Ask your AI about ASU

From copy to a pilot

01
Copy the brief
It includes the product, task examples, and instructions for the assessment.
02
Paste into a new chat
Use the Codex workspace or AI account where your relevant business context is available.
03
Review the estimate
See hours and cost by job, the audit status, and one task to try first.

Create your first task

Ask about any ASU feature. When you’re ready to start, paste this request into the same chat.

Register my ASU manager account and create my first task. Give me the task link and a message I can send to my worker. Show me where to track progress, review the proof, and request a redo.

11 chapters

What your AI gets

A brief that separates tested Calling behavior from approved screens, planned features, and sample cases.

Preview the copied text

# 1. Where ASU gives a manager time back

Assess where ASU could save time in my business. Use the reference below and the authorized records you can already access. Start with the actual active workers by role and the number of eligible jobs they handle. Show the strongest credible monthly savings in 250 to 350 words, with a compact table and the arithmetic. End by inviting any ASU question and offering first-task setup. I explicitly ask you to delegate the numbers and workflow audits to two separate independent reviewers before releasing the answer. Use gpt-6.1-sol for both reviewers in Codex. Apply the audit and pilot instructions in chapter eleven. The assessment and audits are read-only business analysis. A local draft and audit receipts are fine. Use only company facts supported by records from the applicable period.

The reason to consider ASU is the work that starts after a manager assigns a job. Someone chases the worker for an update, opens the material the worker sends, checks every requirement, asks for a fix, and repeats the check. The manager also answers questions about cases the instructions missed. That work repeats even when the company has a clear procedure and the worker knows the job. It consumes attention that could go to staffing, exceptions, and improving the standard itself.

With ASU, the company defines what counts as done and what evidence can demonstrate it. A worker opens a task, follows the instructions, and submits the proof. ASU checks the submitted evidence against the applicable criteria. If the expected view of a finished room is missing, the worker gets the exact request for a better photo. If a question requires a company decision, the manager gets the question with the relevant proof and context. A checked box is a report from the worker. The evidence check determines which requirements the submission supports.

The same pattern works across a concierge's recurring shift, a freelancer's regular assignments, a one-time repair, and a partner's calling campaign. The company can reuse its own form and criteria across people in a role. It can give a different role its own standards and ask for evidence that fits the work. The first task can use an existing reference or procedure and a worker link. The worker continues using the tools appropriate to the job and submits the relevant proof in ASU.

The product idea came from the owner's account of managing hospitality operations remotely and wanting fewer routine checks, clearer feedback, and fewer repeated answers. The construction and client follow-up scenarios express the same problem. A reported quantity needs readable measurements. An approved answer needs proof that the client received the required reply. A manager can help an employee finish without inspecting every routine submission personally.

The useful question for your company is which repeated check can be defined clearly enough to delegate. Consider frequency, time currently spent, the cost of a false pass, and how easy it is for a worker to collect the proof while doing the work. Start with one real process with a reliable standard and observable evidence. Preserve the original due time and the history of corrections. Judge the trial by the manager time it recovers and the quality it maintains. Chapter eleven gives the detailed response and pilot instructions for this assessment.

# 2. Product map and evidence

ASU has a Calling implementation with code and controlled test records. The broader task constructor, reusable role forms, cameras, evidence checking, worker access, reports, and hierarchy are described through individual designs and integrated simulations. Use the exact evidence level for each capability. An owner-approved interface confirms the interaction and any specifically approved example values. A browser test proves the stated preview behavior. A controlled Calling test proves only its exercised path. An open question about complete production acceptance remains an open question when recommending a live pilot.

The Calling workflow has a manager who sets contact work and acceptance criteria, and a caller who submits the original recording and related proof. Controlled records show why the difference between work and business outcome matters. In one housekeeping recruitment test, the conversation provided useful answers and a next step, while the separate introduction criterion remained unmet. That specific finding can be identified without discarding the supported facts. Approval, in-task access, and external delivery are different events. The Calling sources and the recorded test supply the reference for checking implementation claims.

The task creation master and shared page register hold the approval history for the wider system. Proof first, time and frequency, human-authored requirements, distribution, recipient details, review, and confirmation are separate design decisions. The approved action on screen 10SN is Continue, with draft gaps carried through to the creation check. Screen 11D, the extra owner selector, was retired. Responsibility follows assignment. The proposed backward deadline buffer H-E2 was deferred. Company working-time escalation, the original due time, and visible exceptions supply the active design. These distinctions prevent old previews from becoming accidental defaults.

The integrated task creator combines the separate screens into a local simulation and includes Review all inputs, conditional editing, and demo invitations. The historical DEMO-6B6B12E7 example shows a daily task at 08:00 San Francisco time beginning November 9, 2026, with an explicit 30-minute proof window and permission to start after assignment. Approved recurring rule 8R is a separate choice. It allows that occurrence to open at 08:00 when the offset is zero. A start rule alone does not set a proof deadline. Treat each demo value as belonging to its own setup.

The signed-in worker and manager cabinets were returned for rework at revision one. Revision two is a repaired simulation with separate worker, manager, and handoff checks passed by the recorded reviewers. Owner review of the combined revision remains pending. It uses a separate organization assignment to demonstrate the release boundary. The worker proof results revision one and the initial finding and recheck entry are approved designs. The later timestamp and image callout, blocked correction help, and general appeal argument stages have their own pending review status. The implemented Calling guard supplies an evidence reference for the wider appeal design.

The 15 generated case visuals, three construction visuals, role forms, and channel workflow graphics show how to explain the intended system. They are useful for selecting an operation and specifying proof. Classify a proposed call-center deployment as a use case assembled from available and planned pieces. A case example, author approval, and client deployment are different forms of evidence.

The two-minute task setup, 90 percent routine supervision concept, five-minute feedback after upload, and fivefold manager coverage are targets with separate measurements. Use the screen register to confirm the latest source, exact revision, branch, and what was actually approved. Where a more recent authoritative record changes a status, use that record and preserve the earlier example as history. The companion source index names the relevant design chats, product record, generated cases, and implementation evidence.

