The Crossing Report — Issue #31

The Measurement Issue

Published November 16, 2026 · By The Crossing Report · 54 min read

This issue: 82% of professional services firms don't track AI ROI. The firms that measure are generating 2.3× more from their investment. This issue gives you the full 4-metric scorecard — and the specific actions to run it in two hours a month.


You've Been Using AI for a Year. Do You Know If It's Working?

Thomson Reuters surveyed 1,500+ professionals across 27 countries for its 2026 AI in Professional Services report. GenAI adoption doubled from 22% to 40% in the last year. That's the headline. The number underneath it is the problem.

Only 18% of professional services firms collect any AI ROI metrics. Forty percent don't know whether their organization tracks ROI at all. Among the 18% that do measure, most track internal metrics — cost savings, usage rates — not business outcomes like client satisfaction or revenue. Sixty-six percent report productivity gains. Only 20% report revenue lift.

That gap — productivity gained, revenue not captured — is the measurement problem in one sentence.

SPI Research confirms it from a different angle. AI is now used on 43.6% of professional services projects, up from 23.1% last year. 84.8% of firms report productivity and time savings. Only 29.9% report revenue generation. Among firms that have deployed complex agents or AI-first operating models: 20.5% EBITDA margins. Chat-only firms: 13.3%. 36.8% of firms track no AI outcomes at all.

Two independent data sources. Same finding. You are very likely one of the 82%.

This is not a technology problem. It is a measurement problem. The firms closing the gap — the ones reporting revenue lift, not just time savings — are not using different tools. They are asking a different question first: how much does one workflow cost today?

That number is your baseline. Everything that follows depends on it.


The Pricing Trap Your Firm May Already Be In

Clio surveyed over 1,700 legal professionals for its 2026 Legal Trends report. Here is the number that should stop you:

86% of solo law firms and 78% of small firms have made no pricing changes since adopting AI.

That means every hour AI saves you on a matter, you are handing that savings to your client on the invoice. The efficiency gains are real. The revenue gains are not flowing through to the firm.

The solo-to-small firm breakdown is sharper than the headline: only 33% of small firms have increased revenue with AI, compared to 60% of enterprise firms. The difference is not the tool. It is the pricing model and the measurement discipline behind it.

The client side is moving faster than most firm owners realize. Everlaw surveyed 300 in-house legal teams: 54% say AI has already reduced the volume of work they send to outside counsel. Another 44% expect to reduce it within a year. Among the heaviest AI users, 73% are already sending less. 81% expect outcome-based pricing to eclipse the billable hour within five years — including 47% within three years.

The economics are clear. AI makes you faster on the same hourly invoice. Your clients are figuring out they can do more work themselves. The only stable position is to price on value delivered, not time consumed. And you cannot price on value delivered if you do not know what your work costs.

One action before you close this: identify one recurring matter type where your post-AI delivery time has dropped more than 20%. That is your first retainer conversion candidate — and the first workflow to measure.


The Professional Standards Floor Is Rising

Three things need to happen before the end of this month if you run a CPA or tax practice.

One. The AICPA published its first AI Practice Aid for Financial Statement Auditors on October 13 — 51 pages covering both sides of AI in audit: when clients use AI in their financial reporting, and when you use AI to do the work. The core holding: "the auditor remains responsible for the audit opinion." Not yet a required standard, but the scaffolding for what your peer reviewer will ask about in 2027.

Three requirements the practice aid implies: (1) a written list of AI tools your firm has approved for use; (2) a documented human review step before AI-assisted work product reaches a client or regulator; (3) a staff competency check covering each tool's known limitations. If you do not have all three, you have a professional standards gap heading into year-end.

Two. The Journal of Accountancy published a 12-item AI cybersecurity self-assessment checklist in its October issue — written by CPA/CITP Sanjay Chadha, drawing from NIST, CIS Controls, and ISO 27001. Designed specifically for CPA firms. The most urgent items for a small firm: AI governance (approved tools and output review requirements), data classification, and AI agent safeguards. It is searchable at journalofaccountancy.com.

Three. The IRS is replacing the FIRE system on December 31. If your firm files 1099s, W-2s, or ACA forms directly through the IRS — not through software that handles the transition automatically — you need an IRIS TCC. Applications take up to 45 days. The registration window closes approximately now. Go to irs.gov/iris this week. The penalty for late filing starts at $60 per form.

On November 13 — three days after this issue sends: Microsoft Copilot's accounting skills go general availability in Excel. Journal entry drafting, bank reconciliations, automatic debit/credit balance checking, and an optional audit trail that documents every step Copilot takes. No Frontier enrollment required — a standard M365 Copilot license is enough. Before you turn it on: time one reconciliation batch manually. That 30–60 minute number is your before. The after comes at 30 days. Without the before, you cannot calculate the ROI — which is precisely what 82% of your peers are failing to do.


