AI is already inside CAS and accounting firms. The bigger question is whether firms are actually changing the way work gets done.
For the past several years, the accounting profession’s AI conversation has moved quickly.
In 2023, firms were asking: “What is it?”
By 2024 and 2025, the question became: “What tool should we buy?”
Now the question that matters is: “How should work flow through my firm?”
That’s the shift Aaron Berson, CPA, Founder and CEO of Fringe Advisory, explored during, Beyond the AI Buzz: Leveraging AI & Emerging Technology to Transform Accounting.
And it changes the conversation completely.
AI is no longer simply a software decision.
It’s becoming an operating-model decision.
First, Understand What You’re Actually Buying
One reason firms can be disappointed with AI is that the term itself has become too broad.
As Berson shares, four very different technologies frequently get grouped under one label:
- Rules & RPA follow predetermined instructions.
- Predictive AI learns patterns from historical data and predicts likely outcomes.
- Generative AI produces language, analysis, summaries and drafts.
- Agentic AI combines generative AI with tools and permissions to complete multi-step tasks within established boundaries.
For a CAS practice, those differences can look like this:
- Rules/RPA: recurring journal entries or bank-feed rules
- Predictive AI: suggested coding, anomaly detection or cash-flow forecasts
- Generative AI: drafting a client email or variance narrative
- Agentic AI: running a reconciliation process and escalating only the exceptions
The practical takeaway for accounting firms is simple:
Ask every vendor what kind of AI they’re actually selling.
The answer should change what you expect the technology to accomplish.
AI Adoption Isn’t the Problem Anymore
Approximately 40% of organizations report already using generative AI organization-wide, up from 22% a year earlier. But only 16% say AI is currently central to workflow, and only 18% measure the ROI of their AI initiatives.
That tells us something important. Organizations are adopting AI faster than they’re transforming around it. Meanwhile, professionals themselves are moving even faster.
According to the data from Thomson Reuters Institute, “2026 AI in Professional Services Report:”
- 82% of current AI users use it at least weekly
- 57% use public generative AI tools for work
- 31% use paid industry-specific AI
- 1 in 3 use AI tools their firm never approved
Employees aren’t merely experimenting anymore.
They’re building AI habits.
That creates both an opportunity and a risk.
If leadership doesn’t provide direction, people don’t necessarily stop using AI. They may simply use whatever tools they can access. That’s how shadow AI becomes an organizational problem.
Accounting Firms Have an Embedding Problem
The issue becomes even clearer when looking specifically at accounting firms.
- 88% use AI in at least one client service
- Only 30% have made AI their daily default.
- An average of approximately five hours per professional per week spent moving data between tools
- Professionals use an average of 10 applications and only 41% report full integration
This leads to a strong message:
Adoption isn’t the differentiator anymore. Embedding is.
A firm can own multiple AI-powered applications and still have highly manual operations. Having technology isn’t the same as having a system.
CAS Growth Makes the AI Question Even More Important
This challenge is particularly relevant for CAS practices. CAS continues to grow, but much of that growth remains tied to people.
The AICPA & CPA.com 2025 CAS Benchmark Survey comparison shows median annual CAS revenue increasing from approximately $1 million in 2022 to $1.61 million in 2024—a 61% increase.
Over the same period, however, median staffing increased approximately 31%, while net client fees per professional increased about 29%.
In other words: Growth is still significantly tied to headcount. That’s difficult to sustain in an environment where talent is scarce and expensive.
AI’s biggest opportunity in CAS therefore isn’t simply generating emails faster. It’s helping firms break the traditional link between revenue growth and proportional headcount growth.
That means creating capacity. And then using that capacity for higher-value client work.
The Goal Isn’t More AI. It’s a Better Operating Model.
When evaluating technology, firms need to ask another question: Who’s driving?
Berson breaks AI-enabled workflows into three practical modes.
- Assistant:
You drive. AI researches, writes, analyzes or summarizes when asked. - Automation:
Rules drive. Templates, triggers and feeds move work between systems without someone watching every step. - Agent:
The technology drives a bounded workflow while people supervise exceptions.
That distinction matters.
“A prompt is not an agent. An agent is a process with a brain and a leash.”
The leash matters just as much as the brain.
MCADA: A Map for the AI-Enabled Firm
Rather than jumping straight into autonomous agents, firms need to understand where their processes are today.
Berson’s MCADA Maturity Model provides that map:
0 — Manual: People do everything.
1 — Cloud: Software stores. People move.
2 — Automation: Rules move the work.
3 — Data: Signals drive decisions.
