A 90-Day Blueprint for an AI-Enabled Operating Function

Scaling companies rarely suffer from a shortage of activity. Teams are busy. Meetings are frequent. New tools are introduced. Leaders communicate priorities. Functional plans are created. People work hard to respond to changing information.

Yet execution can still feel unreliable. Priorities are interpreted differently across functions. Decisions are repeatedly reopened. Important context remains inside conversations. Commitments become difficult to trace. Leaders spend increasing amounts of time reconnecting work that the organisation cannot coordinate on its own.

AI is often introduced into this environment as a productivity solution. The company adds meeting summaries, internal assistants, automated research, content generation or workflow tools. Individual tasks become faster, but the underlying operating environment remains largely unchanged. The organisation produces more information without necessarily becoming better at deciding or executing.

The more useful objective is not to make every employee more productive in isolation. It is to build an operating function that helps the company convert leadership direction into coordinated action.

This can begin within 90 days.

What is an operating function?

An operating function is not necessarily a new department. It is the connective layer through which the organisation translates direction into decisions, decisions into commitments and commitments into completed work.

In a smaller company, this responsibility may sit with the founder, Chief of Staff or head of operations. In a larger company, elements may be distributed across strategy, business operations, programme management, revenue operations, finance and functional leadership.

Regardless of organisational design, the function must answer five recurring questions: What matters now? What decision is required? Who owns the next action? What information is needed to move forward? And how will the organisation know whether the action worked?

When these questions do not have reliable answers, coordination increasingly depends on leadership intervention. Senior leaders repeatedly reconstruct context. Cross-functional work progresses through persistence rather than design.

An AI-enabled operating function should reduce that dependency. Its purpose is not to automate leadership. Its purpose is to improve the systems through which leadership attention is prepared, applied and converted into action.

The 90-day arc

The programme runs in three phases of roughly 30 days each. Each phase produces concrete outputs, and the final phase feeds what the company learns back into observation.

Days 1–30: Observe and map the operating reality

The first month should not begin with automation. It should begin with observation.

Companies frequently describe how work is supposed to happen rather than how it actually happens. Process documents show clean sequences. Responsibility charts assign clear owners. Technology systems appear to contain the required information.

The real operating environment is usually less orderly. Teams use private spreadsheets. Important decisions happen in direct messages. Meetings compensate for missing visibility. Senior people act as informal escalation paths. CRM fields are interpreted differently by different functions. Work moves because an experienced person knows whom to contact.

The first 30 days should make this operating reality visible.

1. Create an operating reality map. Identify the major recurring flows through which the company turns information into action — leadership priorities, product decisions, customer escalations, sales opportunities, hiring requests, operational incidents, budget decisions, launch planning, strategic partnerships and cross-functional initiatives. For each flow, document where the work begins, what information enters the process, who interprets it, which decisions are required, where ownership changes, which tools are involved, what creates delay, where leadership becomes involved and how completion is recorded. The purpose is not to document every process in the company. It is to identify the small number of recurring flows that consume disproportionate management attention.

2. Build a decision inventory. List the decisions that are repeatedly made, delayed, reopened or escalated. For each decision, record the decision itself, the owner, the contributors, the information required, the expected frequency, the consequence of delay, the current decision forum, whether the outcome is documented and whether similar decisions are treated consistently. This often reveals that the organisation does not have a general execution problem. It has a smaller number of unresolved decision patterns that generate repeated coordination work.

3. Audit the meeting system. Meetings provide a useful map of organisational friction. Review recurring meetings and ask: What decision or coordination need does this meeting serve? What information must be prepared beforehand? What output should exist afterwards? Is the same issue being discussed elsewhere? Does the meeting produce a decision, a commitment or only awareness? Would the meeting still be required if information and ownership were clearer? The objective is not simply to reduce the number of meetings. It is to understand which meetings are compensating for missing operating infrastructure.

4. Identify repeated management work. Look for activities that senior people perform repeatedly: assembling information from multiple systems, interpreting incomplete updates, rewriting priorities for different teams, following up on commitments, preparing decision context, resolving ambiguous ownership, summarising customer or market signals, translating between functional vocabularies and identifying issues that have become urgent. These are strong candidates for structured workflows and AI assistance. They are also evidence that important company context has not yet been institutionalised.