# 3. Apply your company's standards

A company can bring the rules it already uses. The manager writes what must happen, which facts determine acceptance, and what source can support each fact. Use rules the worker can understand while doing the job. Give each requirement a clear scope and observable pass condition. An existing procedure can contain instructions for people, evaluation rules, and escalation policy in the same document. Translate those into separate fields where they affect the task. Show the current approved wording to the worker and carry its version into the review.

There are five ways to give ASU a comparison reference. An example shows what the result should look like. A manager can supply a photo of a prepared room, a drawing, a video of an acceptable setup, or a document with the relevant section identified. A checklist states the required items and allows each item to be checked. A numerical requirement gives a value, acceptable range, or limit, with its unit and scope. A system record identifies the system, exact record, fields, and expected values. A personal inspection routes a check to the person authorized to examine and record the result. The designer chooses the relevant path after the manager selects a reference type. Each requirement can have multiple checks and each task can have multiple requirements.

The source of a standard and the source of worker proof have different purposes. A reference photograph defines the expected room. A newly taken photograph shows what the worker submitted for this room and request. A company form can record the reported result and the worker's answers, while separate photos or video support visible facts. A system screenshot can show a booking, item price, client address, or delivery status. The manager specifies the field, expected value, and identifying details the proof needs. The worker supplies fresh material inside ASU, alongside the tools they already use. This creates a starting route based on evidence upload, a reusable standard, and a task link. Direct connections to existing software have their own evidence and authorization.

For example, a front desk procedure can require the assigned lobby, badge and uniform, and the full desk setup against a reference. A desk photo that cuts off the check-in area can support the visible checks while leaving the setup check open. The worker sees which view to retake. Set the required form once for the role and reuse it on each applicable shift. Keep each worker's result and each dated request separate, even though the standards are shared. The same approach works for a cleaner, an electrician who needs to show the assigned panel and its labels, or a plumber who needs to record the repaired joint during the required test.

A construction role form can ask for the worker wearing the required helmet, visible workwear, the worker next to the assigned site marker, and evidence of the finished work. One live clip can support several visible checks if the relevant parts are clear. A separate full-area photo can support the result against the approved reference. A form marked Yes can coexist with three supported checks and one open check. If a measurement is required, the image and measurement must identify the same area. A video establishes what can be seen within its scope. A company requirement concerning concealed work can specify the applicable test record or authorized inspection.

For a conversation task, define required meaning and any values that must be exact. Let a caller use natural wording when the procedure allows it. If the recipient has already supplied a clear answer, use the answer and its source. Ask the next applicable question. Exact contact details, prices, dates, approved promises, and permitted dispositions can be required separately. For a meeting, the manager writes criteria for which promises, deadlines, questions, or answers to capture. Each commitment then has an owner and proof that fits its action.

Use the smallest proof set that can establish the standard. Require distinct materials when they prove different facts. Carry an already selected worker proof type into the relevant comparison step, and ask for the expected details. Review any incomplete checks before creation. An example can illustrate what the worker should capture without turning unrelated sample details into company policy. When a business changes a rule, publish its effective version, identify affected work, and retain the previous decision. Review already started work explicitly if the change affects what counts as done.

If the reader supplies their own procedure, draft a requirement card with the expected result, acceptance conditions, proof source, required visible details, and manager decision route. Mark any scope question that changes acceptance. For example, if a drawing shows a tiled lobby and the work order also refers to an entrance, ask whether the entrance belongs in the measured area. The answer affects the criterion. A question about the worker's display name can wait if the task link and proof can be reviewed without it.

Choose the reference before choosing the required detail. For a price check, specify the offer currency and whether the limit includes delivery or additional charges.

# 4. Create a task in two minutes and send it

The quick start demonstrates a simple task whose standard is already available. The timer starts when the manager opens Create task and stops at a valid, ready-to-copy link. A new procedure may require separate time to define or approve criteria. Record that time as setup. In a live pilot, measure the actual setup and link handoff with the authorized workflow. The integrated task creator is useful for practicing that route and showing all inputs, including what would block creation. Its simulated link and controls illustrate the design.

Start with the evidence. Select one or more required types, such as a form, a fresh photo, a live video, an original call recording, or meeting audio. Carry the selected types into later steps. Add conditional details when they are needed to establish the result. A lobby opening task may need one live photo and a short form. A call assignment may need the original audio and a screenshot for a separate communication requirement. Choose the type based on what the worker actually has to demonstrate.

Choose one-time or recurring work. For a one-time task, ask whether there is a completion deadline. A dated task has a specific due time and zone. An undated task remains open until the outcome and evidence are resolved. Its due date stays unset. Its applicable action windows can still be used for a blocking issue. For a recurring task, start with a simple daily time and allow additional times. Advanced scheduling can include selected days, intervals, or random checks. Each scheduled occurrence has its own dates, proof, status, and history. A new occurrence does not erase an older unresolved one.

Keep the schedule, permission to begin, and proof deadline separate. San Francisco time is the approved starting default for the example settings, with the local clock carried as America/Los_Angeles. Use the actual company calendar and time zone for the reader's task. A choice to restrict start leads to the applicable one-time or recurring start settings. For the approved recurring variant B, zero offset means a request opens exactly at its scheduled time. A range or offset must be shown clearly before final review. When proof is due within a stated interval, calculate that due time from the relevant opening or assigned event under the task rule. When the proof window is unset, preserve the unset value.

Write the outcome and specific checks using the chosen comparison reference. The manager can add another check or requirement and review what was entered. A system record route asks for the platform, record, and fields the worker's fresh evidence must show. The approved Continue step preserves a draft while missing required values still block final creation. Use what the manager already supplied when moving to review. Select a department when it applies, or carry the company route for work outside departments.

