The Nexairi Dispatch · Friday, May 15, 2026 · Issue #15
39% more errors when finance teams run 4+ agents
BCG measured what happens when finance teams run 4+ AI agents at once. The error rate is 39% higher.
Good morning, friends. Xero and QuickBooks are both moving AI inside live client files, and most CPA firms haven't built a gate yet. BCG found finance teams running four or more AI agents at once make 39 percent more major errors. And most CFOs still can't tell their board what AI is actually worth. Have a productive Friday.
🔒 CLIENT DATA — Xero and QuickBooks Are Now Inside Your Clients' Files
What happened: Xero and QuickBooks have both added AI features that touch live financial, payroll and employee records. CPA firms using these platforms may already have vendor AI reading client files without any formal approval process. The Nexairi piece covers exactly where each platform's AI enters the data flow and what a proper approval gate needs to include.
Why it matters: Most engagement letters and data privacy policies predate AI subprocessors. Partners who haven't defined which features are approved for client data have no documentation when a state board or client asks. Ethics rules haven't caught up, and that gap won't protect anyone in a complaint.
What to watch: State CPA boards are starting to ask about AI vendor controls in peer review. Firms that build the approval gate now will have answers ready.
📊 FINANCE OPERATIONS — BCG Studied Finance Teams Running 4+ AI Agents. The Error Rate Jumped 39%.
What happened: BCG found workers managing four or more AI agents simultaneously experience cognitive overload and make 39 percent more major errors. Finance teams rank as the highest-risk group because they're already running QuickBooks AI, tax research tools, document extraction and general-purpose chat in parallel.
Why it matters: In accounting, you can't fix a wrong number with a quick edit. A variance analysis error or tax return mistake has real consequences for the client and the firm's liability. Running four tools at once isn't a productivity strategy. It's a risk multiplier.
What to watch: The fix is a sequencing plan: one tool per workflow, proven before the next one goes in. Firms that get this right will have cleaner audit trails and fewer surprises at quarter close.
💰 CFO METRICS — Most CFOs Can't Explain AI Returns to Their Board
What happened: Finance teams are seeing real productivity gains from AI tools. CFOs still can't explain those gains to their boards. The problem isn't the tools. It's the metrics. The Nexairi CFO playbook lays out three specific numbers that translate AI spend into something a board can actually evaluate.
Why it matters: Boards want AI ROI measured the same way they measure capex. CFOs who can't answer with a number are walking into those meetings exposed. Seat count and adoption dashboards won't satisfy an audit committee in 2026.
What to watch: AI ROI is heading toward standard board reporting by Q4. CFOs who build the measurement framework now won't be scrambling when the audit committee asks.
🤖 AI TOOLS — OpenAI Published a Finance Workflow Guide. Nobody Covered It.
What happened: OpenAI published a five-workflow guide for finance teams using Codex on May 12. It covers monthly business reviews, variance bridges, reporting packs, model maintenance and scenario planning. No accounting or finance press covered it before Nexairi did.
Why it matters: Codex isn't ChatGPT. It takes structured data (spreadsheets, dashboards and prior reports) and produces complete, source-cited first drafts without writing a line of code. The window to test it before competitors catch on is about 60 days.
What to watch: If Codex gets native Excel or Google Sheets integration, adoption will move fast. Teams piloting it now will already know how to use it when that happens.
Outside Nexairi
Anthropic now has a Claude plan built for small businesses — Anthropic
Anthropic released a small-business plan on May 14 with simpler pricing and workflows designed for teams without a dedicated AI staff. CPA firms and small accounting practices are a natural fit.
AI chatbots are leaking real phone numbers pulled from training data — MIT Technology Review
MIT Technology Review found chatbots surfacing private phone numbers from their training data, with AI-related privacy complaints up 400 percent over last year. It's the same data exposure problem CPA firms manage with vendor AI, just applied to contact info instead of financial records.
70% of executives say they need AI platforms they own outright — MIT Technology Review
A MIT Technology Review survey found 70 percent of executives believe they need AI platforms they own and control, not vendor-hosted models. For finance and legal teams under strict data rules, that view is already changing how they evaluate tools.
Tool Worth Knowing: Agent FM
Agent FM monitors your AI agents in real time so you can see what each one is doing, catch stalls early and flag errors before they land in a deliverable. For finance teams running multiple tools across reporting and document work, it's a control panel for the whole stack.
Deeper Read
Data readiness for agentic AI in financial services — MIT Technology Review
MIT Technology Review makes the case that most financial services firms aren't data-ready for AI agents yet, and explains what readiness actually looks like in practice.
Quick Hits
- AI ROI beyond seat count — what finance teams should actually measure
- AI policy template for small CPA firms and bookkeepers
- Google Workspace Gemini: what CPA firms should turn on first
- CPA vendor due diligence checklist before any AI tool touches client data
- Claude AI cracked an 11-year-old Bitcoin wallet holding $400K
- Agent FM: monitor all your AI agents from one place
- Linchpin: open-source self-hostable AI agent runtime
- Notion launches a full developer platform to build on top of it