The Case for a Scheduled, Periodic Practice
The classification framework described in Section 2 and the source evaluation discipline described in Section 4 provide the practitioner with the tools to manage their engagement with the AI information environment on a continuous basis, sorting developments by their professional relevance as they encounter them, applying appropriate scepticism to different source categories, and identifying when a development meets the criteria for Tier One or Tier Two attention. What neither of these tools provides is a structured mechanism for acting on the accumulation of Tier Two developments the practitioner has noted over time, assessing whether the AI landscape has moved in ways that warrant changes to the established practice, and conducting the integration performance review that Module 4.3 established as a regular governance requirement.
This structured mechanism is the quarterly review. It is a scheduled, time-bounded practice that brings together the full range of currency management activities that the practitioner's AI practice requires, conducted at a frequency that is sufficient to catch professionally significant developments before they create problems while not consuming the professional time that AI assistance is intended to free. The case for quarterly rather than monthly or annual review rests on a specific analysis of the pace at which the AI landscape moves, the pace at which professional obligations and integration configurations evolve, and the minimum frequency at which deliberate, comprehensive practice assessment produces value rather than merely reassurance.
A monthly review would be more responsive to fast-moving developments but would consume disproportionate professional time relative to the volume of professionally significant development that actually occurs at that frequency. Most months do not produce Tier One developments for any specific practitioner: the regulatory frameworks relevant to their domain do not publish new guidance every month, the AI tools they are using do not revise their data handling terms every month, and the capability improvements of available AI models do not cross the professional relevance threshold that would justify practice changes every month. A monthly review conducted in the absence of substantive developments to address is a review that produces no professional value while consuming professional time.
An annual review, by contrast, risks allowing Tier Two developments to accumulate without assessment for long enough that a development that had been moving from Tier Two toward Tier One for several months creates professional risk before the review identifies and responds to it. The twelve-month period is also too long for integration performance review to function as effective quality assurance, since an integration that has been performing unreliably for six months before the annual review identifies the problem has been producing unreliable professional outputs and undermining the practitioner's confidence in AI assistance for a period that good governance practice should have prevented.
The quarterly interval is specifically calibrated to avoid both of these problems. It is frequent enough that Tier Two developments can be assessed for their cumulative implications within a timeframe short enough to prevent them from creating professional risk before they are addressed. It is infrequent enough that each review is likely to find substantive matters to address, making the investment of review time consistently productive. It aligns naturally with the financial, reporting, and professional planning cycles that structure many professional environments, making it straightforward to schedule as a fixed commitment within the practitioner's existing professional calendar.
Scheduling and Protecting the Review
The quarterly review is most effective when it is scheduled as a fixed commitment in advance rather than arranged when convenient or when a specific trigger suggests the time has come. The practitioner who schedules the review on an ad hoc basis when they feel sufficiently informed to make it worthwhile is applying a criterion that is poorly calibrated to when the review is actually needed. The practitioner who feels most comprehensively informed about the AI landscape is often the practitioner who has been most recently and intensively engaged with the AI information environment, which is not the same as the practitioner whose practice most needs review. Conversely, the practitioner who has been focused on other professional priorities and has had limited engagement with the AI information environment for an extended period is precisely the practitioner whose practice most needs the structured assessment that a quarterly review provides.
The review should be scheduled at a specific time and protected from displacement by other professional demands with the same discipline that a client commitment or a regulatory deadline receives. The practical challenge to this discipline is that the quarterly review, unlike a client meeting or a filing deadline, does not carry an external accountability mechanism that makes its displacement professionally costly in the immediate term. The practitioner who cancels or defers a client meeting faces an immediate professional consequence. The practitioner who cancels or defers the quarterly AI practice review faces no immediate consequence, since the professional risk created by the practice assessment that did not occur accumulates gradually rather than presenting itself immediately. This asymmetry means that the quarterly review is systematically vulnerable to displacement by more urgent-seeming demands unless it is treated with a discipline that is self-imposed rather than externally enforced.
The mechanism that supports this discipline is the explicit recognition, at the point of scheduling, that the quarterly review is a professional governance activity rather than an optional professional development exercise. The practitioner who understands that their AI practice involves ongoing data handling decisions, integration configurations that require periodic governance review, and regulatory compliance obligations that evolve over time, and who understands that the quarterly review is the mechanism through which these governance obligations are discharged, is in a position to treat the review's protection from displacement as a professional obligation rather than a personal preference.
