Where we're going with AI
The Next 12–18 Months
The CAS practice you're running today won't exist in 18 months. That's not a threat — it's the opportunity.
A practitioner's view on what's actually coming, what it means for your firm, and why the next year and a half will separate the practices that lead from the ones that scramble to catch up.
Where Things Stand
Let me be direct with you.
I've spent the last two years inside the AI transformation — not reading about it, not attending webinars about it, but building it into the daily operations of my own CAS practice. And from where I'm sitting, what I see coming in the next 12 to 18 months is more consequential than everything that's happened so far. Combined.
That's not hype. I don't do hype. It's a practitioner's honest read of the trajectory.
Here's what I know: AI's theoretical task coverage in financial roles has already hit 85%. Actual deployment across the profession sits around 20%. That 65-point gap is where the next 18 months play out — and it won't close gradually. It'll close in waves, the same way data entry went from negligible AI coverage to 67% in just two years. One day the gap is there. Then a new model drops, a new tool ships, and an entire category of work tips over into automation.
The question isn't whether this happens. It's whether you're positioned when it does.
What's Coming First
The commodity layer has a timeline.
Let's start with the most immediate impact: the work that pays most of your bills today is headed toward near-zero delivery cost.
Bookkeeping. Bank reconciliations. Transaction categorization. Basic compliance filing. Routine financial statement preparation. This is the commodity layer — repeatable, pattern-based work that AI handles faster, cheaper, and increasingly more accurately than a human team.
The proof points aren't theoretical anymore. Five autonomous bookkeeping products launched in a single two-week window earlier this year. BILL's AI agents eliminated 80% of W-9 collection steps. KPMG used AI's capability as a lever to negotiate a 14% fee reduction from Grant Thornton — at the enterprise level. AI can categorize 960 bank transactions in 12 minutes.
In 12 to 18 months, expect autonomous bookkeeping to be table stakes for any tech-forward firm. Not a differentiator — a baseline. The firms that haven't automated their commodity layer will be competing on price against tools that are essentially free. And they'll lose.
This isn't an argument to panic. It's an argument to stop defending work that's already being commoditized and redirect your capacity toward what AI can't do.
Where the Advantage Shifts
The dividing line moves to knowledge.
If the commodity layer compresses, what survives?
Context. The knowledge you carry about your clients, your industry, your practice's way of doing things — structured into a form that AI can actually use. This is context engineering, and it's the skill that separates firms that use AI from firms that are transformed by it.
Here's what I mean concretely. You know that your client's Q4 revenue spike is seasonal, not a growth trend. You know the owner is planning to exit in two years. You know the new hire in AP is the owner's nephew. None of that is in the chart of accounts. All of it shapes the right advice.
Over the next 12 to 18 months, the firms that pull ahead will be the ones that encode this knowledge — client profiles, decision histories, SOPs with real decision logic, pricing frameworks, industry-specific playbooks — into structured, machine-readable systems. Not because it's trendy, but because every piece of context you encode makes your AI more useful. More useful AI produces better output. Better output justifies higher fees. Higher fees fund more context development.
That's the compounding flywheel. And it rewards early movers disproportionately.
The firms that don't do this work will have the same AI tools as everyone else — producing the same generic output that requires the same heavy editing. Same tools, wildly different results. The difference is the context.
The Economic Reality
Your pricing model is about to get tested.
Most CAS practices still price by volume. Hours worked, transactions processed, returns filed, employees on the client's payroll. That model made sense when delivery required proportional human labor.
AI breaks that math.
When your cost to categorize 960 transactions drops to near-zero, but you're still billing the client based on transaction volume, you're in a race to the bottom — against software. Your revenue drops as your costs drop. The margin compression won't hit all at once, but it doesn't need to. It just needs to happen enough that clients start asking why they're paying what they're paying.
In the next 18 months, I expect value-based pricing to move from "something progressive firms do" to "the only model that makes economic sense." Bookkeeping becomes the data layer — the infrastructure that feeds advisory. Advisory becomes the product.
The benchmark I'm seeing in my own practice and across the firms I work with: $20K in annual revenue per client, supported by $200 to $2K per month in AI tooling costs. That's a 20:1 ratio. And with AI costs dropping — GPT-5.4 at $2.50 per million input tokens — the economics only get better.