Decide distribution before choosing people. A shared task produces one result under the selected allocation. Individual assignments require each worker's own result. A queue gives each eligible worker a distinct item and advances that worker after the current item is accepted and all required redo is cleared. Choose one worker, several people, or a group, then choose the organization or task-only recipient path and the actual recipients. Assignment determines responsibility for proof and corrections. A task created from a selected worker's details carries that worker forward and skips the duplicate allocation question.

The accountless invitation opens only the assigned task. For one recipient, the manager can copy a message and task link with an optional label. For several recipients, the designed flow lets the manager set the count, then copy each separate invitation so attempts and progress stay separate. The manager sends the link through their ordinary channel. Opening a link or copying a message is an observable setup action. Sending, receipt, proof submission, and verified completion are later actions with their own evidence. An organization invitation is a separate reusable link used by people joining the company and signing in.

For the approved one-time reminder interface, the starting setting is one editable ASU reminder 15 minutes before the deadline. More reminders or an AI reminder call can be added. The call is a reminder in the worker's selected language and can target eligible recipients as configured. In the approved dated one-time escalation example, wait 15 minutes after both the deadline and the last configured reminder, alert the manager, and try one authorized transfer to an eligible backup in the assigned group. If no eligible replacement exists, the manager must address the task. Preserve accepted evidence, the original due time, and assignment history. Pause the specified overdue reassignment while proof is under review. Company working-time response windows govern the separate hierarchy design. They belong to the same open obligation when applicable.

Show a final review with the outcome, criteria, selected proof, timing, distribution, recipient, reminders, and decision route. Give the manager a way to edit an input directly and return to review with other inputs retained. A valid configuration can reach confirmation and show a clear message and link for each recipient. The worker can then see the brief and the requested evidence. A recurring invitation and a new company registration have different purposes.

A useful demonstration is one front desk opening with the correct lobby and prepared counter visible in a fresh image. Use a shift time and proof deadline appropriate to the business. The manager chooses the existing counter reference, requires the lobby marker in the image, selects the assigned concierge, and copies the task message. If the reader works with a contractor instead, use the same form of brief for that contractor's work. The worker sees one actionable request and can reopen it to see the current result. Record the elapsed setup, test the link and proof path, and compare the actual time with the two-minute target.

For a queue, the recipient count and the work list are separate inputs. A group can exist before any item is assigned. Review the proposed eligibility, distinct items, and assignment settings before copying links. If a worker has a pending correction, show it with the reason the next item waits. For an undated task, preserve any open decision and show its configured response window separately from a completion date.

# 5. What the worker sees and how proof is checked

A worker can open a task-only invitation directly or use the signed-in company cabinet. The task link keeps its own assignment and proof history. An organization invitation lets a person register or sign in, verify email, and confirm joining the company. The company can reuse that registration link for multiple people. These paths give a different scope of access. In the signed-in cabinet, the worker can see assigned tasks and the permitted company information. In the task-only path, the link provides the instructions, submission, status, and follow-up for that task.

The approved My tasks design distinguishes current work from completed work. Sort actionable work by the relevant deadline, and keep planned work hidden until its start rule allows it to open. Respect that rule when restoring a saved view. One recurring parent can contain dated requests and separate attempt histories. Each occurrence has its own due time, progress, and result. The worker can go back to all tasks after reading an assignment. Opening the next occurrence leaves an earlier missed deadline visible. An undated item can still appear as open and actionable.

The task view shows the goal, instructions, permitted answers, mandatory questions where applicable, and the exact proof the manager selected. Show a separate slot for each required file or capture. If the task requires both a photo and video, each slot can be reviewed separately. The approved live photo and video designs capture inside ASU, with the available camera choices and a confirmation step. They use fresh captures for those proof types. A form records fields the worker submits. A call may use the original recording, and a meeting may use audio. A screenshot, export, or additional document can support an applicable check according to the task requirements.

The worker can ask a scoped clarification about the assignment. Show an applicable manager-approved answer if one exists, and use the remaining open question when a new decision is needed. A required question can have two separate facts: the caller asked it and the conversation established the required answer. An answer volunteered earlier may already satisfy the substantive requirement. A request to check a price or send a reply may require evidence of the actual external action as well as the conversation. An unavailable source needs recovery or clarification tied to the affected criterion.

The submission review distinguishes checking, passed, and needs redo. It shows results by required question and proof item, the reason a check needs attention, and what to submit next. If the desk photo misses one part and the video already passed, the worker replaces the photo while the accepted video carries forward. A new upload is a new attempt on the same obligation. If a planned timestamp or highlighted image points to a problem, it should identify the exact place relevant to that check. W-RES1 revision one is the approved result flow. Revision two previews Show evidence and Can't redo this? with a scoped manager question. Carry their review status forward when explaining what is available.

Proof can be under review, require a correction, await a manager, or be verified. Track when the worker acted, when the proof was uploaded, when it was checked, and the task's original due time. If complete timely proof is waiting for review, show Review pending and route the outstanding action to the reviewer. A genuine late submission or missed action remains in the history even if verification comes later. If the worker cannot obtain a replacement, the manager receives the affected item, the reason, and the next decision.

Verification requires the applicable evidence and requirements, cleared mandatory redos, and resolved decisions that affect acceptance. The manager's answer can guide the next action. It still takes proof to establish that the worker followed it. A finding can be opened for the approved first appeal step tied to the task and evidence. The linked path carries the attempts forward. The result tells the worker what they can do next and tells the manager why a decision reached them.

When a shared queue is used, the worker advances independently from the other workers. Their current item stays active while evidence is being reviewed or while a required fix or decision is open. Once the item meets the acceptance condition, the next eligible item can be assigned. A completion report counts the same obligation once across its attempts. For the reader's example, show a specific worker message with the original requirement, evidence that passed, the missing detail, and the smallest useful new submission. Include the time at which the next action becomes due under the agreed policy.

An optional recipient label helps managers distinguish assignments. A check about identity or attendance needs its own evidence tied to the worker, request, site, and time.