The time commitment for the quarterly review is bounded at approximately two hours. This boundary is deliberate and important. A review process that regularly exceeds two hours is a review process that has either expanded beyond its appropriate scope or that reflects a practice state requiring more intensive remediation than a standard quarterly review is designed to provide. A review that consistently requires more than two hours to complete suggests that the practitioner's practice has accumulated more complexity than their quarterly maintenance discipline can address efficiently, which is itself a finding that the review should surface and that should prompt simplification of the practice rather than extension of the review commitment.
Within the two-hour boundary, the review covers four domains, addressed in the sequence described in the following sections. Each domain has its own defined scope and its own output, and the two-hour budget should be allocated approximately across the four domains rather than allowing any single domain to expand without limit. A rough allocation of thirty minutes per domain provides sufficient time for substantive assessment of each without allowing the overall review to exceed the two-hour boundary. The allocation need not be rigidly equal, since some reviews will find more substantive matters in one domain than others, and the time allocation should flex accordingly while the overall boundary is maintained.
Domain One: Integration Performance Assessment
The first domain of the quarterly review is the assessment of how the practitioner's current AI integrations are performing against the expectations that justified their construction. This domain is addressed first because integration performance problems are the category of AI practice issues most likely to have been creating professional risk or reducing professional efficiency between reviews, and addressing them is therefore the most urgent dimension of the review.
The integration performance assessment draws on the practitioner's accumulated experience of using each current integration over the preceding quarter. For each active integration, the practitioner should assess three questions. The first is whether the integration has been performing reliably: whether the outputs it produces have been consistently of the quality and accuracy required for professional use, whether the integration has functioned without significant technical failures, and whether the verification steps applied to the integration's outputs have revealed error rates within the acceptable range for the task type. An integration whose error rate in professional use has been higher than expected, or whose technical reliability has been inconsistent, is not delivering the value that justified its construction and requires either reconfiguration or replacement with a manual workflow.
The second question is whether the integration's access scope remains appropriate given any changes in the practitioner's role, the nature of the matters they have been handling, or the regulatory or data handling guidance applicable to the information the integration accesses. An integration's access scope is set at the time of its construction based on the practitioner's assessment at that point of what access is necessary and appropriate. Circumstances change in several ways that may affect the appropriateness of an integration's access scope. The practitioner may have taken on work in a new practice area involving more sensitive information than the integration's current scope contemplated, the AI tool's data handling terms may have been revised in ways that affect the compliance status of the current access scope, or new regulatory guidance may have been published that changes the assessment of what access scope is permissible for the type of information involved. The quarterly review is the mechanism through which these changes are identified and the integration's access scope is adjusted to remain within appropriate boundaries.
The third question is whether the integration is still earning its maintenance commitment.As established in Module 4.3, integrations carry ongoing maintenance costs, with access tokens requiring renewal, API changes requiring reconfiguration, data handling terms requiring monitoring, and the ongoing assessment of whether the integration remains compliant with evolving regulatory guidance requiring professional attention. The integration whose maintenance commitment exceeds the professional value it delivers is an integration that should be decommissioned, and the quarterly review is the appropriate moment to make this assessment without the inertia that often causes underperforming integrations to persist beyond the point at which they should have been retired.
The output of the integration performance assessment is a specific decision for each active integration, whether to continue without change, reconfigure to address an identified performance or compliance issue, reduce access scope in response to a changed practice or regulatory context, or decommission because the maintenance commitment is no longer justified by the value delivered. Where reconfiguration or decommissioning is indicated, the review should produce a specific plan for the action required and a timeline for its completion.
Domain Two: Model Landscape Developments
The second domain of the quarterly review is the assessment of model landscape developments that have accumulated since the previous review, with specific attention to whether any Tier Two development has progressed to the point where it meets the Tier One criteria and requires action, and whether any development has reached the capability threshold that would make a new AI-assisted workflow viable for the practitioner's professional context.
This domain draws on the Tier Two developments the practitioner has noted during the preceding quarter using the classification framework from Section 2. The practitioner should review each noted Tier Two development and assess, in light of the quarter's subsequent developments, whether the development has progressed in a direction that changes its classification. A Tier Two development representing a capability improvement in a specific task type, noted at the beginning of the quarter as warranting awareness, may have progressed during the quarter to the point where independent professional assessments have confirmed that the capability improvement is reliable in professional use conditions, a progression that would support an assessment that the development now meets the threshold for evaluation as a potential new workflow component. Alternatively, a Tier Two development representing a regulatory trajectory may have crystallised into a published guidance document that now meets the Tier One criteria and requires a specific compliance response.
