What Is Actually Stopping Small Firms From Using AI

Ask a managing partner at a 50-person law firm why they have not deployed AI in any meaningful way, and you will usually hear one of three answers. The tools are too expensive. The implementation is too complex. And the risk of getting locked into the wrong vendor is too high.

All three concerns are reasonable. Enterprise AI contracts are not cheap. A serious deployment with a major provider can run into six figures annually before you have built a single workflow. For a firm billing 30 to 40 lawyers, that arithmetic is difficult to make work. The numbers look even worse when you factor in the time a senior operations or IT person needs to spend managing the vendor relationship, configuring the system, and then reconfiguring it every time the provider changes the product or the pricing model.

The result is that AI adoption in professional services has largely split along firm size. Large firms with dedicated technology budgets have moved. Everyone else has watched, run a few pilots, and then quietly shelved the project.

That split is now breaking down.

What Has Changed

The release of Qwen 3.8, an open-source AI model from Alibaba's Qwen team, is a useful marker for a broader shift that has been building for several months.

Qwen 3.8 is a small model, at least by the standards of the last two years. It has 3.8 billion parameters. By comparison, the large commercial models that set the benchmarks in 2023 had hundreds of billions. The assumption built into most enterprise AI conversations was that capability scaled directly with size, and that size required enormous compute resources, enormous data centres, and enormous price tags.

Qwen 3.8 breaks that assumption. On standard reasoning and language benchmarks, it matches or outperforms models that cost vastly more to build and run. It runs locally on consumer hardware. And it is free to use under an open licence.

This is not a one-off curiosity. It reflects a pattern playing out across the open-source AI space: the performance gap between free, locally-run models and expensive commercial APIs is closing fast. The practical implication for a professional services firm is significant.

What This Looks Like in Practice

Consider a senior associate at a mid-size commercial law firm who spends three to four hours a week reviewing first drafts of commercial agreements, flagging non-standard clauses, and summarising positions for clients. That is a task a well-configured local AI model can assist with today, without sending client data to an external server, without a per-query fee, and without a procurement process.

The same applies to an accountant at a 20-person firm preparing client-facing reports. Or an analyst at a financial consulting firm who spends significant time each week pulling data from multiple sources and summarising findings in consistent formats. These are not exotic automation problems. They are the ordinary, repetitive, high-volume tasks that make up a substantial portion of billable and non-billable time in most professional services firms.

The barrier was never that AI could not do these things. The barrier was that running AI at scale cost too much and introduced too much third-party risk. A locally-run open model changes both of those constraints at once.

Running a model locally means client data stays inside the firm's own environment. That matters for legal privilege, for accounting confidentiality obligations, and for any firm that has clients in regulated industries. It also means the firm is not subject to a provider changing their terms, raising prices, or deprecating the model version a workflow was built around.

What This Means for Competitive Position

Firms that deploy AI effectively in the next 12 to 18 months will create cost structures that are genuinely difficult for slower-moving competitors to replicate quickly. This is not about replacing staff. It is about changing the ratio of senior judgment to junior task completion, which changes the economics of service delivery.

A firm that can review a contract, prepare a due diligence summary, or produce a client-ready report in half the time does not just have lower costs. It has more capacity. It can take on more work, reduce turnaround times, or price more competitively in market segments where cost sensitivity is high.

Smaller firms that move early on this have a real opportunity to compete above their weight class. A 30-person accounting firm running well-configured local AI tools is not the same animal as a 30-person firm that is not.

Our Take

The conversation in professional services about AI has been dominated for too long by the assumption that serious capability requires serious spend. That assumption was defensible in 2022. It is not defensible now.

Open-source models like Qwen 3.8 are not a compromise position for firms that cannot afford the real thing. For many professional services use cases, they are the right technical choice on both cost and data governance grounds. Any firm that is holding off on AI adoption because of budget or vendor concerns should revisit that position immediately. The calculation has changed.

The firms that will look back on this period with regret are not the ones that moved too fast. They are the ones that waited until the market made the decision for them.

If you want to understand what a locally-run AI deployment would actually look like for your firm, get in touch with the ROOVOLT team and we will walk you through a specific use case.