# 6. Manager work, reports, and escalation

A manager needs to know which task requires action and why. The organization design supports departments, a department manager, optional team managers, and placement of workers under their actual manager. A task can belong to a department or follow the company route. New tasks inherit the relevant authorized reporting structure where configured. The directory can show compact worker and manager counts, then open an individual's task history. A manager can create a task from a selected worker without selecting the same person again. Keep permissions to create tasks, inspect evidence, manage people, and remove a task bound to the authorized role.

In the ordinary review, show the requirement, evidence, check result, open correction, and decision dependencies. A manager can answer a question for the current task or explicitly approve reuse for matching situations. A reusable answer needs the relevant authority, question and criterion context, applicable version, effective time, and any limits. An exception approved for one client can stay with that client. A general answer can guide later workers when the same facts match. When the facts change, the new case gets the applicable rule or a new decision. The authorized answer can then be checked against what the worker actually communicated. The scenario with 10,000 workers shows this reuse principle as a scale example.

The company-wide escalation design applies to existing and future company tasks when the policy is activated. It includes work outside a department, recurring requests, and applicable task-link assignments. The company defines its working days and hours, closures, time zone, and how long each level has to act. Departments can have working calendar and response window overrides. Apply to all sets the intended company profile across the affected departments. Team shifts and any required authorized routes still matter. The working calendar interface was approved, with actual values and activation awaiting configuration.

A task deadline describes when the outcome is due. An action window describes how long the responsible person has to take a particular action before the case moves up. Preserve both. At the end of an applicable window, show what is missing, the evidence and blocker, the next action holder, and the reason for escalation. If complete evidence was timely and a review is pending, the pending action belongs to the reviewer. Use one incident for the unresolved obligation, with its real dates and handoffs. The separate earlier deadline buffer screen was retired. A recurring schedule can continue while the older case stays open.

The reviewer personally follows up and records what happened. They can work with the assigned person, explicitly reassign to an eligible person, arrange a substitute, or take over. A proposed substitute is different from a confirmed change. Reassignment updates actual responsibility and retains the original deadline and evidence. The new performer still owes the required proof and corrections. If the reviewer fails to act within the applicable window, the case moves through the authorized hierarchy. At the final level it stays visible to the CEO or owner. A configured authorized deputy can take the next action when the stated rule allows. Viewing or acknowledging a case is recorded separately from a substantive action.

Administrative removal is its own outcome and permission. A designated removal admin confirms the exact task or occurrence and writes the reason. Keep prior attempts, the original due time, why it escalated, and who recorded the removal. Completion remains the outcome of ordinary proof verification. A manager taking over or a decision on an appeal returns work to that same evidence standard. Removed work does not count as completed or qualify the worker for the next queue item. The approved follow-up and confirmation screens preserve these boundaries.

The Day Concierges example shows three concierges as rows and six regular tasks as columns. The checkpoints are arrival confirmation, start of day, ready for midday check-ins, going to eat, returning after eating, and end of day. Week and Month views summarize completed requests, resubmissions, and missed deadlines. Open a cell to see dated history, original due times, criterion results, and proof. A recheck is another attempt on the same request. A worker-history export provides the relevant complete context for their own LLM analysis and is distinct from this product context. It remains available independently of the reporting period.

The 1,000-meeting campaign shows progress and remaining work, with results by worker and criterion. A count traces to the accepted meeting and its evidence. Count each completed meeting once across attempts. A first-check rate uses the applicable checked items as its denominator and keeps pending cases visible. Follow the individual link when a total suggests a problem. Use the same definitions for comparisons across workers, and check assignment mix before drawing a conclusion about productivity. A total measures the work recorded under the defined standard.

The meetings design lets the manager define which commitments and unanswered questions to find for that meeting task. Bind a candidate commitment to its source, owner, timing, and required evidence. An unclear owner, due date, or answer goes to the manager with the relevant record. If a client asked about Saturday installation, an approved answer can enable the reply. Record who will send it and when, through the required channel. The linked reply closes when proof supports the recipient, approved meaning, action time, and other applicable requirements. Calendar and task effects must be evidenced when the workflow calls for them.

In the repaired signed-in cabinet demonstration, dated worker and manager receipts can pass between preview views for the same sample occurrence. The handoff keeps prior misses, accepted proof, and local drafts with their owners. In production reasoning, verify identity, scope, current state, and actual effect before accepting a receipt. A worker's submission supports review. Approval authority stays with the authorized review route.

When an overdue case moves between levels, the manager's action and the reason remain separate records. A changed deadline requires an authorized decision. Preserve earlier dates alongside the new date so a later report can show actual performance.

# 7. Appeals, unclear evidence, and recovery

Give the worker the finding they can act on. It should point to the exact task, attempt, requirement, and evidence. Explain what the submitted material supports and what remains open. A cropped screenshot may require a wider screenshot of an action already performed. A missing substantive call point may require the next permitted correction. Unreadable or unavailable material can require a technical hold and a retry against the same evidence. These situations have different remedies and different responsible people.

The approved first general task finding step lets the worker ask ASU to recheck a specific finding tied to the original attempt. The worker can say which action they took or which fact they dispute, and connect it to the check. Existing proof may already establish that link. If the source is insufficient, ask for the missing detail or allow supporting photo or video. Keep the original proof, previous wording, versions, and deadlines visible. An explicit resubmission can return the same issue for review. A resubmission is a new attempt or explanation on that issue.

Calling has an implemented argument check before a case reaches human review. It is separate from the evidence recheck and from the manager's final decision. After the applicable recheck, the worker's own specific action or disputed fact must connect to the actual finding or requirement. A short, concrete linked dispute can meet the minimum even if the final finding is uncertain. A generic objection calls for a precise clarification. A new file, quote, or timestamp can help and is optional when the existing evidence already supplies the connection. When voice explanation is used, confirm the transcript the worker intends to submit.