5. Establish baseline measures. Do not begin with an extensive measurement programme. Choose a small number of useful baseline measures, such as time from issue identification to decision, time from decision to assigned action, the percentage of strategic commitments with a clear owner and date, the number of decisions reopened because context was missing, the number of initiatives requiring executive intervention, preparation time for recurring operating meetings, the number of overdue cross-functional commitments and the proportion of meeting time spent reconstructing information. The measures do not need to be perfect. They need to make operating improvement visible.

“

The organisation is building an accurate operating map, not an automated fiction.

The role of AI during the first 30 days

AI can support the observation process without being prematurely embedded into the company's workflows. Useful applications include synthesising interview notes, clustering recurring operational problems, comparing process descriptions across teams, extracting decisions and commitments from meeting records, identifying repeated information requests, summarising patterns across customer, sales or operational data, and generating initial process maps for human review.

The output should always be validated against the lived experience of the people doing the work.

By the end of the first 30 days, the company should have an operating reality map, a decision inventory, a meeting and coordination audit, a list of repeated management tasks, baseline operating measures and a prioritised list of intervention opportunities.

The objective is shared understanding. The company should be able to explain where execution friction originates rather than describing it as a general lack of accountability or urgency.

Days 31–60: Design the operating layer

The second month turns observation into infrastructure. The goal is not to design a complete company operating system. It is to introduce a small number of mechanisms that improve how priorities, decisions, commitments and context move through the organisation.

1. Establish an operating cadence. Define the minimum set of recurring forums required to run the company: a weekly operating review focused on current performance, material changes, cross-functional commitments and emerging risks; a monthly priority review focused on whether the company's major priorities remain correct and adequately resourced; decision-specific forums used only where a recurring category of decision requires structured preparation and clear authority; and a quarterly strategic review focused on changes in the market, company position, resource allocation and strategic assumptions. Every forum should have a defined purpose, a standard input, a clear decision or output, named participants, an owner, a record of commitments and an explicit relationship to other forums. A meeting without a defined operating purpose should not become part of the permanent cadence.

2. Introduce a decision brief. Create a lightweight standard for preparing consequential decisions. A decision brief should contain the decision required, why the decision is needed now, relevant context, available options, key assumptions, expected implications, areas of uncertainty, the recommended course, the decision owner and the date by which the decision is required. The purpose is not to bureaucratise every choice. It is to reduce the amount of leadership time spent reconstructing basic context. AI can help assemble and update these briefs, but ownership must remain with the person responsible for the quality of the recommendation.

3. Create a visible commitment system. Strategic execution often breaks because decisions and commitments are stored in different places. A decision is recorded in meeting notes. The action appears in a project tool. The owner interprets the commitment differently. Leadership assumes the work is underway. A useful commitment system should make five elements visible: the commitment, the owner, the expected completion date, the current status and the evidence of completion.

4. Define escalation conditions. Escalation should not depend entirely on personal confidence or organisational status. Define the conditions under which an issue should move upward or across functions — a strategic commitment is materially off track, customer risk exceeds an agreed threshold, two functions cannot resolve ownership, required information remains unavailable, a decision has exceeded its expected timeframe, an assumption underlying the plan has changed, or an exception falls outside an established rule. Clear escalation conditions reduce both silent failure and unnecessary executive involvement.

5. Establish a context repository. The company needs a durable home for the context required to understand important work. This does not mean placing every document into one knowledge-management platform. It means ensuring that important decisions and initiatives can be understood without relying entirely on personal memory. For major initiatives, preserve the objective, relevant history, key assumptions, major decisions, responsible owners, linked evidence, unresolved questions, current status and material changes. AI assistants become substantially more useful when the company can provide reliable context rather than asking the model to infer organisational reality from disconnected files.

