AI for Firms
Three team members use ChatGPT for different tasks. The tax senior swears by an AI research tool. The bookkeeping manager tried two categorization tools last quarter. The founder calls this "our AI initiative." But nobody has documented what works, nobody has defined graduation criteria, and none of these experiments have changed how the firm actually operates.
Trying AI tools is not the same as building operational AI capability. Experimentation creates enthusiasm; strategy creates operating change. Most accounting firms are stuck in experimentation mode — individual team members trying tools based on personal curiosity, with no framework for evaluating results, graduating successful experiments, or scaling validated approaches across the firm. Without a bridge from experiment to integration, AI remains a collection of side projects that never become firm capability.
Why firms with active AI experimentation see no operational improvement — and why the gap between trying tools and building capability is structural, not motivational.
Founders, managing partners, and operations leaders in accounting firms who want to move from scattered AI experimentation to systematic AI capability.
Every month spent in experimentation mode without a strategy is a month where the firm generates AI enthusiasm without building AI capability — and competitors with strategy pull further ahead.
The founder surveys the team about AI usage. The results are encouraging on the surface: eight of twelve team members report using AI tools regularly. Three use ChatGPT for drafting client communications. Two use an AI-powered research tool for tax questions. One uses a transcription tool for client meeting notes. The bookkeeping manager has tried three different categorization tools and has opinions about all of them.
The founder presents this at the partners meeting as evidence that the firm is "embracing AI." The partners nod approvingly. The firm appears to be ahead of its peers in AI adoption.
But a closer examination reveals that none of this experimentation has changed the firm's operating model. The client communication process is the same — some people draft with AI, some do not, and nobody has evaluated which produces better results. The tax research workflow is unchanged — the AI tool supplements individual research but has not been integrated into how the team collectively approaches research questions. The bookkeeping categorization experiments produced opinions but no decision about which tool to standardize, how to integrate it, or what success would look like.
The firm has AI activity but no AI capability. Individual team members have personal AI habits but the firm has no AI operating model. The experimentation is real. The progress is illusory.
The root cause is the absence of a bridge between experimentation and integration. The firm has no framework that answers: How does an experiment become a standard? What criteria determine whether a tool is worth adopting firm-wide? Who decides? Who implements the transition from "some people use this" to "this is how we operate"?
Without this framework, experimentation is self-contained. Each team member's AI usage exists in its own bubble — generating personal value for the individual but no organizational value for the firm. The experiments do not build on each other. The results are not captured. The learnings are not shared systematically. The firm as an organization does not learn from its members' individual experiments.
This is the AI-specific manifestation of the broader pattern where workflow improvement fails without change discipline. The firm can identify better ways of working (through experimentation) but cannot implement them at scale because it lacks the organizational mechanism to move from "this works for me" to "this is how we do it." The enthusiasm is genuine. The change management infrastructure is absent.
The result is permanent experimentation — a state that feels innovative but produces no operational change. The firm invests time and subscription costs in experimentation that never graduates to integration. Each quarter looks like the last: individual AI usage grows, organizational AI capability does not, and the firm's operating model remains unchanged despite genuine effort from motivated team members.
The first pattern is experimentation that lives in individual team members rather than in the firm's knowledge base. When a senior discovers that an AI research tool is effective for a specific type of tax question, that knowledge stays with the senior. It is not documented, evaluated, or shared in a structured way. Another team member may independently discover the same thing — or may try a different tool and reach a different conclusion. The firm's AI knowledge is scattered across individuals with no mechanism for aggregation.
This pattern persists because documentation feels like overhead when you are experimenting. Nobody wants to write up their AI experiments when they are still trying things. But without documentation, the experiments are invisible to the organization. The firm cannot learn from what it does not know about, and it does not know about the experiments that individual team members are conducting.
The second pattern is the absence of defined criteria for when an experiment becomes a standard. The bookkeeping manager has been "trying" a categorization tool for four months. She has opinions about it. But nobody has defined: what would this tool need to demonstrate before we adopt it firm-wide? What accuracy rate is sufficient? What time savings justify the cost? What integration requirements must be met?
Without graduation criteria, experiments run indefinitely. The tool is always "being evaluated" and never "adopted" or "rejected." The firm accumulates experiments in various stages of informal trial without ever making a decision about any of them. This indecision is comfortable — it avoids the organizational effort of adoption — but it also means the firm never builds capability from its experimentation.
The third pattern is measuring AI progress by team sentiment rather than operational metrics. "The team is excited about AI" becomes the success metric, substituting for "AI has improved our operational performance by a measurable amount." Enthusiasm is valuable — it creates willingness to adopt — but it is not progress. Progress is measurable change in how the firm operates.
This conflation is reinforced by industry conversations where firms compare AI activity rather than AI outcomes. "We are using three AI tools" sounds more advanced than "We are not using AI yet" — but if the three tools have not changed the firm's operating model, the practical difference is zero. The firm with no AI tools but a clear strategy may be closer to operational AI capability than the firm with multiple tools and no strategy. The same principle applies to process documentation: having documents is not the same as having documented processes that actually guide work.
The client experiences no change. The firm's internal AI experimentation is invisible to the client because it has not changed how work is delivered. The client receives the same service, at the same speed, with the same quality. If the firm mentions its AI initiatives in a business development context, the client may form expectations that the firm's actual capability cannot meet — creating a credibility gap between what the firm claims and what it delivers.
Conversely, firms that have moved from experimentation to strategy — integrating AI into their actual delivery model — can demonstrate tangible improvements: faster turnaround, more consistent quality, proactive insights that were not possible before. The client experiences the difference between a firm that talks about AI and a firm that delivers with AI.
