AI Operations

Manage your AI team, review business health, and trace decisions, approvals, workflows, and outcomes

5 min. Updated 2026-09-30.

Direct and Watch Your AI Team

AI Operations is the daily workspace for directing AI employees and understanding their work. Select an employee to see messages, completed tasks, live progress, visible specialist consultations, and review requests in one time-ordered conversation. Use the composer below the feed to continue talking. Important client-facing and financial actions pause for your approval.

  1. Select an employee

    In Agents, choose the Account Manager, Delivery Planner, Proposal Strategist, Client Growth Manager, or Business Analyst from the employee list.

  2. Choose a task

    Use the Conversation selector to resume a task, or choose New conversation for a separate thread. Select Load older conversations to find earlier threads. Task messages keep their animated progress and completion status.

  3. Watch the work progress

    Expand Work progress inside a task message to see completed, working, blocked, failed, and approval-waiting steps. The task summary describes the business result without exposing raw prompts or internal model reasoning.

  4. Talk to the employee

    Use the composer at the bottom of the same conversation. The selected employee is already addressed. Ask questions or describe changes in your own words, including corrections to earlier replies. Proposal pricing changes apply to the current proposal preview and preserve other choices. Expand Internal pricing breakdown to inspect the calculation. Confirm the latest review before creating the unsent draft. Replies stay within this thread, and Load older activity retrieves earlier messages and task updates.

  5. Review proposed actions

    Choose Confirm, Reject, or Cancel on proposed chat actions. Approval applies only to the exact proposal draft, invoice draft, concession, record change, or message shown. For an email draft awaiting review, select Review draft here to inspect and edit the message before approving and sending or rejecting it.

  6. Follow specialist consultations

    The Account Manager can consult the Delivery Planner, Proposal Strategist, Business Analyst, or Client Growth Manager while keeping responsibility for the opportunity. Each read-only consultation appears in the task history with its request, evidence, assumptions, risks, and recommendation.

  7. Review project email actions

    From a project page, ask the Delivery Planner to review recent project email. It can propose adding explicitly requested organization contacts or removing one matching service from an untouched draft Smart File. Confirm the exact access or service change before approval. Sent, signed, versioned, or paid documents require a revision instead.

  8. Manage capabilities

    Select Manage employee to review the employee canvas and control optional skills. Live task activity stays in AI Operations while capability settings remain in the management view.

Tips

  • A moving step represents persisted task state. SupaBook does not display decorative progress when no work is running.
  • A specialist consultation provides advice only. It cannot change records, contact a client, approve another action, or expand an employee's authority.
  • A handoff does not expand either employee's authority. Approval and skill settings still apply after work changes hands.
  • Resolving one proposed action does not automatically resolve other actions in the same task.
  • If an approved action fails temporarily, retrying uses the same reviewed action. A changed scope, price, recipient, concession, or total requires a new approval.
  • The Activity tab remains the detailed audit timeline for decisions, drafts, deliveries, workflows, knowledge use, and AI requests.
  • Nothing sends, charges, discounts, waives a fee, or marks an invoice paid merely because a task reaches the approval step.

Manage the AI Account Manager

The AI Account Manager coordinates the work around an opportunity, not just the email. It can request visible, read-only advice from specialist employees while remaining responsible for the client objective. Qualify and respond to new leads researches and evaluates each inquiry. Enforce opportunity standards blocks proposals when required checks or approvals are incomplete. Prepare discovery calls builds a briefing for confirmed scheduler bookings linked to leads. Monitor opportunities watches next actions, deadlines, and recent stalled leads.

  1. Review active skills

    In AI Operations, select the Account Manager and choose Manage employee to review which skills are active and what each skill does.

  2. Control each skill

    Turn a skill on or off without changing the other Account Manager skills.

  3. Set opportunity standards

    Open the Account Manager canvas, then select Opportunity standards to edit qualification values, the current operating constraint, ideal-client profiles, deposit and revision rules, discovery requirements, out-of-service-fit declines, and approval thresholds. When out-of-service-fit declines are enabled, clearly unsupported requests use the active Project Not a Good Fit email template and remain pending approval.

  4. Review the workflow

    Open the canvas to see how new inquiries are researched, normalized, evaluated, routed, and held for approval.

  5. Prepare for a booked call

    Keep Prepare discovery calls active to automatically build company context, recent CRM history, a call goal, questions, missing details, risks, and opportunities for linked bookings.