6. Select a small number of AI-supported workflows. Do not attempt to apply AI across the organisation simultaneously. Select two or three workflows where the work occurs frequently, information is fragmented, preparation consumes meaningful time, the expected output is reasonably clear, human review remains practical and improvement can be measured. Suitable early workflows may include preparing weekly operating reviews, assembling decision briefs, summarising customer or market signals, extracting and tracking commitments, identifying exceptions in recurring processes, producing role-specific updates from shared source information and maintaining context on strategic initiatives. The AI should initially reduce preparation and coordination work. It should not be given authority over decisions the organisation has not yet clearly defined.

Days 61–90: Implement and calibrate

The final month moves the operating design into live use. This is where apparently sensible systems meet actual organisational behaviour. The objective is not perfect adoption. It is to learn which mechanisms improve execution, which create unnecessary work and which require stronger ownership.

1. Run the operating cadence. Begin using the new forums and information standards. The weekly operating review should focus on material changes since the previous review, current company priorities, commitments requiring attention, decisions required, cross-functional dependencies, emerging risks and exceptions that do not fit the current system. Avoid turning the review into a sequence of functional status presentations. The purpose is to direct collective attention towards the issues that require coordinated action.

2. Automate preparation before automating authority. Use AI and workflow automation to reduce the preparation burden. The system might assemble data from agreed sources, summarise material changes, identify overdue commitments, compare current performance with prior expectations, draft a decision brief, surface missing information and prepare role-specific views of the same operating context. A person should remain accountable for validating the output. Preparation can be automated more quickly than authority can be delegated.

3. Introduce structured reason capture. When a recommendation, classification or proposed action is changed, capture the reason. Useful reason categories may include missing context, an incorrect assumption, a strategic exception, a customer-specific consideration, outdated source information, an ownership change, a threshold incorrectly applied or judgement required. This turns human correction into organisational learning. Without reason capture, the company repeatedly fixes individual outputs without improving the system that produced them.

4. Measure the operating effect. Compare the new operating environment with the baseline established in the first month. Relevant measures may include faster decision preparation, shorter decision cycle time, fewer reopened decisions, clearer commitment ownership, fewer overdue cross-functional actions, reduced meeting preparation, less time spent reconstructing context, fewer unnecessary executive escalations, stronger completion rates for strategic commitments and increased use of shared operating information. Do not use the number of AI-generated outputs as the primary success measure. The objective is better company execution.

5. Review adoption friction. If a new mechanism is not being used, do not immediately conclude that employees are resistant to change. Investigate whether the system duplicates existing work, the required input is unclear, the output is not useful, the workflow sits outside existing tools, ownership is ambiguous, the process creates reporting without improving action, leadership is not using the same system, or the mechanism is too complex for the value it provides. Adoption is often a design signal.

6. Calibrate and remove. At the end of 90 days, review every new operating mechanism. Keep what improves clarity and execution. Revise what is useful but cumbersome. Remove what creates activity without changing decisions or outcomes. A functioning operating system should reduce organisational burden over time. It should not create a permanent administrative layer whose purpose is to maintain itself.

By day 90, the company should have a functioning operating cadence, live decision and commitment practices, selected AI-supported workflows in regular use, baseline-to-current operating measures, documented adoption feedback, a prioritised refinement plan and clear ownership for continued operation.

“

Preparation can be automated more quickly than authority can be delegated.

A practical operating scorecard

The company should be able to assess the new operating function across five dimensions. AI should contribute to these outcomes. It should not become a separate category of activity with no relationship to company performance.

What not to do

The most common failure modes of a programme like this are predictable, and most of them involve starting with technology rather than the work.

The expected end state

After 90 days, the company should not expect to have solved every execution problem. It should expect to have a more deliberate operating layer.

Leadership direction should be translated into clearer priorities. Important decisions should arrive with stronger preparation. Commitments should be more visible. Cross-functional work should depend less on repeated personal intervention. Company context should be easier to retrieve and reuse. AI should be supporting a small number of defined workflows whose value can be observed.

Most importantly, the company should have created a mechanism for continued improvement.

A useful operating function does not eliminate uncertainty. It gives the organisation a more reliable way to recognise uncertainty, decide what it means and act accordingly.