A managing partner at a mid-sized law firm buys 30 Microsoft 365 Copilot licenses in January. By December, nine people use it daily, a handful use it weekly, and the rest forgot it existed by March. The firm pays for all 30 licenses anyway. This is the most common AI story in professional services right now, and it’s the gap between buying AI and getting value from it.
MIT’s Project NANDA put numbers on this last year. Of 300 enterprise AI deployments analyzed, 95% delivered little to no measurable impact on the P&L. The 5% who did weren’t using better technology. They were using it differently.
For firms past the curiosity phase, past the demos and trial seats, the question becomes “What does AI do here, for us, that we can prove?” Once you’ve worked through where AI sits in your business today, the next stage is operational maturity, which most firms find harder than the first stage.
What operational AI looks like in practice
Operational AI looks dull from the outside. A paralegal’s intake checklist runs through a Copilot prompt that pulls relevant precedent into a standard format. An accounting practice manager’s month-end reconciliation has three steps trimmed because a tool flags anomalies before they hit the partner’s desk. The flashy stuff, like a partner asking ChatGPT to draft a client memo from scratch, is usually where firms get distracted and stuck.
The shift from experimentation to operations involves narrowing the use cases. Most firms try to spread AI across everything and end up with thin adoption everywhere. The MIT research found that the firms extracting real value picked one well-defined pain point and built around it before moving to the next. Back-office and administrative workflows tend to pay back faster than client-facing ones, even though most early budget gets pointed the other way.
Measuring productivity in numbers the business respects
“Saved 40 hours a month” is the kind of metric AI vendors love and managing partners distrust. The measures that hold up are the ones tied to how the firm makes money or loses it.
For a law firm, that might be hours billed against hours worked, an improving ratio without anyone working longer. For an accounting firm, it might be time-to-close on a tax return or the reduction in outsourced back-office spend that MIT’s research highlights as one of the more reliable returns on AI investment. For pharma services or financial advisory work, it might be the speed at which a regulatory document gets through review without errors slipping through.
These metrics come from the firm’s existing operational reporting, with AI tracked as one input among several. That requires honest baseline numbers from before the deployment, which is the part most firms skip, and the reason a proper IT assessment ahead of rollout pays back.
What it looks like by role
The shape of AI use varies more than vendor marketing suggests. A managing partner uses it differently than the admin who supports them.
Partners tend to get the most value from horizon-scanning work: drafting board packs, summarizing long documents before a meeting, and reformatting analysis for different audiences. Practice managers use it for intake summarization and deadline tracking. Operations leads use it for invoice reconciliation, supplier review, and contract abstraction. Administrative staff use it for meeting notes that flow into client files, document templating, and the low-value formatting work that used to absorb whole afternoons. In every case, the workflow matched the role, not the other way around.
The quarterly review keeps the summit reachable
Mountains stay climbed only because climbers keep coming back. AI environments are the same. The capability that worked in March doesn’t work in October, partly because the tools have changed, partly because the business has.
A quarterly review of an AI environment asks five questions: What’s being used? What’s been abandoned and why? What new capabilities have shipped that fit? Where has shadow AI crept in (staff using personal ChatGPT accounts for client work)? And what needs retiring? Microsoft’s 2025 Work Trend Index found that 82% of leaders expect AI agents to expand workforce capacity in the next 12 to 18 months. That’s a moving floor. Without a regular review, the firms that moved early end up overtaken by the ones who moved later with the benefit of better tooling.
This is where a vCIO arrangement earns its keep. A part-time strategic resource that knows the firm, understands the tools, and runs the quarterly cycle so the partners don’t have to. Between reviews, that looks like tracking which seats get used and which sit dormant, watching what the major vendors have shipped, picking up on the workarounds that flag a gap in the toolset, and arriving at the quarterly meeting with a recommendation rather than a question.
What “done well” looks like 12 months in
A firm that has done this properly has three or four workflows where AI saves measurable time, with the numbers visible in the operational reporting. There’s a documented acceptable use policy that staff follow, an IT partner running the review cycle, and a license footprint that matches usage. When a new tool is released that might fit, and there will be one next quarter, the firm can evaluate it in a week instead of a year.
Reaching operational maturity takes work. Holding it through four moving quarters takes more. The firms that stay there over the long run usually have help from an IT partner who understands professional services firms and treats AI as part of the broader IT strategy, running it through the same disciplined cycle as everything else in the technology stack.
Book a consultation with BASE Solutions to map your path from AI experimentation to operational value.