  6. Monitor active opportunities

    Keep Monitor opportunities active to receive an alert when a next action is overdue, a project date is approaching, or a recently active opportunity has stalled. Open the lead's Opportunity tab to see the evidence and recommended next step.

  7. Record the outcome

    On a lead, open Why this suggestion and select Record outcome. Confirm or correct the recommendation, then add known discovery, proposal, project value, margin, and client-quality results.

  8. Review decision quality

    Open the Quality tab in AI Operations to compare recommendation confidence, evidence completeness, human overrides, and known commercial outcomes.

Tips

  • Qualify and respond to new leads is the research and evaluation step. A prepared email is one output of that analysis.
  • Enforce opportunity standards is always active because proposal locks and approval gates are workflow rules, not optional writing behavior.
  • Opportunity standards are saved for the current account. Updating them does not change another studio's Account Manager.
  • Operating constraint affects capacity routing only. It does not improve service, budget, or reputation fit.
  • Ideal-client profiles can strengthen strategic fit when lead evidence matches them, but they do not bypass discovery or approval requirements.
  • Outcome feedback updates quality reporting. It does not change account policy automatically.
  • The Account Manager uses approved studio knowledge and account-specific instructions when deciding what to recommend.
  • Recommendations for discounts, waived fees, free work, changed terms, or client promises remain proposed concessions until a person approves the exact change.
  • Open a linked booking in Scheduling to read its Discovery brief.
  • Opportunity monitoring does not email leads, change statuses, or complete actions automatically.
  • Nothing is sent from this approval-first workflow until a person reviews it.

Plan Delivery and Build Proposals

The AI Delivery Planner checks how approved work can be delivered responsibly, including resources, effort, schedule, logistics, assumptions, and operational risks. It does not set margins, discounts, or the final selling price. The Account Manager can request this read-only consultation before the AI Proposal Strategist combines the plan with the qualified opportunity, validates commercial policy, and prepares a proposal draft for human review. Studios can use account-specific titles such as AI Producer and AI Sales Strategist.

  1. Qualify the opportunity

    The Account Manager confirms fit, objectives, required discovery, scope readiness, and proposal permission.

  2. Record the completed discovery

    On the lead, open Discovery and record the confirmed objective, audience, scope, investment, timeline, usage, logistics, stakeholders, approvals, risks, and unresolved questions.

  3. Reassess proposal readiness

    Select Save and reassess. The Account Manager uses saved scope and deliverables as evidence. Follow-ups are grouped into client confirmations, venue confirmations, and internal Delivery Planner and Proposal Strategist reviews. Internal costing and capacity checks are not questions for the client.

  4. Draft from discovery

    When the lead shows Ready for proposal, open Proposal Strategist and ask for a draft. You can discuss wording or pricing first, or prepare the unsent Smart File directly from saved discovery.

  5. Review the readiness result

    Proposal Strategist checks the latest assessment, discovery status, default proposal template, and saved pricing quantities before showing the draft action. If something is missing, use the offered link to correct it, then ask for the draft again.

  6. Review the draft

    Open the unsent Smart File to review the filled template, service quantities, days, and calculated package prices. Ask for changes in plain language. Saved reference rates remain flagged for owner confirmation.

  7. Create the review draft

    Select Prepare proposal draft. SupaBook checks discovery, fills the existing proposal template, and saves editable Service Template quantities and prices. Generation runs in the background. Nothing is sent automatically.

  8. Complete high-value review

    When an option reaches an account approval threshold, the review-only draft flags owner or producer approval. A person must resolve it before presenting or sending the proposal.

  9. Review before presentation

    A person reviews scope, pricing, exceptions, and commitments before the Account Manager presents or sends the proposal.