The 4-Metric Scorecard: What to Measure and How to Run It in Two Hours a Month

You do not need new software to track AI ROI. You need four numbers, pulled from systems you already have.

Utilization rate. What percentage of billable or client-facing time is tracked and productive? For a CPA firm, this is actual hours worked on client files against total hours in the week. For a law firm, it is billable hours against available hours. AI should move this up — less administrative overhead, faster delivery. If utilization is not improving after six months of AI use, your tools are beside your workflow, not inside it.

Realization rate. Of the fees you bill, what percentage do you actually collect — and at what margin? AI efficiency that gets handed back to clients as faster turnarounds without pricing changes will show up here as flat or falling realization. If utilization is rising but realization is flat, you are absorbing the gains instead of capturing them.

Project margin. Revenue minus direct cost on a per-engagement or per-matter basis. Track this for each major service line, not firm-wide. AI's ROI is concentrated: a practice area that has adopted AI fully will show materially different margins than one that has not. The comparison between service lines is your signal for where to invest next.

Client retention. What percentage of your top-20 clients are still active after 12 months? Firms using AI to deliver faster, more proactive insight retain clients at higher rates — but only when the AI output is reaching the client, not just saving internal time. Track it explicitly and connect it to what you changed.

The two-hour monthly process: pull these four numbers on the last Friday of the month. The source is your practice management system, your billing software, and your CRM. Compare to the same month six months prior. Identify the number moving fastest and the one moving slowest. The slowest one is your next action.

Kanopy Labs measured first-year ROI across seven AI agent deployments: lead qualification at 340%, customer support at 280%, invoice processing at 190%. The pattern across all categories: firms that set a dated baseline at launch and measured at 30/60/90/180 days saw consistent ROI. Firms that deployed without a baseline could not demonstrate return at all — and were more likely to abandon the tool at the first friction point.

If you do not have a baseline today, run the scorecard this month as your starting point. The measurement gap is not abstract. It is a 7.2-point EBITDA gap, documented by SPI Research, between the firms that measure and the firms that do not.


Three Platforms, One Window (For Accounting Firms)

In the last 30 days, four of the largest U.S. accounting platforms launched agentic AI. If you are on any of them, the measurement window is open now — not in Q1.

CCH Axcess Tax (Wolters Kluwer, October 23): agentic tax return preparation. The system classifies source documents, maps data to tax forms, executes calculations, and flags exceptions before you open the file. It is inside your platform if you are on CCH Axcess — no additional purchase. Measurement requirement: before you run a client return through the agent, time one return manually. That is your before number.

Agentic Book Close (Intuit, October 28): firms upload their own close SOPs; the AI builds automated close workflows from them — reconciliation, balance sheet review, financial summaries — executed according to your documented process, not a vendor template. The measurement prerequisite is built into the feature itself. You must document your close process to use it. That documentation is also your baseline.

CoCounsel Tax, Audit & Accounting (Thomson Reuters, October 22): fully rebuilt agentic platform grounded in Checkpoint — 12 million primary sources maintained by licensed CPAs and tax attorneys. It builds deliverables for human review before anything files. If you are on a Thomson Reuters platform, this is a conversation with your rep.

Sage Intacct AI Workspaces (November 5–6): AP agent in limited rollout, Agent Marketplace live, additional agents in development. Four major platforms, 30-day window.

The BDO Audit Innovation Survey puts the risk in numbers: 99% of finance leaders use AI for financial judgments. Only 49% can fully document or substantiate their AI-assisted outputs. 86% worry about overreliance. The AICPA CART framework — completeness, accuracy, rights/obligations, presentation/disclosure — maps the four AI risk categories every audit firm should be able to answer. These are peer-review criteria.

The measurement prerequisite for all four platforms is identical: document what your team does manually today, then run the same workflow through the agent. The before-and-after comparison is your ROI proof case — and the documentation the AICPA practice aid, the JofA checklist, and IRS OPR guidance all require simultaneously.


The Law Firm Measurement Toolkit

Anthropic added six Legal Artifacts to Claude in late October — ready-made templates, no setup required, available on a standard Professional or Team plan. Search "Legal Artifacts" inside Claude; no vendor contract, no IT ticket.

The six templates and what they measure:

  • Matter Tracking — the foundation of project margin measurement. If you do not currently track time and cost per matter type, start here. This is the before number for the 4-metric scorecard applied to legal work.
  • Citation Checking — surfaces mismatched quotations, incorrect page references, and unsupported claims in filings and memos. Directly relevant for SB 574 compliance (attorney verification of AI-cited authorities, effective January 1, 2027).
  • Regulatory Monitoring — set up for the practice areas where your clients face compliance deadlines. Proactive guidance is what reduces client churn; this makes it repeatable.
  • Contract Review — flag deviations from your own standard terms on incoming documents.
  • Jurisdiction Comparison and Multi-Contract Review — for transactional and multi-jurisdiction practices.

