECS architects and coaches while your employee drives the build in Claude Cowork, Codex, Copilot agents, or a comparable workspace.From $650

Prices from the published AI pricing catalog.

Start here

Accounting AI Readiness Assessment

ITECS starts by mapping the business case, client-data flows, vendor terms, and governance model before recommending an agent build.

1

Map busy-season priorities

Identify where returns stall, where staff chase status, and which deadlines drive extensions and rework.

2

Trace client data and your WISP

Document where return information lives, which tasks need it, and how each AI service fits Section 7216 and the firm's written security plan.

3

Rank use cases by economics

Prioritize by returns cleared, reviewer hours returned, extensions avoided, and feasibility, starting where documents already arrive.

4

Write the governed path

Document the AI policy, approval points, the consent position from counsel, audit logging, and vendor terms before any production build.

Governance

Built Around Section 7216 and Professional Review

ITECS designs accounting AI around the rules that govern return information and the professionals who stay responsible for each return.

Return information boundaries

Section 7216 limits how preparers use and disclose return information. Whether an AI vendor fits the auxiliary-services exception is unsettled, so the design minimizes exposure and counsel decides consent.

  • US-only processing and storage, with offshore access excluded
  • Enterprise terms that exclude model training on firm data
  • Written Section 6713 and 7216 confidentiality terms with ITECS and subprocessors
  • A consent workflow in the IRS format if counsel requires one

Professional responsibility

IRS guidance under Circular 230 and the Texas board both treat AI output as a draft that a qualified professional verifies.

  • Preparers, reviewers, and partners keep each judgment and signature
  • The agent recorded as an overseen service provider in the firm's WISP
  • Independence review before any agent touches an attest client's systems
  • SOC 2 Type II applies only if Promus managed cloud hosting is selected

Security

Security for Accounting AI Workflows

Accounting AI touches tax return information, payroll files, and client financial records. ITECS designs these systems with MFA, activity logging, retention rules, and staff approval before anything reaches a client or the IRS.

Multi-factor authentication and user-activity logging, as the FTC Safeguards Rule requires
Each extracted value linked to its source page and reviewer
Retention and disposal set by the firm's written security plan
Staff approval before client messages or readiness releases

FAQ

Accounting AI FAQ

Does using AI on client tax data violate Section 7216?

Section 7216 bars preparers from using or disclosing return information without written consent unless an exception applies, such as a US provider of auxiliary services that makes no substantive determinations. Whether an AI vendor fits is unsettled, so ITECS designs for US-only processing, no training on your data, and written confidentiality terms, and your counsel makes the consent call.

Will the agent prepare, sign, or file returns?

No. It classifies documents, checks completeness against last year, drafts follow-ups for staff approval, and hands preparers a source-linked packet; preparers, reviewers, and partners keep each judgment and signature.

How does this fit our WISP and the FTC Safeguards Rule?

Tax preparers are financial institutions under the Safeguards Rule, so the agent is documented as an overseen service provider in your WISP, with MFA, activity logging, retention rules, and incident response coverage. IRS Publication 5708 is the template many firms use.

We already pay for SurePrep or CCH Axcess Scan. Why add this?

Those tools extract document data into your tax software. The readiness agent works around them, using their output to decide which returns are complete, track K-1s, draft requests, and update your workflow tool, and discovery maps what your licenses already cover.

Can we use it on audit or CAS clients?

Yes, with guardrails. An audit agent can roll forward PBC lists and match documents to selections, but the Texas board reminds firms that AI-generated analyses are not audit evidence, and your independence partner signs off before any agent touches an attest client's systems.

Start with an AI Blueprint

The AI Blueprint reviews your workflow, data sources and integrations, specifies the architecture and the human-review model, and ends with a fixed-fee not-to-exceed for the build.