A 22-person CPA firm with offices in the South Hills and a satellite in Cranberry Township came to us in early fall. Two senior accountants had retired the prior spring, the partners couldn't backfill at the rate the local market was asking, and the firm was looking down the barrel of another 1040 and 1120 season with fewer hands. The managing partner had been reading about generative AI and wanted a serious answer to one question: where can this actually shave hours off our work without putting client data or our license at risk?
That's the conversation behind most of our AI workflows for CPA firms engagements. The partners are not looking for a chatbot demo. They want fewer late nights in March, faster turnaround on review, and a defensible story for clients and regulators about how data is handled.
The challenge: a 22-person CPA firm staring down another tax season
The firm's pain points were specific and measurable. Source-document intake was almost entirely manual — clients emailed PDFs, dropped paper at the front desk, or uploaded to a portal that nobody loved. Staff accountants spent the first 30 to 60 minutes of every 1040 keying numbers from W-2s, 1099s, K-1s, and brokerage statements into the tax software. Review notes bounced back and forth in email. Engagement letters and PBC lists were rebuilt from last year's Word docs every January.
On top of that, the firm did audit and attest work for a handful of municipalities and a regional credit union, which meant any AI rollout had to respect IRS Publication 4557 safeguards, the FTC Safeguards Rule, AICPA confidentiality standards, and — for the credit union work — the realities of GLBA. "Just paste it into ChatGPT" was not an option.
The partners were not looking for a chatbot demo; they wanted fewer late nights in March and a defensible story about how client data is handled.
How it was solved: building AI workflows for CPA firms around real engagements
We started with a two-week discovery focused on the engagements that consumed the most hours: individual returns, monthly close for write-up clients, and the audit PBC process. Rather than buying a single AI product and looking for places to use it, we mapped the firm's actual workflow and identified eight points where AI could remove typing, summarization, or lookup work — then ranked those by risk and ROI.
TL;DR: The win is not "add AI to the firm" — it is identifying the three or four steps in a tax or audit workflow where AI removes typing and lookup, then wrapping those steps in controls a CPA can sign off on.
The architecture used Microsoft 365 as the data boundary, with Copilot licensed for the partners and managers and a tenant-scoped Azure OpenAI deployment for document extraction. Client documents stayed inside SharePoint and a dedicated Azure resource group in the East US 2 region. No client data was used to train external models, and we documented that in writing for the firm's WISP and for client-facing FAQs.
What was included in the rollout
The deliverables were concrete:
Document intake automation that read incoming W-2s, 1099-NEC/MISC/INT/DIV/B, K-1s, and Consolidated 1099s, extracted the fields, and produced a reviewer-ready worksheet alongside the original PDF. A human preparer still signed off on every return — the AI produced a first pass, not a filing.
A private firm assistant built on the firm's own document library: prior-year workpapers, the firm's tax memo archive, and AICPA guidance the partners had licensed. Staff could ask "how did we handle the Section 174 capitalization for the manufacturing client last year" and get a cited answer pointing to the actual workpaper.
Audit PBC and engagement letter generation from prior-year templates, with variances flagged for partner review. Monthly-close email drafts for write-up clients, generated from the trial balance and reviewed before sending.
We also delivered written acceptable-use guidance, a one-page client disclosure the firm could share, MFA and conditional access tightening, and four hours of role-based training split between partners, seniors, and admin staff.
Outcomes after the first quarter
By the end of the following April, the partners reported that preparer time on straightforward 1040s had dropped meaningfully — enough that two staff accountants finished the season without the Saturday rotation they had worked the prior three years. Review notes were shorter because the first-pass workpapers were more consistent. The firm did not need to hire the contract preparer they had budgeted for. None of the AI features touched a return that wasn't reviewed and signed by a credentialed preparer, and the firm passed its peer review without a finding tied to the new tooling.
We are deliberately not publishing a percentage figure here. Every firm's mix of returns and starting baseline is different, and the honest answer is that the gain shows up most on high-volume, document-heavy work and least on complex advisory engagements.
Why CPA firms in the Pittsburgh region work with PGH Networks
PGH Networks has supported professional services firms across Pittsburgh, Cranberry, Wexford, Monroeville, Robinson, Washington, and the Mon Valley for years. Our team is on-site within the 75-mile radius of 15220 when an engagement calls for it, and we hold the compliance posture — SOC 2 awareness, HIPAA for the medical-adjacent CPA work, CMMC for firms serving DoD contractors in the region — that an accounting firm's clients increasingly ask about during vendor reviews.
What separates this work from generic IT support is the vertical AI practice behind it. We are not reselling a single product. We design AI workflows for CPA firms around the engagements you actually run, inside the Microsoft 365 tenant you already pay for, with the controls your peer reviewer and your insurance carrier expect to see.
Takeaway and next step
If your firm is heading into another season understaffed, or your partners are nervous about staff using consumer AI tools without guardrails, the path forward is not a product demo. It is a short, focused review of where AI workflows fit your engagements and where they do not.
We offer a fixed-scope AI readiness assessment for CPA firms in the Pittsburgh metro: two weeks, a written workflow map, a risk register, and a phased rollout plan you can take to your partners' meeting. Call PGH Networks at the number on our contact page, or send a note describing your firm size and software stack (UltraTax, Lacerte, CCH Axcess, Drake, QuickBooks, Sage Intacct — we have worked with all of them) and we will get back to you within one business day.