If a decisive source cannot be read, show the technical hold and keep the same issue attached to its original scope. Restore the relevant material before trying the check again. Preserve findings already supported by other sources. When a linked case qualifies for manager review, show the dispute, earlier evidence, and the worker's reason. The manager decides the disputed point under their authority. Record the decision and rationale. Preserve the original due time and the rest of the evidence. A decision can resolve one finding while other task requirements remain open.

The general task design applies this pattern to a form, live image, video, meeting audio, or other applicable evidence. Its later W-AP2 revision five is a preview awaiting review. The earlier stages show status labels, a clear reason when routing is blocked, and an editor that asks what the worker did, which fact or criterion they dispute, and why the result would change. The signed-in cabinet follows the approved W-AP1 endpoint for that integration. The Calling guard is the implementation reference for its own route. Carry the state of each capability when explaining which path could be tested now.

For the reader's business, show one example in which the worker provides a better view and the ordinary correction clears. Show another in which the evidence is unchanged but the worker identifies a specific fact the reviewer missed. Label the outcome according to the evidence. A human approval should cover its stated scope. Return to normal verification, where every required criterion and outstanding correction must be resolved before a task counts as done. Measure reopened findings and holds.

# 8. Employees, freelancers, one-time work, and partners

The same method applies whenever the company can express the outcome and agree on evidence. Choose the access model and approval route that fit the relationship. A permanent employee can join the organization and see recurring work, completed requests, and allowed guidance. A recurring freelancer can get the same repeated standard through the relevant assignment route. A one-time contractor can receive a task-only link with a single required result and proof. A partner company can carry a set of agreed assignments under a company procedure, with the evidence scope and responsible people defined for that engagement.

For full-time concierges, use the role's regular checkpoints and the same opening standard across people. Each dated request gets the assigned worker, original time, and evidence. If the image misses the check-in setup, ask the worker for that specific new view. Route access problems or unresolved decisions to the supervisor. Review Week and Month totals by criterion with a drill-down into dated attempts. This gives the manager a way to examine the work that needs attention without opening every ordinary submission.

For a freelancer, assign a repeated or one-time deliverable with a clear reference and deadline when applicable. A researcher could submit a form with the requested fields and a source screenshot showing the relevant record and date. If the record ID is missing, request the missing proof. A property photographer might show the required views of the assigned room, using the specified capture method. Give the person the instructions and the proof they owe. A freelancer's contract rate and time spent producing required evidence belong in the savings calculation. Keep their task access scoped to the work assigned.

For a one-time repair, use the equipment or work-order identifier, the required visible result, and any test record or authorized inspection the company requires. A worker can open the brief by link, perform the work, and submit the result. If the video starts after the required test, ask for an authorized recording that shows the relevant interval. If access is blocked, show the manager the case and the missing evidence. The manager can arrange access, reassign, or follow another authorized path. The original outcome and proof requirements remain visible through the handoff.

A partner call center is a proposed application with a concrete evidence path. The client provides the version of its procedure, allowed contact list, required questions, required exact terms, approved answers, permitted follow-ups, and accepted outcomes. The partner assigns each contact under the chosen distribution, and the agent sees only the instructions and context relevant to their assignment. The work can be handled with the partner's existing calling and communication tools. The worker submits the required originals and linked proof in ASU. A record should support which contact and attempt the original belongs to, when the action happened, and the requirement being checked.

For example, a vendor handles 20 contacts for a recruiting campaign. A verified result for 19 leaves one contact open. For each call, check the introduction, required facts and questions, permitted treatment of negative or volunteered answers, and agreed next step. Natural wording can satisfy a semantic requirement. An exact contact, price, term, or deadline is checked as specified. A specific omission can trigger a permitted callback, with unresolved points for the same contact grouped into one actionable request. If the original already contains the answer, keep it and move to what remains. Respect the company's rules for stop requests, unsupported contact claims, and next actions.

The client's policy owner decides new terms and scope exceptions. The vendor supervisor can follow up with the agent, check the original, arrange a permitted correction, and resolve issues within delegated authority. ASU can reuse a previously approved answer when its conditions match. If approval creates a new follow-up, track proof of the actual action. A correct call and a converted prospect are distinct outcomes. Payment eligibility uses the defined contract and acceptance evidence. Any payment execution has its own records. An uploaded summary or reported total is checked against its underlying attempts where required.

Give the auditor the procedure version, sampled original records, matching contact and attempt identifiers, per-criterion findings, corrections, exception decisions, and dates. Define the sampling rule before the trial. Include ordinary passes, failed checks, late work, and unresolved cases. Record which criteria were independently checked and which still lack an original source. If a sample pass is overturned, reopen the relevant check, preserve the reason, and examine related items under the agreed audit policy. The auditor can verify whether the partner used the right procedure and whether the sampled evidence supports the reported results. Show the sample coverage and count unique accepted contacts across resubmissions.

Start with one partner campaign and a clear recipient and approval route. Choose an authorized sample of calls, compare the partner's current review and the ASU-assisted review under the same procedure, and measure manager minutes, extra agent effort, false passes, time to correct, and outcome coverage. Use actual contract fees for any change in paid vendor work. When the vendor invoice remains fixed, report verified changes to client oversight hours or contracted rework separately. This makes the relationship model useful for a staff team and for an external service without confusing the contract, the standard, and the evidence of work.

Use an auditor view that keeps the client procedure, applicable approval scope, and vendor performance records together. Give each disputed check a responsible person and a due action. When evidence cannot be accessed, keep that check open and record the gap. For each metric, distinguish the number of unique contacts assigned, the number supported as accepted, the number corrected, and the number independently reviewed by the auditor.

# 9. The closed-loop case library

Choose a case to adapt. Each is a scenario from the design or generated visual set, with any deployment or test status separately stated. Transfer the pattern and substitute the reader's actual rules, identity, and dates. For each case, preserve the source, action, and decision that make acceptance possible. The source pack has the 15 individual graphics and the three additional construction graphics.