The most common misdiagnosis is that the firm is "being innovative" by encouraging experimentation. Innovation requires a path from idea to implementation. Experimentation without that path is exploration — which has value but is not innovation until it produces operational change. The firm congratulates itself on innovation when it is actually stuck in exploration.
The second misdiagnosis is that more experimentation will eventually produce a breakthrough. "If we try enough tools, we will find the right one." But the constraint is not finding the right tool. The constraint is having a framework for evaluating, adopting, and integrating any tool. A firm with a strong adoption framework will succeed with a mediocre tool. A firm without a framework will fail to adopt even a perfect tool.
The third misdiagnosis is that the team needs a champion. "If we had one person dedicated to AI, we would make progress." A champion can accelerate experimentation, but without organizational authority to standardize and integrate, the champion becomes another individual experimenter — just with more visibility. Strategy requires leadership authority, not individual enthusiasm.
Firms that build AI capability rather than AI activity define a structured adoption pathway.
They experiment with purpose. Each experiment has a defined hypothesis, a specific workflow it targets, and a success criterion. "We are testing whether Tool X can reduce standard return preparation time by 30 percent over a two-month period using ten engagements." This is an experiment. "Try this tool and see if you like it" is not.
They capture and share results systematically. Experiment results — what worked, what did not, what was learned — are documented in a shared format that the firm can reference. When someone asks "have we tried AI for categorization?" the answer is a documented evaluation, not a conversation with whichever team member happened to try something.
They define graduation criteria before experimenting. Before the experiment begins, the firm defines what would need to be true for the tool to be adopted firm-wide. Success criteria, integration requirements, cost thresholds, and implementation timeline are established upfront. The experiment then produces evidence for or against adoption — not an open-ended "we are still evaluating."
They own strategy at the leadership level. Individual team members own experiments. Leadership owns the strategy that evaluates experiments, approves graduation, allocates integration resources, and ensures that validated approaches become operating standards. Without leadership ownership, AI strategy remains a collection of bottom-up experiments with no top-down integration authority.
Experimentation is the beginning of AI capability, not the destination. Without a strategy that bridges experimentation to integration, firms generate AI activity without building AI capability — and the gap between the two widens with every month of undirected experimentation.
The strategic discipline is to build the bridge: define an adoption pathway with clear stages (experiment, evaluate, standardize, integrate), assign leadership ownership, create graduation criteria, and measure progress by operational outcomes rather than tool count or team enthusiasm. This framework transforms experimentation from an end in itself into a stage in a deliberate capability-building process.
Firms working with Mayank Wadhera through DigiComply Solutions Private Limited or, where relevant, CA4CPA Global LLC, typically build AI adoption frameworks that connect experimentation to integration — defining evaluation criteria, graduation pathways, and implementation plans that turn validated experiments into operational capability. The goal is not to stop experimentation but to give it purpose and direction — because the firms that win with AI are the ones that move from trying tools to building capability.
AI experimentation generates enthusiasm. AI strategy generates operating change. Without a bridge between the two, experimentation becomes a permanent state that feels innovative but produces no capability.
Measuring AI progress by tool count and team enthusiasm rather than by operational metrics that demonstrate measurable improvement in how the firm delivers work.
They define adoption pathways: experiment with purpose, evaluate against criteria, graduate successful experiments, and integrate validated approaches into the firm's operating model.
If the firm's AI experimentation has not changed how work actually moves through the firm, it is activity without strategy. Build the bridge from experiment to integration.
AI experimentation is individual team members trying tools to see what works. AI strategy is a firm-wide framework that defines which workflows AI should address, what success looks like, how experiments graduate to operational use, and how the firm builds AI capability systematically. Experimentation creates enthusiasm. Strategy creates operating change.
Because experimentation is easy and strategy is hard. Trying a new AI tool requires only a subscription and curiosity. Building an AI strategy requires workflow mapping, success criteria, graduation pathways, and organizational change management. Most firms take the easier path and stay in experimentation indefinitely — generating excitement without building capability.
Through a defined pathway: experiment with the tool on a specific workflow, measure results against defined success criteria, document what works and what does not, standardize the validated approach, train the broader team, and integrate the tool into the firm's operating model. Each step has a defined gate — criteria that must be met before proceeding to the next stage.
Experimentation generates useful information — which tools show promise, which workflows might benefit, what the team responds to. But without a strategy to capture and act on that information, the learning is lost. Individual team members know what worked for them, but the firm does not systematically capture, evaluate, or scale those findings. The experimentation has value. The value is wasted without strategy.
Key signs include: multiple team members using different AI tools for similar tasks, no documentation of what has been tried and what results were achieved, no defined criteria for when an experiment becomes operational, AI usage that varies entirely by individual preference, and enthusiasm about AI that does not translate to measurable operational improvement.
AI strategy should be owned at the leadership level — typically the managing partner or COO — because it requires decisions about workflow priorities, resource allocation, and organizational change. Individual team members can own experiments, but the framework that evaluates, graduates, and scales successful experiments must be a leadership responsibility. Without leadership ownership, AI remains a collection of individual initiatives rather than a firm capability.
The timeline depends on the workflow, but the structure matters more than the duration. Define what you are measuring before the experiment starts. Run the experiment for enough cycles to produce meaningful data — typically one to three months for most accounting workflows. Evaluate against the predefined criteria. If the criteria are met, proceed to integration. If not, adjust or abandon. The key is having criteria, not having a specific duration.