The model landscape assessment in the quarterly review is also the appropriate moment to apply the model selection framework from Module 4.2 to any new AI tools that the practitioner has identified as potentially relevant during the quarter. The practitioner who noted during the quarter that a new AI tool with specific capabilities relevant to their professional domain had been released is in a position, at the quarterly review, to apply the four-dimension selection framework to assess whether the tool warrants further investigation. This assessment requires the practitioner to apply the classification criteria to determine whether the tool appears to meet the threshold for Tier Two or Tier One attention, and to decide whether to invest in further evaluation before the next quarterly review, without demanding a full depth evaluation at the review stage.
The output of the model landscape assessment is a specific decision for each noted Tier Two development: continue monitoring at Tier Two, reclassify to Tier One and identify the specific action required, or reclassify to Tier Three because subsequent developments have clarified that the development does not meet the criteria for professional relevance. For any new AI tool identified as potentially relevant, the output is a decision about whether to conduct a fuller evaluation before the next review and, if so, a specific allocation of time for that evaluation.
Domain Three: Regulatory and Compliance Developments
The third domain of the quarterly review is the assessment of regulatory and compliance developments that affect the practitioner's AI practice. This domain is the most critical from a professional risk perspective because regulatory non-compliance in AI practice creates specific professional, legal, and reputational consequences that make it the highest-priority category of Tier One development.
The regulatory assessment should cover three areas. The first is developments in the data protection regulatory environment: new guidance from national data protection authorities on the application of GDPR to AI processing, enforcement decisions that clarify how existing obligations apply to specific AI practices, and any consultations or draft guidance that indicate the direction of regulatory development in areas relevant to the practitioner's practice. The practitioner should assess whether any development in this area requires them to review the data handling terms of their current AI tools, to revise the information they submit to AI tools in light of new guidance on what processing is permissible, or to seek specialist input to assess whether their current practice satisfies the standard the new guidance articulates.
The second area is developments in the implementation of the European AI Act and its associated guidance. The Act's implementation is a multi-year process that will produce a progressive stream of more specific guidance and regulatory decisions as its risk classification framework is applied to specific AI system categories and use cases. The quarterly review should assess whether any development in this implementation has direct implications for the practitioner's specific AI practice, particularly where the practitioner's AI use involves higher-risk applications in regulated professional contexts.
The third area is sector-specific regulatory developments from the professional and supervisory bodies that govern the practitioner's domain. Legal professional regulatory bodies, insurance supervisory authorities, financial services regulators, and the professional standards bodies that govern consulting practice are all engaged to varying degrees with the implications of AI use in their regulated domains, and their publications on AI governance and professional obligations are among the most directly applicable sources of regulatory intelligence for practitioners in those domains. The quarterly review should include a check of recent publications from the relevant bodies to identify any statement, guidance, or consultation that affects the practitioner's AI practice.
The output of the regulatory assessment is a specific action for each identified development: no action required, review specific aspect of current practice against new guidance, seek specialist input to assess compliance implications, or revise specific practice elements in response to clear regulatory guidance. Where specialist input is required, the quarterly review output should include a specific instruction to seek that input, including identification of the appropriate specialist and a timeline for doing so.
Domain Four: Capability Threshold Assessment
The fourth domain of the quarterly review is the assessment of whether any AI capability that the practitioner has been monitoring at the Tier Two level has reached the threshold of reliability and professional relevance that would justify its incorporation into the established practice. This domain is the most forward-looking of the four and is the domain most directly connected to the positive dimensions of the production-to-judgment shift: the progressive expansion of the execution tasks that AI assistance can reliably support, and the corresponding progressive recovery of practitioner capacity for the judgment work that the shift makes more important.
The capability threshold assessment requires the practitioner to have a clear prior view of what the threshold is for each capability they are monitoring. A capability that the practitioner has been unable to use in their AI practice because AI tools have not been sufficiently reliable for the task type in question has a threshold defined by the reliability standard that the task type requires for professional use. A capability that the practitioner has been monitoring because it might enable a new workflow that would recover time from a currently manual execution task has a threshold defined by the combination of reliability, cost, and data handling compliance that the workflow would require.