**01. Room 14 cleaning.** The manager supplies a reference photo and the required areas of the room. The worker cleans room 14 and submits the original walkthrough video. Check that the room is the assigned one and each relevant area is visible and matches the company's reference. If one required area is dirty, name the spot, ask for the correction and a new view, and carry forward what passed. A reference that leaves the requirement unclear goes to the manager. Close after the required visible result is supported. Record the due time, the action the evidence supports, and the later upload.

**02. Two daily bookings.** The supplied example calls for two bookings each day, at least four hours apart, with the matching Sirvoy cancellation and a separate one-use card with a €20 total limit for each booking. Ask for the corresponding booking, cancellation, time, and permitted limit proof. Match the records and check the interval. If one screenshot belongs to the other booking, ask for the missing matched record. An impossible timing requirement or policy exception needs the manager's decision. Close the daily request when both distinct chains pass. Keep the original dates and create the following day's obligation separately.

**03. Twenty contacts.** Publish the full list of 20 unique contacts and the evidence needed for each permitted outcome. The worker supplies the original and any required message or attempt proof. If 19 are supported, identify the one contact that remains open and the reason. A verified decline can be a permitted work outcome when the company's rule says so. A stop request or a genuinely exceptional disposition follows the authorized path. Count each contact once and retain earlier attempts. The full assignment closes when every contact reaches an allowed, evidence-backed outcome, with conversions shown separately.

**04. Recruitment call by meaning.** The company requires the applicable hotel or hostel experience, cleaning time including linen, availability, and the next step. Check the original audio and contact and time linkage. Credit a clear answer volunteered earlier. Natural wording can convey the approved meaning. Preserve separately any exact information or mandatory introduction. A controlled Calling example isolated an unmet introduction even though the candidate supplied relevant answers. If a required detail remains missing, give the permitted corrective action for that contact. A new question beyond approved guidance reaches the manager. The work result and recruiting outcome have separate records.

**05. Meeting promises.** A salesperson says they will send a proposal Friday and speak again Monday at ten. The manager's meeting criteria specify which promises to extract. Bind the two commitments to the recording or meeting notes, their owners, and due times. A sent proposal needs the correct recipient and attachment evidence. A call or appointment has its own required proof. An unclear person, time zone, or date needs clarification. Keep a draft visibly open until the missing decision is made. The meeting can be verified while its promises are still open. Each commitment closes on its own proof.

**06. WhatsApp follow-up.** A worker sends the authorized proposal or information to a specified client. The submission needs the recipient, required meaning or exact terms, time, and required sent or delivery status. A cropped header can hide the recipient even if the text is readable. Ask for clearer proof of that message before considering another send. A new commercial request outside the approved answers goes to the manager. Keep the action time and the later screenshot upload distinct. Close the communication when it meets its stated standard, and keep any later promise as a separate open obligation.

**07. Telegram agreement.** A conversation changes an appointment or adds a follow-up. The worker provides the relevant messages, correct client context, and the earlier agreement. Establish what changed and whether both sides actually agreed to it. Keep the original time and the source for the accepted change. A proposed time that still needs agreement goes to the responsible person. If an updated calendar record is part of the job, ask for proof of the updated record. A later obligation still needs its own performance evidence. Use dates from the source and retain the reasons for any change.

**08. Email proof.** The company requires an email to the address the client gave, with specified information and an attachment if applicable. A sent message can establish the recipient, body, attachment, and time. The worker can provide a clearer original when the attachment or header is obscured. If the address in the source is ambiguous, route the question to the right person before requesting a send. When the task also requires the client's reply, keep that dependency open until the agreed evidence arrives. A matching email action and the resulting business outcome are checked separately.

**09. Call back in an hour.** Keep the source request and its actual time. Translate the agreed interval into a due action for the right contact and caller. When the customer changes the time, retain both records and the basis for the new appointment. If the worker reports calling back, check the original and the required action time. Keep an overdue action visible while proof is outstanding. An ambiguous agreement needs clarification, and a refusal follows the permitted disposition. Close the callback when the evidence shows the required resolved outcome and keep any original miss in the history.

**10. Missed mandatory question.** The original call may support the cleaning time while leaving whether it includes linen unresolved. Show that exact gap with the source. Group open corrections for the same contact into one permitted next action. Carry forward the time and any other established answers so the caller can focus on the missing point. If the relevant answer requires a manager's decision, send the question separately with its context. When the caller submits new evidence, recheck the unresolved points and preserve the first attempt. Completion follows the approved substantive checks for the task.

**11. Unclear proof.** A cropped screenshot, partial original, or unreadable recording can leave a check unresolved. Tell the worker exactly which fact needs support. A screenshot with a hidden phone-number header can be replaced with a clearer view of the same message. A source failure can go to technical recovery against the same attempt. Preserve anything already supported. If the underlying action is uncertain, obtain the required original or ask the authorized person about the conflict. Close only after the appropriate evidence clears the relevant check. Keep the action time, upload time, and processing history.

**12. Missed deadline.** A task reaches its original due time while required fulfillment is still unsupported. Keep it open and record what the evidence establishes about the action time. Material uploaded later may prove an earlier action if the company standard accepts that source. Complete timely proof waiting for a decision belongs in review pending. An actual missed deadline stays recorded when the work later verifies. Follow the configured reminder and escalation route, and show what action is needed next. An exception that requires approval is recorded with its reason and scope.

**13. New manager question.** A client asks a question for which no approved answer applies. Capture the source and why the answer matters. The authorized manager chooses the answer and where it applies. Track the reply as an action for a specific worker, recipient, and deadline. When proof arrives, check the approved meaning and required delivery evidence. A published answer enables that action. The question and reply obligations can remain open at different times. Retain the recorded decision for the current case and for later cases that match its explicit conditions.

