<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Mullion Studio blog</title>
    <link>https://mullionstudio.com/blog/</link>
    <description>Articles on AI in the recurring processes of accounting, law, insurance and engineering firms.</description>
    <language>en</language>
    <lastBuildDate>Thu, 05 Nov 2026 09:00:00 +0000</lastBuildDate>
    <atom:link href="https://mullionstudio.com/blog/feed.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>Section 7216 and AI tools at CPA firms</title>
      <link>https://mullionstudio.com/blog/section-7216-and-ai-tools-at-cpa-firms/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/section-7216-and-ai-tools-at-cpa-firms/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Accounting firms</category>
      <description>What IRC 7216, the consent regulations and the IRS guidelines of June 2026 mean when a firm puts Copilot or ChatGPT near tax return information. Key points: Under IRC section 7216, a tax return preparer who knowingly or recklessly discloses or uses tax return information without authorization commits a misdemeanor, with a fine of up to $1,000 ($100,000 where section 6713(b) applies), up to one year of imprisonment, or both. Tax return information under Treas. Reg. § 301.7216-1 includes statistical compilations even when they do not identify the taxpayer, so removing names from a file may not take it outside the rule. No IRS primary source we found says whether sending tax return information to a generative AI service falls within the contractor exception in § 301.7216-2(d)(2), so firms should treat it as an open question for counsel. The IRS Office of Professional Responsibility guidelines of June 24, 2026 say practitioners must handle client data only with secure, enterprise-approved AI and must have internal policies for Circular 230 compliance. Where consent is needed, it must be a written document signed and dated by the taxpayer, in the form set by Rev. Proc. 2013-14, with separate documents for use and for disclosure.</description>
    </item>
    <item>
      <title>AI disclosure in engagement letters</title>
      <link>https://mullionstudio.com/blog/ai-disclosure-in-engagement-letters/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/ai-disclosure-in-engagement-letters/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Accounting firms</category>
      <description>What ABA Opinion 512, state bar guidance, the AICPA code and the 7216 rules say about telling clients you use AI, and what a useful paragraph covers. Key points: ABA Formal Opinion 512 says general, boilerplate provisions in engagement letters are not sufficient to obtain a client&#x27;s informed consent to the use of generative AI. ABA Formal Opinion 512 requires informed consent before information relating to a representation is entered into a self-learning generative AI tool. For CPA firms, AICPA 1.150.040 says the client should be informed, preferably in writing, before confidential information goes to a third-party service provider. For tax return information, the section 7216 consent is its own signed document, and a clause in the engagement letter does not replace it. A useful AI paragraph says which tasks use AI, which tools and terms apply, who reviews the output, how fees reflect the work and how the client can opt out.</description>
    </item>
    <item>
      <title>How CPA firms use Microsoft Copilot, process by process</title>
      <link>https://mullionstudio.com/blog/how-cpa-firms-use-microsoft-copilot/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/how-cpa-firms-use-microsoft-copilot/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Accounting firms</category>
      <description>Five recurring processes where Copilot can draft, check or sort, what Microsoft&#x27;s data terms cover, and what stays with the firm. Key points: Microsoft states that Copilot prompts, responses and data accessed through Microsoft Graph are not used to train foundation models. Copilot only surfaces organizational data that the individual user can already view, so SharePoint and OneDrive permissions should be reviewed before rollout. Microsoft&#x27;s data commitments do not replace a CPA firm&#x27;s own duties under AICPA 1.700.040 and section 7216. Five recurring CPA firm processes where Copilot can prepare drafts for a person to approve are new client intake, document follow-ups, engagement letters, repetitive client emails and meeting notes. A first month with Copilot works on one process, measured before and after, with a named reviewer for each step.</description>
    </item>
    <item>
      <title>AI client intake at law firms: ABA Opinion 512 step by step</title>
      <link>https://mullionstudio.com/blog/ai-client-intake-at-law-firms/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/ai-client-intake-at-law-firms/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Law firms</category>
      <description>The intake process of a business law firm, one step at a time, with what ABA Formal Opinion 512 and California&#x27;s 2026 guidance ask at each one. Key points: ABA Formal Opinion 512 requires a client&#x27;s informed consent before information relating to the representation is entered into a self-learning generative AI tool. At a business law firm, AI can summarize the inquiry, structure intake answers, prepare the list of names for the conflict search and draft the engagement letter from a template. The decision to take a matter, the conflict decision and every message that leaves the firm stay with a person. ABA Formal Opinion 512 says managerial lawyers must set clear policies on generative AI, and supervisory lawyers must make reasonable efforts to ensure lawyers and staff comply. California&#x27;s 2026 guidance says agentic AI must not transmit client information outside the firm on its own without appropriate safeguards and human review.</description>
    </item>
    <item>
      <title>AI for insurance agency workflows, with the tools you already have</title>
      <link>https://mullionstudio.com/blog/ai-for-insurance-agency-workflows/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/ai-for-insurance-agency-workflows/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Insurance and engineering</category>
