Professional AI tools have a cost structure that differs fundamentally from the software that professional firms have adopted and managed over the past two decades, and this difference has direct consequences for how AI-assisted work should be designed, allocated, and governed at the level of the individual practitioner and the professional team.
The software that most professional firms are familiar with charges on a licence or subscription basis. A practice management system, a document management platform, a CRM, or a case management tool carries an annual or monthly fee that covers the firm's access to the system regardless of how intensively or infrequently that access is used. A firm that pays for fifty user licences can have all fifty users active simultaneously, conducting hundreds of transactions, generating thousands of documents, and running queries against the full database, without the cost of the system changing as a result of that activity. The economic model is predictable, the budget line is stable, and the incentive for individual users is straightforwardly to use the system as much as it helps them because usage carries no marginal cost.
AI tools operate on an entirely different economic model, and professional practitioners who approach AI cost with the assumptions developed from decades of licence-based software will systematically miscalculate both the economics of their AI practice and the governance requirements that those economics create. AI tools charge on a consumption basis. Every interaction with the tool consumes computing resources, specifically the processing power required to read the input the practitioner has provided, reason across it in the context of the model's training, and generate a response. The cost of each interaction is determined by the volume of text processed, both the text submitted by the practitioner and the text generated by the model in response, and by the capability tier of the model handling the interaction. Longer inputs, more detailed outputs, more complex reasoning tasks, and more capable models all produce higher costs per interaction. The billing accumulates across every interaction the practitioner conducts, across every member of the professional team using the tool, across every working day of the month.
This consumption-based structure means that the way practitioners use AI tools directly determines how much those tools cost the firm, and that disciplined professional use and economically sustainable professional use are the same thing approached from two different angles. A practitioner who submits well-constructed requests, provides precisely the context documents relevant to the specific task rather than attaching entire document sets, specifies the required output with sufficient precision to produce a usable first draft, and structures complex tasks into defined sequential steps rather than submitting them as single sprawling requests, produces professional-quality output at a cost that reflects the actual work required. A practitioner who submits vague requests that require multiple clarifying follow-ups, attaches documents in their entirety when only specific sections are relevant, or repeatedly requests revisions to outputs that could have been correctly specified in the initial instruction, generates weaker results at substantially higher cost. The economics of AI use reward exactly the disciplines that produce the best professional outcomes, and they penalise exactly the habits that produce the worst ones.
There is a further dimension to the economics of professional AI use that moves beyond the direct cost of individual interactions and into the operational economics of AI-assisted workflows at scale. When AI assistance is embedded into professional workflows that recur across a practice, the cost and time characteristics of those workflows compound across volume in ways that make the design choices consequential at a level that single-interaction economics do not reveal. A workflow that processes forty matters per month, each involving three AI-assisted steps, generates a hundred and twenty interactions per month from a single practitioner. The cost and latency profile of each of those interactions, multiplied across this volume, and then multiplied again across a team of practitioners all running similar workflows, produces the actual economic footprint of the AI practice. Decisions about which model tier to use for each workflow category, how to structure the context documents that inform each interaction, and how to sequence the steps within each workflow, have budget and throughput implications that are only fully visible at this scale of analysis.
The tradeoff between response speed and output quality adds a further economic dimension that practitioners need to understand to make sound workflow design decisions. More capable AI models, which tend to produce higher-quality outputs on complex tasks, also tend to respond more slowly because they perform more computation per interaction. Faster, more cost-efficient models respond quickly but may require more practitioner effort in review, verification, and correction for the most demanding tasks. The economically sound approach calibrates the model tier to the task rather than applying a single model across all workflow categories, using faster and more cost-efficient models for the high-volume, well-structured tasks where their performance is adequate and reserving more capable models for the complex, high-consequence tasks where the additional reasoning quality they provide reduces professional risk in ways that justify the higher cost and longer response time.
Building an AI practice that remains economically sustainable as usage grows requires practitioners to understand these economic mechanisms at the level of their own workflows rather than at the level of abstract principle. A practice that is economical at low usage volume may become expensive or behaviourally distorting at higher volume if the workflow design has not accounted for how costs compound. A practice that appears well-designed for a single practitioner may create budget pressure or governance complexity when extended across a team. Module 3.2 addresses these economic realities at the level of granularity where practitioners actually make decisions, providing the analytical framework required to design AI-assisted workflows that deliver professional value at a cost structure that supports sustained, disciplined use rather than constraining it.