**14. One answer for matching cases.** When another worker meets the same approved situation, identify the current version, rule scope, and matching facts. Apply the permitted answer and check the worker's communication against its meaning. The 10,000-worker illustration shows why one reusable decision can remove many repeated manager questions. Different facts, conflicting guidance, or a personal exception may need another decision. Ask for a focused correction if the delivered message does not meet the applicable requirement. Keep a record of the approved answer used and the evidence that the current worker fulfilled the action.

**15. The criterion changes.** A manager updates an approved script, answer, goal, or proof requirement. Publish the intended scope and effective version. Show affected workers what applies to their work, and keep the old version with the earlier decisions. If work was already underway, determine which rule applies to that attempt and record the decision. A review checks the result against the version bound to the work. Request a targeted correction if the wrong rule was used. Keep the previous result traceable. A test of the changed rule should establish that the intended new behavior actually reached the worker.

**Construction A. Daily checkpoints.** Record the starting condition, midday progress, and end-of-day result for the same site and work area. Bind the assignment, worker or crew, original times, and relevant reference. A fresh video can show visible PPE, workwear, and an assigned site marker in addition to the work. Check the facts separately. A result claim needs its own evidence. If the final area is cropped, preserve supported checks and request a full-area capture. Carry a blocked access or unclear scope question to the manager and retain the original checkpoint history.

**Construction B. Materials request and delivery.** The request identifies the approved material, quantity, allowed cost, needed date, and job. A purchase approval records the decision. The receipt supports what was purchased. Delivery evidence supports what reached the job, where and when. Installation or consumption needs its own proof when applicable. Match the item, amount, and identifiers across records. Keep missing quantities or delivery proof open. Ask for the exact missing record or route a changed price or substitution to the authorized person. Show material waits as blockers on the work rather than treating them as accepted output.

**Construction C. Monthly verified output.** Use accepted new quantity and verified labor hours from comparable work. Record unit, worker or crew, location, scope, and measurement basis. In the tile example, readable measurements initially support 8.0 m² out of 12.5 m² reported. The additional 4.5 m² needs evidence and any scope decision. Keep rework, access issues, and materials waits visible separately. Credit the same accepted area once. Electrical or plumbing work can have its own visible result and required test or inspection record. A month's verified output can be traced to its accepted tasks and evidence.

# 10. Time, cost, and San Francisco wages

Choose a comparison period and a specific set of comparable tasks. Measure the effort each person spends supervising or collecting evidence for those tasks. Include chasing, reviewing, sending corrections, answering questions, checking resubmissions, resolving disputes, and auditing a sample. Keep active human minutes separate from time waiting for a system or worker. Record setup once and ongoing changes separately. For a partner, distinguish client oversight time, vendor supervisor and agent time, and the contract fee. Use the same quality and timing requirements before and after.

For each task type, multiply the number of assigned tasks by the fraction reviewed manually and the minutes per reviewed task. Divide by 60 for the baseline review hours. If the baseline also includes other measured manager work, track those hours in the same period. The owner's starting assumptions are 30 minutes for a reviewed task and 99 percent of tasks manually reviewed. On those assumptions, 30 × 0.99 gives 29.7 review minutes per in-scope task. This describes baseline review coverage. Actual company logs can replace both inputs.

The target of five times as many workers under the same manager-hour budget gives 29.7 ÷ 5 = 5.94 manager minutes per comparable task for that review component. It assumes the same number and mix of tasks per worker and equivalent standards. Measure the manager's actual time after adding the exception workload, spot checks, and routine upkeep. Report the two views separately. With the same team, how many hours return? With the same manager hours, how many comparable workers could be supervised? Avoid counting one set of recovered hours twice.

For each role, hours recovered equal baseline human hours minus ASU-assisted human hours. The worker result can be negative if taking or preparing proof takes longer. Net team hours are the sum of those differences. Multiply each role's difference by its applicable hourly labor cost to get the capacity value. Add verified costs avoided, such as paid rework avoided, where they are separate from the same hours already counted. Subtract recurring software and operating costs, and account for one-time setup in the first period or use an explicit amortization period. Distinguish the value of time available for other work from cash costs the company actually reduced.

The default wage reference is the latest published U.S. Bureau of Labor Statistics Occupational Employment and Wage Statistics for the San Francisco–Oakland–Fremont metropolitan area. Match the occupation to the work the person does and record the role, area, statistic, reference date, and source. The May 2025 release reports a mean wage of $85.21 per hour for general and operations managers. Use it where that occupation fits, and identify it as a wage benchmark. Choose a closer occupation for a different job. The wage survey measures wage and salary jobs and reports wages separately from employer benefit costs. Look up current estimates from the official series when applying the model.

The benchmark is available at https://www.bls.gov/regions/west/news-release/occupationalemploymentandwages_sanfrancisco.htm and the method at https://www.bls.gov/oes/methods_25.pdf. If authorized company records are available, use comparable base wages to show how company pay differs from the San Francisco role reference. Use the client's fully loaded cost for their own savings estimate when that figure is known. Convert annual wages using the actual paid hours or an explicitly stated standard. Record currency and conversion assumptions where needed. Use actual freelance, vendor, or one-time contract charges for those costs. Collect role totals when individual payroll detail is unnecessary.

Take 300 comparable tasks per month across ten workers as an illustration. Baseline manager review takes 148.5 hours under the owner's starting inputs. At the fivefold target, that review takes 29.7 hours, returning 118.8 hours. At the published management wage example, the gross capacity value is $10,122.95 for those hours. At an assumed $10 per worker per month, the software cost in this illustration is $100. Confirm the price and billing period for an actual proposal. If workers spend an additional 2.5 hours at an illustrative client rate of $20 per hour, deduct $50. The illustrative monthly capacity value is then $9,972.95 before setup and other costs. Label each rate by source.

For fixed manager hours, the same model suggests that a team of ten comparable workers could grow to fifty under the target, with task volume, software cost, and worker effort adjusted for the larger team. Prove the target on the relevant workload before using it as a hiring or budget decision. If the team works on harder cases or sees more unresolved exceptions, measure the new mix and replace the estimate. A limited number of recurring reviews can offer meaningful savings even if the overall capacity target varies.