4 — Autonomy: AI works. Humans supervise.
Each stage builds upon the one before it. You don’t simply purchase Stage 4. You climb to it.
Stage 0: Manual — “The Process Is a Person”
At Stage 0, work lives in inboxes, spreadsheets and people’s memories. The process stops when the person who knows how it works isn’t available. The danger is that this may feel perfectly normal. But it creates a direct relationship between volume and labor:
More work = more people.
That’s the exact relationship CAS firms eventually need technology to break.
Stage 1: Cloud — Digital Doesn’t Necessarily Mean Automated
Moving into the cloud was transformational. But many firms simply moved manual processes into cloud applications.
“The software remembers. People still move everything.”
You may have a cloud General Ledger, practice management and bill-pay applications, yet employees still chase every document, move every piece of information and manage every handoff.
One clue that you’re here? You have 14 logins and you’re still copying and pasting between systems. The lesson: Subscriptions aren’t systems.
Stage 2: Automation — Rules Move the Work
At this stage, processes begin moving without constant human intervention. Bank feeds categorize routine transactions. Task templates create assignments. Reminders fire automatically. But there is an important warning:
“Automating a bad process gives you bad output, faster.”
That’s why process design needs to precede aggressive AI adoption. AI doesn’t eliminate the need for good processes. It makes them more important.
Stage 3: Data — The Work Starts Talking Back
Stage 3 represents a bigger shift. Firms begin managing with dashboards, exception reports, KPIs and capacity forecasts. Instead of asking: “What happened?” the organization begins asking: “What does this tell us?”
That’s particularly important for CAS because this is also where data begins supporting proactive advisory conversations.
Instead of reviewing every client the same way, teams can focus attention where the signals indicate something needs investigation.
Stage 4: Autonomy — AI Works, Humans Supervise
At Stage 4, technology begins to:
Observe → Reason → Act → Escalate
Confidence thresholds allow routine work to proceed while unusual items are sent to people for review. Humans aren’t eliminated. Their role changes.
Instead of reviewing everything, they spend their time reviewing the things AI wasn’t confident about.
This is why Berson describes autonomy as earned rather than bought. It rests on the processes, systems, data and controls created in the earlier stages.
What Does This Look Like in the Monthly Close?
Bring MCADA to life by applying the model to one familiar CAS workflow: month-end close.
At the manual stage, the checklist may live in Excel—or someone’s head—and employees chase documents through email.
At the cloud stage, everyone can see the checklist, but people still push every task.
With automation, bank feeds, rules, recurring assignments and client reminders begin handling predictable steps.
At the data stage, reconciliations and schedules draw from connected information and teams work the six closes that are stuck instead of treating all sixty the same.
At autonomy, AI can draft reconciliations and variance notes, post routine entries above established confidence thresholds and escalate anomalies with explanations attached.
Score Workflows, Not Firms
An accounting firm shouldn’t necessarily describe itself as: “We’re Stage 2.”
Different processes will be at different stages:
- Client onboarding might be Stage 1.
- Bookkeeping could be Stage 2.
- Accounts payable could be Stage 3.
- Monthly close might still be Stage 1.
That’s normal. Score workflows, not firms.
Before deciding where AI belongs, firms should first assess their current CAS workflows and identify where manual work, capacity constraints and inconsistent processes remain. Assess your current CAS workflows with the Infinite Ties CAS Assessment.
And follow the Rule of One:
Move a workflow one stage at a time.
Because:
“Autonomy on top of chaos is just faster chaos.”
That is why firms need to standardize their CAS processes before layering automation or AI on top of them.
Buy the Tools. Build the System.
Accounting firms will still purchase technology. They should. But purchased AI alone doesn’t create an AI operating model.
The presentation identifies four reasons purchased technology can stall:
- Bought AI stays siloed.
Each vendor automates its own part of the process, but no one owns the handoffs. - Bought agents are built for the average firm.
Your approval rules, review culture and unique processes may differ. - Product-hopping erases savings.
Repeatedly changing technology means repeatedly implementing, training and rebuilding processes. - Bought AI can make every firm look alike.
If everyone can buy the same technology, technology alone doesn’t differentiate the firm.
Buy the plumbing. Build the judgment. Clients only notice the part you own.
How Do You Pick Your First AI Project?
Once firms understand where their workflows are, another question appears: Where do we start?
That’s where the FIRST framework comes in.
Score a potential AI project from 1–5 on five dimensions:
F — Frequent: Does it happen weekly or monthly at meaningful volume?
I — Irritating: Does the team genuinely dislike doing it?
R — Rules-Based: Can you clearly explain the decision logic?
