Ground-Up or Top-Down? How Small Firms Actually Roll Out AI

At BKC, a New Jersey accounting firm serving thousands of clients, green ledger sheets are still making their way downstairs to the tax department. The reason, according to Patti Storms, is that two people at the firm have flat-out refused to use QuickBooks.

Storms, who supervises BKC’s client accounting team, has run out of patience. “Why are we accommodating this?” she asks. She expects AI to go the same way:

“You either use it, or you’re not working here,” she predicts.

She figures the firm has a few years before it gets to that point. For now, BKC guides staff on which AI tools to use, and the tax department gets its own specific training.

Chad Davis, co-founder of the remote Canadian firm LiveCA, made the opposite call after his team debated adding more training, documentation, and rules around their AI processes. 

“We’ve ended up at no,” Davis says.

They chose more one-on-one help instead. Some of his staff are using AI tools every day. Others haven’t logged in at all, and he isn’t ready to force the issue.

Two respected firms with plenty of AI mileage under them have chosen opposite playbooks, and neither is finished with their AI adoption journey—far from it, in fact.

Generally speaking, the question for accounting leaders in 2026 is not whether to adopt AI, but how tightly to manage its rollout—and seasoned experts land in very different places.

The Rollout Spectrum

At this point in the game, there’s no one-size-fits-all playbook for rolling out AI in an accounting firm. The current state of adoption and regulation spans a wide spectrum.

In fact, according to Financial Cents’ recent survey of 486 bookkeeping and accounting professionals, 95% of firms are somewhere on the AI adoption curve, but only 11% identify as “running” with it. In other words, almost everyone has started, and almost no one has finished.

Organic or Individual-led

For their part, LiveCA is in no hurry to cross any type of finish line—if one exists at all. They’ve chosen not to push usage mandates or require staff training. Instead, leadership has communicated clear expectations about what’s off-limits and urged staff to ask questions when they’re unsure. Beyond that, people explore at their own speed.

We’re not leaning into hard and fast rules of governance,” Davis says.

Chad Davis

Peer-led

In some firms, the curious “early adopters” step up to teach the rest of the team. And that instruction doesn’t necessarily have to come in the form of a structured course or learning session. At VP CPA’s, for example, the team hosts a recurring “Tech Tip Tuesday,” where those who tried something new that week show the rest of the team what worked and what to watch out for.

It’s like peer-to-peer education, which always lands better than ‘we’re going to have a formal training for everyone,’” says Bri Como, the firm’s manager of operations and systems.

Bri Como

Structured or Formal

Further along the curve, adoption tends to get more defined. At First Steps Financial, AI is a standing item on the firm’s quarterly goals. An internal AI committee regularly reviews tools and updates the firm’s overarching policy. Additionally, the firm’s handbook is currently being rewritten with an AI addendum. Staff use one approved, paid platform; free or personal accounts are off the table.

I don’t think it’s a choice,” says Gena Graziano, the firm’s integrator and chief of staff, referencing the firm’s decision to go all-in on AI.

Gena Graziano

Top-down or Mandated

In some firms, AI adoption is practically a condition of employment—which, as Storms mentioned, appears to be the trajectory BKC is on. 

It’s tempting to read all of this variation as one firm doing it right and another doing it wrong, but that’s definitely not what’s happening. These firms are simply catering to different needs and priorities. VP CPA’s is building habits through low-stakes visibility, First Steps Financial is promoting speed through committee-level clarity, and BKC is fostering consistency through pressure.

What remains consistent across all of these firms is that they’ve chosen their position intentionally. Let’s take a closer look at these stances and (hopefully) help you land in the right place for your firm.

The Case for Bottom-Up

The strongest argument for letting adoption happen organically is a practical one. It comes down to asking who actually knows where AI belongs in a firm, and the answer is likely not leadership.

When LiveCA debated formalizing AI training, the team leaned toward the counterintuitive—fewer group sessions, not more. This supports the idea that while a mandate can tell someone to use a tool, and formal training can help them get comfortable with the technology, it can’t replicate the moment someone sits next to them and guides them through their specific adoption journey.

People tend to get more out of a safe one-on-one fixing a specific problem, clicking the buttons together, than they do in a group setting where it’s kind of passive,” Davis says.

Chad Davis

It’s slow, person-by-person work, and it’s why Davis has refused to push out staff who haven’t logged in at all, even as others race ahead. Where a top-down system tends to treat non-adoption as a performance problem, Davis considers it a tradeoff: whatever ground someone hasn’t covered on the tools, they may be making up for in client relationships and institutional knowledge the firm can’t easily replace.

