On Tuesday morning, Oct. 6, Citrin Cooperman announced that it had acquired Rosetree Solutions, an AI consultancy that has served as a launch partner of both Anthropic and OpenAI. The deal closed Oct. 1. Terms weren't disclosed.
According to Accounting Today's Paige Hagy, Rosetree brings roughly 60 employees and expands Citrin's ability to design agents, applications and workflow integrations for middle-market clients. CPA Practice Advisor's Jason Bramwell covered it too. Rosetree co-founder David Rosenbaum becomes Citrin Cooperman's Applied AI Practice Leader.
Here's the part I keep coming back to. The pitch is a pairing: Citrin's knowledge of client margins, pricing and cash flow, plus Rosetree's ability to actually build and integrate. The release also cites a Goldman Sachs survey: 76% of small businesses use AI, but only 14% have it fully embedded in core operations.
That gap is the market. And larger firms are clearly deciding they'd rather turn their client-finance knowledge into implementation work than leave AI to standalone consultancies.
Most of you reading this are not going to buy a 60-person AI shop. You don't need to. But you do need a deliberate answer to a simple question: Where does my firm play, if at all?
I see four entry points. Diagnose, design, implement, or govern. Pick one or two on purpose. Refer out the rest.
Skills. You already have most of them. You know the client's close process, where the spreadsheets live and which reports nobody trusts. Add a working knowledge of what today's AI tools can and can't do, and a structured way to document processes.
Where to stop. A diagnostic is an opinion about readiness and opportunity. It isn't a promise that a tool will work. Put that in the engagement letter. Be careful recommending specific vendors if you have referral or reseller arrangements, and disclose them.
Pricing. Fixed-fee assessment. Scope it tight — a few workflows, a written findings memo, a prioritized list. This is a great door-opener.
Partner or build? Build. This is the natural first step for almost any firm.
Skills. Now you're mapping the future-state workflow: what data feeds what, who reviews, where the controls sit, what "done" looks like. Process mapping, data literacy and an internal-controls mindset matter more than coding.
Where to stop. Designing a workflow that touches financial reporting is a different thing from being responsible for its output. Spell out that the client owns the decisions and the final system.
Pricing. Project-based, priced by the number of workflows or processes in scope. Not hourly, if you can help it.
Partner or build? Build if you have one or two people who enjoy this work. Otherwise, partner with an implementer and keep the design role yourself — it's where your client-finance knowledge earns its keep.
Skills. This is the Rosetree layer: building, configuring and integrating — connecting the AI tool to the ledger, the bank feeds, the document store. It takes technical staff, testing discipline and ongoing support capacity.
Where to stop. This is where independence gets real. If you perform attest work for the client, building or running systems that feed their financial statements can put that independence at risk. Check the rules before you take the engagement, not after. And implementation brings real liability if something breaks — confirm your coverage.
Pricing. Project fee for the build, then a monthly support or managed-service fee. Recurring revenue is the prize here, but so is the support burden.
Partner or build? For most small and mid-sized firms, partner. Look for a partner who knows accounting workflows, will put roles and responsibilities in writing, carries their own insurance, and lets you stay the client's trusted advisor. Ask them how they handle data security before you ask about price.
Skills. Controls, risk assessment, policies, monitoring. Who can use which tools, with what data, and how outputs get reviewed. Honestly, this is the most "accountant" of the four.
Where to stop. Govern and implement for the same client, and you're checking your own work. Keep those roles separate, or pick one. If you're offering anything that looks like assurance over AI systems, follow the professional standards that apply and don't call it assurance if it isn't.
Pricing. Annual policy-and-review engagement or a recurring retainer. Quarterly check-ins work well.
Partner or build? Build. It plays to skills your firm already has, and it pairs naturally with diagnose — even when someone else handles the build.
Citrin Cooperman bought its way into the implement layer. That's a capability-development signal, not a template. Your version might be diagnose plus govern, with a trusted implementation partner on speed dial. That's a perfectly good model — as long as it's deliberate and the boundaries between advisory, implementation and assurance are written down.
(Full disclosure: I used AI tools to help organize my notes and write this piece. The opinions — and any mistakes — are mine.)
So here's my challenge. Before the end of the month, write one paragraph: which of the four levels does your firm offer, which do you refer, and where do you stop? If you can't write it, your clients can't understand it either.