The assessment of whether a monitored capability has crossed the practitioner's pre-defined threshold is an empirical question that requires evidence rather than impression. The practitioner should assess the evidence available about the capability's performance in conditions comparable to their professional use case: independent professional assessments of reliability in the relevant task type, peer practitioner experience with the capability in comparable professional contexts, and where available, the practitioner's own preliminary evaluation of the capability using the test-with-non-sensitive-data approach established in Module 4.3. Where the evidence is sufficient and the threshold appears to have been crossed, the output is a decision to evaluate the capability for incorporation into the practice through the careful, sequential approach that Module 4.3 describes. Where the evidence is insufficient or the threshold has not been crossed, the output is a decision to continue monitoring at the Tier Two level and reassess at the next quarterly review.
The capability threshold assessment is also the appropriate moment to assess whether any capability that the practitioner has already incorporated into their practice should be expanded. An AI-assisted workflow that has been operating reliably for two or three quarters, delivering consistent performance within acceptable error rates and without significant integration problems, may be a candidate for expansion to additional task types or a higher proportion of the practitioner's work in the relevant task category. The quarterly review is the appropriate moment to make this assessment deliberately rather than allowing the practice to expand informally through the gradual lowering of verification standards that sometimes accompanies familiarity with an AI tool's capabilities.
Acting on the Review's Outputs
The quarterly review produces value only when its outputs are acted on, and acting on them requires that the review produces outputs that are specific enough to be acted on rather than general impressions that do not translate into concrete professional decisions. The discipline of the quarterly review is therefore not only the discipline of conducting the review but the discipline of producing outputs that are actionable and then following through on the actions they specify.
Each domain of the review should produce a specific set of decisions, as described in the preceding sections. This assessment requires the practitioner to:
- Apply the classification criteria to determine whether the tool appears to meet the threshold for Tier Two or Tier One attention
- Decide whether to invest in further evaluation before the next quarterly review
- Stop short of a full depth evaluation at the review stage This action list should be recorded in a form that is accessible to the practitioner between quarterly reviews, so that the actions committed during the review can be tracked and completed within the specified timelines rather than being forgotten as the demands of daily professional practice reclaim the practitioner's attention.
The actions that a quarterly review produces will typically fall into one of three categories:
- Actions that can be completed immediately or within a short period without specialist input
- Actions that require specialist input before they can be completed
- Actions that require a fuller investigation or evaluation before a decision can be finalised The first category should be acted on within a week of the review. The second category requires the practitioner to initiate the specialist engagement promptly and to set a deadline for receiving the specialist input that is consistent with the professional risk profile of the issue. The third category requires the practitioner to schedule the investigation or evaluation within a defined period and to ensure that it is completed and its findings assessed before the next quarterly review.
The quarterly review that produces no actions is evidence that the review's scope was too narrow, its standards were too low, or the practitioner's assessment of their current practice was too optimistic, rather than evidence that the AI practice is in excellent condition.A practice that is operating effectively and is well-calibrated to the current AI landscape and the current regulatory environment will still produce actions in each quarterly review, because the landscape moves even when the practitioner's response to it is timely and well-calibrated. The practitioner who regularly completes quarterly reviews without producing any actions should consider whether their review is examining the right dimensions of their practice with sufficient rigour rather than concluding that their practice requires no attention.
The Review as a Professional Discipline
The quarterly review rhythm is most productively understood not as an administrative maintenance activity but as a professional discipline: a regular, structured engagement with the quality and currency of the practitioner's AI practice that reflects the same commitment to professional standards that governs their engagement with the substantive professional work their clients and counterparties rely on. A practitioner who applies rigorous professional standards to the advice they give and the documents they produce, but who does not apply equivalent rigour to the governance of the AI tools and workflows that contribute to producing that advice and those documents, is applying professional discipline inconsistently.
The production-to-judgment shift described in Module 5.1 is making the practitioner's AI practice an increasingly important component of their professional work. As that importance grows, the governance discipline that surrounds the practice should grow proportionately. The quarterly review is the primary mechanism through which that governance discipline is exercised, and the practitioner who maintains it consistently is also the practitioner who is best positioned to expand their AI practice with confidence, to identify and address problems before they create professional risk, and to incorporate new capabilities as they reach the reliability threshold that professional use requires. The quarterly review is, in this sense, not only a currency management practice. It is a professional development practice that keeps the practitioner's AI capability growing in a direction that is controlled, compliant, and continuously aligned with the professional standards that their work requires.