Jan Haugo, an AI implementation coach who’s run several adoption boot camps for accounting and bookkeeping firms, has watched this dynamic play out for years, and it’s shaped a rule of thumb she now gives every firm she works with.

I say let the people who are doing production tell you where they need the AI, because they’re gonna be the best,” she says.

Jan Haugo

Leadership can guess. Only the person doing the task knows where the opportunity really lies.

Haugo also points to a research study she saw suggesting that AI adoption stalls when leadership drives it top-down. “The people at the top don’t tell the people at the bottom how to do their work; they just do the work,” Haugo says. “But when the people at the top start telling the team members how to do the work with AI, AI doesn’t get adopted either. And if it does get adopted, it may not be in harmony or in unison.”

Even Haugo admits the hands-off approach has a blind spot. When she polls firm leadership on their team’s AI usage, the answer is often a bit wishy-washy: “‘I’m sure we’re using AI. I just don’t know how they’re using it,’” they’ll tell her. So, she asks: “Do you know what tools? They have no idea.”

She reads that less as an indictment of the organic approach than as proof of its premise that the energy is already there. People are already exploring. What’s missing is visibility, and the fix might be as simple as leaders asking their teams what they are actually using.

Adoption that grows out of real problems, solved by the people closest to them, tends to take root. Adoption handed down as an edict, even a well-intentioned one, tends to produce compliance without the right kind of buy-in.

The Case for Structure

The bottom-up philosophy says the AI adoption energy in a firm just needs room to run. The structured approach holds that, unsupervised, that energy can just as easily create serious risk—by pushing client data into a free tool without the proper security guardrails, for example.

That’s roughly what happened at Accounting Therapy. Until a few months ago, some staff were on a free ChatGPT subscription. Nothing sensitive, but no one had checked what protection the free tier actually offered.

We didn’t realize that it didn’t have that SOC 2 Type 2 security,” says Taylor Ammons, the firm’s integrator.

Taylor Ammons

Ammons didn’t stop at limiting access to certain tools. The firm’s quarterly “spirit rallies” now include a standing AI education block that walks staff through the firm’s written policy alongside hands-on training in two building segments: skills (reusable instructions for a recurring task) and projects (a bundle of context an AI tool can draw on).

As soon as you onboard [at] Accounting Therapy, you’re getting a paid AI account,” Ammons says. “It’s also just like training and educating on that too, as we learn and get better ourselves.”

Taylor Ammons

At one rally, every staffer built a “strategic thinking partner” project loaded with the firm’s brand voice, ideal client profile, and real client reviews. One bookkeeper messaged Ammons the next morning to say how much fun she’d had building it.

Como’s team at VP CPA’s created the same rule because of a public scare. After a widely reported incident where shared AI conversations turned up indexed on Google, the firm implemented strict security protocols: work accounts only, no personal logins, and no client data (e.g., bank numbers, Social Security numbers, or uploaded documents) going into any general-purpose tool.

This is something that we had to implement,” Como says. “If you’re uploading stuff and it has sensitive information on it, you become blind to it, and you don’t think anything of it. But that’s not good.”

Bri Como

VP CPA’s also built a firm-specific knowledge base that its internal agents are restricted to querying. When someone asks the firm’s process bot a question, it pulls from the team’s own documented workflows and guides. The security rule keeps data in; the knowledge base ensures that what comes back out is the firm’s answer, not a generic one.

Deborah Harris, CEO of Grow Group Global, took structure a layer deeper. Rather than let each staffer build their own private relationship with an AI tool, she created a shared “context engine”: a continuously refined profile of the business, its model, its strategy, and its voice. The whole team draws from the engine before any tool touches client work.

“One without the other is [a] no,” Harris says of pairing context with skills. “There’s no point.” Skip the shared context, and every staffer’s output reflects their own guess at what the firm sounds like. Build it centrally, and the team works from one source of truth.

But firms like these are still in the minority. Just 13% of AI-active firms represented in the AI report have a formal, documented AI policy; 87% don’t. But those that do aren’t just more organized; they’re also getting a measurably better return, reporting 35% ROI, compared to 14% for firms with no policy.

That’s what structure is really for: taking whatever’s already working for one curious person and making it available to everyone else. Otherwise, ten people exploring AI independently can produce ten half-built systems, ten different levels of exposure, and even ten different answers to the same client question.