California Bar formal guidance (July 30, 2026) adds an obligation most engagement letters do not yet reflect: consent is required before inputting confidential client information into AI systems — not disclosure, specific consent. The distinction matters. If your engagement letter says you use AI, that is disclosure. If it asks the client to agree that their documents may be processed by Claude or Copilot, that is consent.

Three-step engagement letter update: (1) add a specific AI consent clause naming the systems your firm uses; (2) identify which client data flows through each tool; (3) add a vendor review obligation — when you add a new AI tool, your obligation to update client agreements triggers. California is the standard. It will be the national standard within 24 months.


The Consulting Margin Trap

IBM reported Q3 2026 consulting revenue of $5.335B — 2% constant-currency growth, with gross margins declining from 28.9% in Q2 to 27.4% in Q3. IBM's strategic response is "services as software" — embedding AI-powered assets into client engagements to protect margin as the per-hour model compresses.

McKinsey moved to performance-based pricing in late 2025, explicitly citing AI: when AI compresses weeks of analyst work into hours, the traditional model of pricing that work by the team-week no longer reflects value. BCG and Bain are in active internal discussions. 25% of McKinsey's global fees are now outcome-based.

For a 10-person consulting firm, the translation is concrete: identify two or three workflows you do repeatedly for clients — market landscape reviews, operational audits, pricing analyses — and build a deliverable template from them. That template is your first "services as software" asset: expertise codified once, priced as a deliverable, not an hourly rate.

The SPI Research EBITDA gap (20.5% for agent-first firms, 13.3% for chat-only) and the IBM margin compression data tell the same story from opposite ends. Firms embedding AI into repeatable, productized work are pulling ahead. Firms selling hours of AI-assisted work are facing the same margin pressure IBM reported this quarter.

The measurement prerequisite: you cannot price on outcomes if you do not know your cost per outcome. The 4-metric scorecard applied to consulting means tracking project margin per engagement type — specifically the ones where AI has changed your delivery cost. If your fees have not moved since your AI tools went in, you are absorbing the efficiency gain.


Three Questions Before You Go Agentic

If you are evaluating agentic AI tools — the systems that execute multi-step tasks without re-prompting — answer three questions before you sign anything.

Can you currently measure the workflow you plan to automate? If you have no baseline for how long the workflow takes and what it costs, you cannot calculate ROI on the automation. Set the baseline before you deploy.

Is the workflow documented well enough for AI to follow? Every agentic tool — Intuit's Agentic Book Close, Clio's Vincent, the Claude Legal Artifacts — requires you to describe the process in enough detail that an agent can execute it consistently. If your workflow varies by person or is undocumented, the agent will not deliver consistent output. Documentation is the prerequisite, not the afterthought.

Is human oversight built into the workflow by default, or does it need to be configured? The AICPA practice aid, the JofA checklist, and California Bar consent guidance all require human review of AI-assisted outputs before delivery. Some tools build this in by default — CoCounsel Tax prompts for review before anything files. Others require you to configure it. Know which is which before you deploy.

Three questions. If you cannot answer all three, you are not ready for agentic. Build the measurement and documentation baseline first.


This Week: Pick One Workflow

The measurement gap is not theoretical. It is the difference between 29.9% and 84.8% — between firms reporting revenue gains and firms reporting only time savings. Between 20.5% EBITDA margins and 13.3%.

Pick one workflow. Time it manually this week. Write it down. That is your baseline.

You do not need a scorecard yet. You do not need new software. You need one number for one workflow.

For accounting firms: one tax return, or one monthly close. For law firms: one contract review, or one motion. For consulting firms: one client deliverable of the type you produce most often. For staffing firms: one sourcing run, from intake to submit.

That number is your before. Everything that follows is a comparison.

IRS FIRE action item (accounting firms): If your firm files information returns directly through the IRS — not through software that handles the FIRE-to-IRIS transition automatically — apply for an IRIS TCC at irs.gov/iris this week. The registration window is 45 days. December 31 is the hard stop.


Issue #31 | The Crossing Report | November 16, 2026

This is the kind of intelligence premium subscribers get every week.

Deep analysis, cross-sector patterns, and the frameworks that help professional services firms make the crossing.

Subscribe free

Related Reading

This is a sample issue — new ones go to subscribers

New issues of The Crossing Report ship exclusively to subscribers every week. Free in your inbox.

Free weekly digest. No spam. Unsubscribe anytime.