Run a pilot with comparable records and an agreed sample. Time the existing manual process and ASU-assisted work for the same criteria. Have an authorized reviewer check sampled passes, misses, redos, and unresolved cases against the originals. Record how often the initial result needed a correction and how many tasks met the original deadline. Set the required quality level before starting. Compare manager minutes, net human hours, incorrect passes, open work, worker effort, and cost. Use actual results to decide what to keep, change, or scale.

If wages, task volume, or actual times are missing, give the formula and calculate a labeled scenario with the missing inputs listed. The first assessment can still identify the best task to test. Ask for the specific number or evidence sample that would change the decision. Explain what additional measurement would turn an estimate into a reliable result for this company.

To estimate break-even, divide the recurring fee and added worker cost by the applicable manager hourly cost. The result is the manager hours that need to be recovered before other effects. At the benchmark rate, an assumed ten-dollar monthly fee needs about seven minutes of manager time recovered per worker each month before other costs. Actual break-even changes materially when capture, exception handling, or implementation adds time. Keep those cost inputs visible so the reader can replace them.

# 11. Calculate the opportunity and audit the answer

Find the jobs consuming the most review effort in the reader's business. Use authorized records. In the owner's Codex workspace, use the ASU product review, shared page register and operating records. Compare up to three jobs. Check dated rosters, shifts, task logs and authorized wage or vendor records. Report aggregate counts and keep personal details in their sources.

Count unique active workers and eligible jobs in the same period. Distinguish workers from shifts, test accounts and overlapping assignments. Count shared work once per job. Include relevant employees, freelancers, one-time assignments and vendor agents when supported. Use measured volume first. If only workers and assignment rates are known, multiply them and label the projection. Use the job's actual language, location and schedule.

Record each input with quantity, unit, period, source and status. Use the company's loaded rates where verified and the appropriate San Francisco wage benchmark separately. Treat 30 review minutes and 99 percent reviewed as starting assumptions. Fivefold coverage is a target. Confirm the proposed ten-dollar worker fee and its billing basis.

Calculate baseline and assisted manager hours for the same volume and quality. Include the review fraction, active minutes and follow-up without duplication. Deduct added worker effort and independent audit hours for net team time. Multiply each role's time difference by its applicable rate and subtract recurring costs. Show setup separately. Count cash savings when spending can actually be avoided. Rank by credible net monthly value or by net time when rates are missing. Use a calculator. When key inputs are missing, mark actual savings pending and give at most one labeled scenario with its assumptions.

Create one draft. Delegate two independent read-only reviews to separate subagents when the host supports them. In Codex, use gpt-6.1-sol for both. The evidence and numbers reviewer opens original records, verifies counts and assumptions, and recalculates totals and units. The workflow reviewer checks the company standard, proof, operational fit and capability status against current product sources. Send both the same draft and sources. Ask for verdict, evidence checked and findings. Wait for both. Repair returned findings and obtain both verdicts on the revised draft. State the actual audit result. If independent review is incomplete, show Audit pending. A correct scenario may pass an audit of its arithmetic with assumptions plainly labeled.

Review the relevant distribution, access, partner roles, timing, reminders, proof, criteria, redos, guidance, escalation, appeals, reports and history. Distinguish available functionality, controlled tests, approved designs and unverified routes. Keep delivery and business outcomes separate from completion where applicable. Verify capability claims and the suggested pilot against the exact current source and approval scope.

Compare manual and assisted review on the same pilot cases and criteria. An independent reviewer checks ASU's verdicts against original proof, covering passes, misses, redos and exceptions. Track active time, critical false passes, deadline misses and unresolved jobs. Retain original due times and evidence history. Keep pilot audit effort separate from ongoing checks. Expand when the agreed quality bar holds and net time savings remain after overhead. Use saved test evidence when live use still needs acceptance.

Write the first answer in 250 to 350 words. Lead with the strongest opportunity and the headline savings or the missing inputs. Use one compact table with up to three jobs, active workers, eligible volume, manager hours recovered, net team hours and capacity value. Show the lead job calculation with its sources, a brief pilot and the real audit verdicts. State any missing inputs in the estimate. Finish with one open question, “What would you like to know about how ASU works?” Then offer this next message as a separate action the reader can choose.

> Register my ASU manager account and create my first task. Give me the task link and a message I can send to my worker. Show me where to track progress, review the proof, and request a redo.

Keep the first assessment separate from setup. Opening the brief or asking about a feature leaves the work in assessment mode. Begin registration and creation when the reader sends a clear request to start that setup in ASU.

When the reader asks about a feature, answer for the situation they describe and point to the supported workflow or design. When they request setup, verify the authorized live ASU path and reuse their existing manager account if present. Help them register and complete verification when needed. Ask only for required missing details about the task, worker access, due time, criteria and proof. Create the supported task, then verify the saved settings, task identity and scope of the share link. Give the manager the link and a short message to send. Use the task-only route when applicable and approved. Show where the manager sees open work, submitted proof, targeted corrections and verified completion. Report the actual state of each step. When a route is still a design or test, identify the available step and what remains pending. Treat account access, saved tasks, share links, sent messages and accepted evidence as separate outcomes.

Start the assessment now using the brief and accessible records.

Download the brief Product evidence guide Download case visuals

01Where ASU gives a manager time back
02Product map and evidence
03Apply your company's standards
04Create a task in two minutes and send it
05What the worker sees and how proof is checked
06Manager work, reports, and escalation
07Appeals, unclear evidence, and recovery
08Employees, freelancers, one-time work, and partners
09The closed-loop case library
10Time, cost, and San Francisco wages
11Calculate the opportunity and audit the answer

From copy to a pilot

Copy the brief

Paste into a new chat

Review the estimate

Create your first task

What your AI gets