The Mandate Question

Regardless of how adoption happens, leaders often eventually face an uncomfortable question: what do you do about the folks who won’t get on board?

Storms offered her answer in the opening lines of this piece. A firm serving thousands of clients, she reasons, can’t let individual comfort dictate the pace of the whole operation. Eventually, everyone gets in line or leaves. “[But] I think we’ve got a few years before that happens,” she says.

Davis has watched firm owners take that stance—“If you’re not on the train, leave,” as he puts it—and he isn’t necessarily calling it wrong. He’s just calculating a different equation.

That person has so much value with customer relationships, with institutional knowledge, that it’s not adversely affecting the firm to let them stay on their own timeline,” he says.

Chad Davis

Haugo’s research—the same study that shaped her earlier bottom-up thesis—backs Davis’ judgment here, too. A mandate can produce compliance. It can’t guarantee the fluency that makes AI actually useful in a client’s file.

So, again, the right approach depends on the firm. Storms runs a client accounting services team inside a firm of roughly 90 employees serving thousands of clients, where consistency matters more than any one person’s preference and two holdouts create real drag. Davis runs a 60-person firm purposefully built around slowness and retention, where losing someone with years of client trust costs more than the time it takes them to adopt a tool.

Both are weighing the same trade—what a mandate costs against what it buys—at different scales and in different cultures. Getting it wrong in your firm could mean either bleeding out your most experienced people or watching your firm-wide adoption sputter along at whatever pace your most reluctant employee sets.

What Both Ends Agree On

When you zoom out far enough on the organic-to-mandated spectrum, you can see that the two ends actually agree on the same three guardrails—and the data tells why.

Security is non-negotiable

Como’s team at VP CPA’s and Ammons’s team at Accounting Therapy arrived at a similar rule from different directions: a public scare in one case, and a gap in the other. Neither firm sits anywhere near the same spot on the spectrum. But both drew the same line for the same reason.

They’re not alone: in the same Financial Cents AI report, the most common way firms said they handle client data was refusing to put it into any AI tool at all. Around 16% said they take this approach with zero exceptions, and another 14% said they redact or anonymize any personal identifiers first.

Humans always check the output

Storms, whose firm leans toward required AI use, is unequivocal about the human-in-the-loop philosophy: “I am not willing to turn over numbers and calculations without human oversight,” she says.

Jean Zick, CEO of Juna Financial Solutions, extends the same instinct to clients now running their own books through AI. She wants those numbers routed back through her firm “so that we can review it…we don’t want them to give bad data to their boards.”

The survey suggests this instinct is almost universal: 90% of respondents agreed human judgment matters more, not less, as AI use grows, and only 19% said they trust AI enough to use it with limited review.

Learning spreads person to person

Even the more structured firms lean on the same mechanism the loosest ones use by default. Zick brought in an outside trainer to help get her team onto equal footing, and it fell short. “The training wasn’t as hands-on as they’d like,” Zick says.

The team members already strongest in AI are now leading the internal trainings themselves.

Graziano’s firm, one of the more formalized firms in this piece, still fills gaps by sending people to conferences and encouraging them to take classes wherever they are offered—and by following people like Hector Garcia and Jason Staats, who are “always teaching something.”

The AI report found this is how most firms actually build AI skills: 56% of firms said they’re building capability by sharing tips informally as opposed to more formal training (35%).

Underneath the disagreement over how much structure a firm needs is a common ground they all seem to rally behind: protect the data, keep humans in the loop on anything that matters, and let the people who are good at this teach the ones who aren’t.

Choosing Your Spot on the Adoption Spectrum

The real divide in AI adoption is decided versus drifted—not organic versus mandated.

BKC still has staff refusing basic accounting software, let alone AI. LiveCA still has staff who haven’t even logged in to their AI tools. Neither firm has solved the adoption problem, but both have decided, on purpose, how they’ll move forward.

Firm size and culture can help with this decision. A ten-person team of self-starters trading prompts organically has earned the right to stay loose. The energy Haugo describes is already there, and a mandate might actually impede it. A 90-person firm with entrenched habits and a client roster that can’t tolerate inconsistency probably needs what Storms and Graziano built instead.

And importantly, none of these decisions are final. The AI landscape is fluid, and your approach to education and adoption should follow suit. Your spot on the spectrum isn’t a decision you make once and forget. You have to keep revisiting it as you go.

Otherwise, just like the folks Storms describes as clinging to green ledger books, you risk being left behind.

Be among the first to know

Enter your email address to receive new issues of the magazine straight to your inbox.