S — Safe: Would a mistake be visible, tolerable and reversible?
T — Traceable: Can you verify the output and measure the improvement?
Then score the project out of 25:
20+ — Green light
15–19 — Tighten the scope
Below 15 — Don’t make it your first project
Example: comparing prepaid amortization with tax-position memos.
Prepaid amortization scores 23/25, making it an ideal early candidate: frequent, repetitive, rules-driven, internal, reversible and measurable.
A tax-position memo scores only 11/25 because it’s less frequent, more judgment-intensive, riskier and more difficult to measure.
This is an important lesson. Don’t start AI transformation with the most impressive use case. Start with the most learnable one.
Governance Can’t Wait
People are already using AI. A firm without an AI policy doesn’t therefore have an AI-free organization. It may simply have unmanaged AI use.
Recommendation: use four guardrails before beginning a pilot:
- Data Rules — Use approved tools and don’t place client data into personal/free accounts.
- Review Lanes — Define exactly what a human needs to review before output leaves the organization.
- Accountability — The deliverable carries the firm’s name, not the AI model’s.
- A One-Page AI Policy — Clearly define approved tools, prohibited uses, where client information may reside and who employees should contact with questions.
The philosophy isn’t: Ban experimentation.
It’s: Give your team a lane, not just a ban.
Pilot AI Like a Business Initiative
Berson’s Four-Phase Pilot.
The formula: One workflow. One owner. One metric.
Then move through four phases.
Phase 1 — Baseline & Setup
Measure current hours, document the process, configure the technology and define review requirements.
Phase 2 — Shadow Mode
Run AI alongside the existing human process. Compare results, tune prompts and rules, and log every miss.
Phase 3 — AI-First, Human Review
Let AI perform the work while the owner approves the output. Track exceptions and time savings.
Phase 4 — Measure & Decide
Compare results to baseline, gather team feedback and make a decision.
And then make one of three calls: Expand. Adjust. Stop.
All three outcomes provide useful evidence.
The bigger threat to an AI pilot isn’t failure, it’s drift: no baseline, scope creep, tool-hopping, skipping human review or failing to establish a decision date.
The unit of evidence is a cycle, not a day. One full cycle per phase, minimum, the cycle sets the pace, you set the calendar. A weekly workflow can wrap in a month. A close-tied one needs two closes, one shadowed and one AI-first, so give it the calendar it needs.
What Happens to the Capacity AI Creates?
This may be the most important question of all for CAS leaders.
AI shouldn’t simply allow firms to complete the same transactional work faster.
The opportunity is to redirect capacity into:
- deeper client conversations,
- financial interpretation,
- exception analysis,
- forecasting,
- scenario planning,
- proactive recommendations,
- stronger client relationships, and
- higher-value advisory services.
And that creates another strategic challenge:
Pricing.
If AI dramatically reduces the time required to deliver a service, firms that remain tied exclusively to hourly billing risk allowing their own efficiency to work against them.
The destination isn’t simply AI-powered bookkeeping. It’s AI-powered advisory—including insight delivery, pricing and building the capacity the firm has gained, and ultimately developing the firm’s own AI operating model.
Three Things to Remember
Berson distills the roadmap into three frameworks:
- MCADA tells you where you are.
- FIRST tells you where to start.
- The Four-Phase Pilot tells you whether it’s working.
That’s a much more useful framework than trying to keep up with a list of 50 AI products that may look completely different six months from now.
The technology will continue changing. The management questions remain remarkably consistent.
The AI-Enabled Firm Starts With One Workflow
Accounting firms don’t need to automate everything tomorrow. They don’t need to build an army of AI agents. And they don’t need another year of simply watching what everyone else is doing.
Start smaller.
Score one workflow.
Pick one pilot.
Choose one owner.
Establish one metric.
Set the guardrails.
Run the cycle.
Then make the call.
As Berson states:
“One workflow. One owner. One metric.”
Two months later, the firm will have either its first meaningful AI win or its first meaningful AI lesson. Both are more valuable than another year of watching.
About Infinite-Ties
Infinite Ties is a training community built specifically for Client Advisory Services (CAS) teams. We help accounting firms equip their people with the skills, frameworks, and tools needed to grow and scale a profitable client advisory services practice. Through weekly education, practical resources, and peer collaboration, Infinite Ties supports firms at every stage of their CAS journey, from getting started to optimizing and scaling mature practices.
Ready to build a CAS practice that scales? To learn more about membership, training opportunities, and how Infinite Ties can support your CAS journey, contact Deneen Dias at: [email protected].