If you're still pricing by the hour in 2027, you're leaving money on the table at best and losing clients at worst.
The Team Question
Fewer people, more clients, different roles.
Here's the part of this conversation that makes people uncomfortable: AI changes the math on how many clients one human can support.
Today, most CAS practices operate at roughly a 1:8 ratio — one staff member per eight clients, give or take. That ratio was set by the volume of manual work each client requires. When AI handles the commodity layer, that ratio shifts. Dramatically.
The 12-to-18-month implication isn't mass layoffs — it's role transformation. The staff accountant who spent 30 hours a month on manual bookkeeping across six clients becomes the person who reviews AI-prepared books across 20 clients and spends the recovered time on advisory prep, client communication, and exception handling. Fewer people doing more work, but higher-value work.
This requires two things most firms don't have yet: a Champion and an Operator. The Champion is your experimenter — the person who looks at a workflow and sees where AI fits. The Operator is your systematizer — the person who takes the Champion's discovery and makes it repeatable across the team. Most firms have one or neither. The firms that have both are the ones closing the adoption gap fastest.
If you haven't identified these people on your team yet, that's your first move. Not buying a tool. Not signing up for a platform. Finding the people who can bridge between what AI makes possible and what your practice actually needs.
The Honest Assessment
The window is real, and it's finite.
I want to be clear about something: I'm not saying every practice needs to be fully AI-transformed by next year. That's not realistic, and anyone telling you otherwise is selling something. Institutional constraints are real — software vendor lock-in, regulatory expectations, professional standards, and a growing wave of legislative attention. Over 78 state-level AI bills are working through legislatures right now. The EU AI Act is in effect. These aren't going away.
But here's what I've learned from inside this: the constraints are narrower than most practitioners assume. The regulatory requirements are on final sign-off and professional judgment — not on the preparation, analysis, and drafting work that makes up the vast majority of your workflow. The platform lock-in applies to work inside Xero or QBO — not to the communication, document processing, analysis, and advisory prep that happens between those platforms.
The gaps between your core platforms — that's where the highest-leverage AI investment lives. And no vendor can gatekeep it.
The 12-to-18-month window matters because the adoption gap is closing in waves. Right now, most firms are clustered around the same 20% adoption level. The separation hasn't happened yet. But it will — and the firms that get to 40% or 60% adoption first don't just gain efficiency. They set the market price. Everyone else either matches them or accepts margin compression.
The Path Forward
What I'd do if I were starting today.
If I were a CAS practice owner looking at this landscape today — knowing what I know after two years inside it — here's what I'd focus on in the next 18 months.
First, automate the commodity layer as fast as possible. Not to cut staff, but to free capacity. Every hour your team spends on work AI can handle is an hour not spent on advisory work that commands premium fees.
Second, start building structured knowledge. Pick one client. Document not just their financials but their context — business goals, owner dynamics, industry constraints, decision history. Build it in a format AI can reference. Then do it for the next client, and the next. This compounds. Start now.
Third, find your Champion. The person on your team who gets excited about this, who tinkers, who breaks things and figures them out. Give them permission and time. Then pair them with someone who can systematize what they discover.
Fourth, revisit your pricing. If any of your engagements are priced purely on volume — hours, transactions, headcount — model what happens when AI cuts the delivery cost by 50%. Then by 75%. If the model breaks, the pricing needs to change before the cost compression forces it.
Fifth, build in the gaps. Don't wait for Xero or QBO to give you better AI tools. Build your AI capabilities in the spaces between platforms — client communication, advisory prep, document processing, knowledge management. That's where you have full control and zero vendor dependency.
None of this requires a massive budget. None of it requires a computer science degree. All of it requires a decision to start — and the honesty to admit that what got you here won't get you where you need to go.
The next 18 months will be the most consequential period for CAS practices in a generation. Not because AI is new — we're past that. But because we're hitting the inflection point where theoretical capability becomes operational reality. Where the gap between what AI can do and what firms are actually doing starts closing fast enough to create real winners and real casualties.
I don't know exactly how it plays out. Nobody does. But I know this: the firms that move with clarity, build with intention, and refuse to mistake activity for progress will come out the other side stronger than they've ever been.
That's where I'm headed. I hope you are too.
— Peter McCarroll, CAS Practitioner & Founder, The AI Accountant