Tips

  • The Delivery Planner does not choose the final selling price.
  • The Proposal Strategist cannot invent rates or bypass the pricing engine.
  • A confirmed proposal action keeps one plan for the lead, template, pricing source, offer count, and revision target through every generation step.
  • Temporary system failures can be retried from the same action card. Discovery, template, pricing, budget, or approval blockers must be corrected before Proposal Strategist prepares a fresh action.
  • When a failed card requires a new action, reply try again. A yes reply to Proposal Strategist's immediate preparation question also creates the fresh readiness-checked proposal action for the same lead.
  • Free resources, consultations, discounts, waived setup or equipment fees, scope alternatives, and changed payment terms are recommendations until a person approves the exact concession. Nothing is applied or promised automatically.
  • Standard and multi-option proposals use the same planning and strategy gates.
  • For three-option proposals, Budget Fit shows the leanest responsible scope against the budget target. A firm maximum stays a limit. Recommended shows the strongest responsible solution and may exceed the stated budget when its documented workload requires it. Full Scope shows the calculated cost of the broader requested outcome.
  • When Service Templates contain private pricing rows, their saved resources, quantities, durations, and estimate sections become the proposal pricing library. The Proposal Strategist selects only what the verified project needs, then server-side math calculates each item before contingency, overhead, margin, and licensing.
  • The service pricing editor groups selected items under their saved sections. Its client preview shows the existing service description and Includes list with one package price. A separate full price breakdown toggle is off by default; private costs and margin stay hidden.
  • When discovery supports multiple useful outputs, the Delivery Planner and Proposal Strategist can organize one flagship asset and derivative edits as a coordinated content library. Shared filming reduces repeated setup, travel, crew, and equipment costs.
  • Each option must have meaningful differences in production coverage, outputs, channel adaptations, locations, or usage rights. The system rejects options that differ only by revisions or minor editing time.
  • Paid campaign or full-buyout usage is calculated from the account's approved pricing policy when configured.
  • Exact dates, final rooms or addresses, scripts, approvers, platform specifications, and optional add-ons can remain clearly labeled assumptions in a review-only proposal when the core scope and effort are known. Confirm them before production.
  • An active rate sheet or priced service catalog is required before the proposal writer can run.
  • If required lead research is unavailable, proposal permission stays on hold until research is rerun or reviewed.
  • Manual discovery notes are internally verified evidence. Enter Unknown instead of guessing when the client did not answer something.
  • From a lead page, answer Proposal Strategist follow-up questions and approve Save discovery update to store and reassess each answer. Select Review with Proposal Strategist when the opportunity is ready.
  • The strategist can prepare a draft directly from discovery. Introduction and investment discussions are optional. Corrections do not count as approval.
  • The Investment review is private to the studio and explains the full calculation and recommendation. The generated client proposal shows the confirmed scope and selling prices, not private costs or margin.
  • If a Service Template rate or package quantity changes after you confirm investment, SupaBook stops draft creation and asks you to regenerate or review pricing before applying the draft.
  • If the verified client request cannot support three honest package differences, choose a real coverage, output, or usage trade-off with the strategist. Revisions alone do not make a separate option.
  • Use Prepare proposal draft to fill the saved template. Review and request changes in the unsent Super File before sending.
  • To choose the account default, open a reusable Super File template and select Use for AI proposals. The proposal writer preserves fixed branding, images, studio information, reviews, cover layout, section order, investment presentation, contract, and payment flow. It fills project-specific sections and replaces placeholder package prices with validated options.
  • From a project, Draft Proposal with AI uses the linked lead's latest saved discovery notes, the current project record, saved pricing, and the account default AI proposal template. The project shortcut fills an unsent draft and retains existing images and contract pages.
  • Clarifying questions are not sent automatically. Record the client's answer, then reassess the opportunity.
  • Automatic proposal workflows must record both employee outputs and the final approval handoff. A recording failure keeps the draft blocked.
  • The Account Manager remains responsible for client presentation and follow-up.

Diagnose Business Health

The AI Business Analyst observes the full business funnel. It uses a context-first analysis method, an industry diagnostic pack, studio-approved targets, and verified outcomes. It checks data reliability before deciding whether the strongest supported constraint is lead volume, response, qualification, proposal conversion, capacity, or missing measurement.

  1. Open Business Health

    In AI Operations, open Business Health and choose a 30-day, 90-day, or 180-day review window.

  2. Review the Analyst profile

    Confirm the business model, current goal, decision context, evidence threshold, sales cycle, seasonality, and studio-approved targets. Blank targets remain unknown.

  3. Check data reliability first

    Review event tracking and lead-field coverage. A low score means the system should improve measurement before recommending a growth or sales change.

  4. Review the constraint

    Read the evidence-based diagnosis, confidence, missing measurements, and recommended next action.

  5. Prepare and approve a handoff

    Prepare an action to assign the issue to the Account Manager or studio owner. Approve it before work begins.