      <description>Renewals, certificates, carrier submissions and client explanations with Microsoft 365 or Google Workspace, and which NAIC rules apply to agencies. Key points: An insurance agency can start using AI in Microsoft 365 or Google Workspace next to its agency management system, which stays the system of record. The NAIC Model Bulletin on the use of AI systems is addressed to insurers and does not mention producers. As of November 3, 2026, the NAIC lists 29 jurisdictions that have adopted the AI Model Bulletin. The NAIC Insurance Data Security Model Law (#668), where a state has adopted it, requires every licensee to keep a written information security program and to exercise due diligence in selecting third-party service providers. Advice on coverage, binding and issuing, and any statement to a client about what a policy covers stay with licensed staff.</description>
    </item>
    <item>
      <title>AI proposal writing at engineering firms, with the tools you have</title>
      <link>https://mullionstudio.com/blog/ai-proposal-writing-at-engineering-firms/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/ai-proposal-writing-at-engineering-firms/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>Insurance and engineering</category>
      <description>Using Copilot or Gemini on past proposals, resumes and project sheets to answer RFPs, and where NSPE&#x27;s Case 24-2 draws the line for sealed work. Key points: AI helps most with the requirements matrix, the search for past projects, tailored resumes and the final check of a proposal against the RFP. The technical approach, team, fees and final approval of a proposal stay with the principal and the project lead. In Case 24-2 the NSPE Board of Ethical Review said AI-generated technical work requires at least the same level of scrutiny as human-created work. Under the NSPE Code, engineers shall not sign plans or documents that were not prepared under their direction and control, whatever tool produced the first draft. NSPE says there are currently no universal guidelines mandating AI disclosure in engineering work, and that ethical principles favor transparency when AI plays a substantial role.</description>
    </item>
    <item>
      <title>What an AI process review covers</title>
      <link>https://mullionstudio.com/blog/what-an-ai-process-review-covers/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/what-an-ai-process-review-covers/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>About the review</category>
      <description>One recurring process, looked at step by step, with a written plan for the tools a firm already pays for. Key points: A Mullion Studio process review looks at one recurring process in a professional firm and shows where AI can take over part of the work. The review is for CPA firms, law firms, insurance brokers and engineering firms with 50 to 300 people. The review runs remotely, with one 90-minute video session on redacted sample documents and no access to the firm&#x27;s systems. About three weeks after the session, the firm receives a written plan with the redesigned steps, the prompts or settings for its current tools and an estimate of the hours freed each month.</description>
    </item>
    <item>
      <title>Choosing the first process to review</title>
      <link>https://mullionstudio.com/blog/choosing-the-first-process-to-review/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/choosing-the-first-process-to-review/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>About the review</category>
      <description>What makes a process a good first candidate, and why files driven by a partner&#x27;s judgment usually are not. Key points: The first process for a Mullion Studio review is chosen in a 20-minute call. A good first process repeats every week, takes hours of skilled people&#x27;s time and is mostly reading, writing or moving information. Typical first candidates are client onboarding, engagement letters or proposals, repeated client questions by email, and meeting notes turned into tasks. Processes where a partner&#x27;s judgment drives every file are usually a weak first choice, because the hours saved are small and the risk is higher.</description>
    </item>
    <item>
      <title>A review without access to your systems</title>
      <link>https://mullionstudio.com/blog/a-review-without-access-to-your-systems/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/a-review-without-access-to-your-systems/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>About the review</category>
      <description>Redacted samples and one recorded session are enough to review a process. Client files stay where they are. Key points: A Mullion Studio process review does not need access to the firm&#x27;s systems or client files. The review works from a handful of sample documents that the firm has redacted and from one recorded video session. Every recommendation has to work inside the firm&#x27;s existing tools and data protection settings. When a step touches confidential information, the report says so and explains what AI should leave alone.</description>
    </item>
    <item>
      <title>What the review report contains</title>
      <link>https://mullionstudio.com/blog/what-the-review-report-contains/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/what-the-review-report-contains/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>About the review</category>
      <description>Five parts: today&#x27;s process, the redesigned map, the risks, the prompts or settings, and the hours freed. Key points: The report of a Mullion Studio process review has five parts: today&#x27;s process map, the redesigned map, the risks of each AI step, the prompts or settings, and an estimate of the hours freed each month. Each redesigned step is marked as drafted by AI, checked by AI, sorted by AI, or kept with the firm&#x27;s team. A 45-minute walkthrough with the team follows the report. After the review, the firm&#x27;s team can apply the plan on its own.</description>
    </item>
    <item>
      <title>How much time a review takes from your team</title>
      <link>https://mullionstudio.com/blog/how-much-time-a-review-takes/</link>
      <guid isPermaLink="true">https://mullionstudio.com/blog/how-much-time-a-review-takes/</guid>
      <pubDate>Thu, 05 Nov 2026 09:00:00 +0000</pubDate>
      <author>mattia@mullionstudio.com (Mattia)</author>
      <category>About the review</category>
      <description>About three hours across a call, a video session and a walkthrough, plus a few redacted sample documents. Key points: A Mullion Studio process review takes about three hours of the firm&#x27;s time. The three hours are a 20-minute call, a 90-minute video session and a 45-minute walkthrough, plus the time to collect a few redacted sample documents. The report arrives about three weeks after the video session.</description>
    </item>
  </channel>
</rss>