  6. Verify the outcome

    Complete and verify the handoff after its measurement window so the Business Analyst can compare the result.

Tips

  • Funnel observability means the platform can see leads but cannot reliably see enough downstream responses, proposals, wins, or losses.
  • AI employees keep firm role boundaries. An explicit request for another specialist’s work creates a visible handoff action instead of a promise or a duplicate confirmation question. Approve the handoff to place the request and verified context in that specialist’s queue.
  • The Business Analyst sends response, qualification, discovery, and follow-up work to the Account Manager, proposal drafting to the Proposal Strategist, delivery planning to the Delivery Planner, and repeat-client growth work to the Client Growth Manager. Demand, pricing policy, capacity, and low-confidence decisions can still require a human owner.
  • Industry guidance supplies relevant questions, funnel stages, and risk checks. It does not silently create conversion, margin, or capacity targets.
  • Studio-approved targets are the only targets used for account performance comparisons. External benchmarks require a source, date, relevant population, and applicability warning.
  • The Video Production pack checks scope readiness, audience, distribution, versions, usage, filming logistics, stakeholders, bidding clarity, production economics, capacity, delivery, and repeat-client value.
  • A handoff does not expand either agent's authority. Material changes and outbound communication still follow their existing approval rules.
  • Business health events, diagnoses, and handoffs are business records and are not automatically deleted.

Check Operational Health

AI Operations combines the existing exception queue with a read-only view of recent AI and automation activity. The summary covers the last 24 hours.

  1. Open AI Operations

    Open AI Operations from the workspace navigation to see current exceptions and system health.

  2. Review the health summary

    Check exceptions, delivery failures, stuck leads, AI request success, knowledge gaps, and workflow failures.

  3. Resolve exceptions

    Open the Exceptions tab to review, edit, approve, reject, or dismiss work using the controls available for that item.

Tips

  • A high exception count means work is waiting for human review; it does not enable automatic sending.
  • Investigate delivery and workflow failures before retrying to avoid repeated customer messages.

Control AI Autonomy

The Autonomy tab controls how much work AI employees may prepare. Begin with Shadow mode, review the recommendations and exceptions, then increase authority only after the account passes its readiness review. Turning on an operating mode does not grant unlimited authority or automatically approve customer-facing actions.

  1. Start with Shadow mode

    Open Autonomy and choose Shadow mode. AI employees observe eligible work and record what they would do without taking the action.

  2. Set spending limits

    Add daily and monthly credit limits. AI work pauses when a limit is reached.

  3. Review readiness

    Confirm the readiness review, spending limits, approval fallback, and emergency stop before managed autonomy becomes available.

  4. Pause when needed

    Turn on Emergency pause to stop new AI work across employees and workflows. The stop overrides every operating mode and action permission.

  5. Observe outbound timing

    Enable the outbound timing policy with Observation only turned on. The platform records proposed email and SMS delivery times without changing current sends.

  6. Set timing boundaries

    Configure response delay ranges, timezone, business and quiet hours, per-contact cooldown, and hourly and daily caps. Timing enforcement remains locked during this observation rollout.

Tips

  • Approval first allows prepared workflows to continue, but important customer and material actions still wait for a person.
  • Managed autonomy is limited to verified, low-risk actions and remains locked until the account passes its readiness review.
  • Credit limits are hard ceilings, not usage targets.
  • Changing autonomy settings does not send queued customer messages.
  • Urgent exceptions may shorten a timing delay, but they never bypass consent, suppression, approval, or final safety checks.

Trace AI Activity

The Activity tab provides a single timeline built from existing AI decisions, drafts, approvals, deliveries, workflows, knowledge searches, and usage records.

  1. Choose an activity type

    Filter by decisions, approvals, deliveries, workflows, knowledge, or AI usage.

  2. Choose a time range

    Review the last 24 hours, 7 days, or 30 days.

  3. Read the trace details

    Timeline rows can show the outcome, prompt version, number of approved knowledge sources, model, token use, and request time when those details are available.

Tips

  • A knowledge gap means the AI could not find an approved source, so it should not invent the missing business rule.
  • Use the Refresh operations button after resolving an exception or retrying a failed workflow.

Browse more Automations & Workflows help articles or return to the SupaBook Help Center.

Canonical page: AI Operations canonical page