Operations Intelligence Workbench

Validate operations data, research actions, and schedule a ChatGPT Plus monitor.

Introduction

30 Second Summary

A polished chart can feel convincing even when its spreadsheet contains mistakes. Without a clear trail from raw data to evidence, a quick answer can send a decision in the wrong direction.

In this project, you will build an Operations Intelligence Workbench in ChatGPT Plus using a ChatGPT Project, synthetic data, Data Analysis, and Deep Research. You will turn a messy operating-spend file into checked metrics, cited recommendations, and a reusable decision brief.

What You'll Build

You open one ChatGPT Project to inspect a cleaned spend table, review a department chart, read cited research, and retrieve a one-page management brief from the same workspace.

By the end of this project, you'll have:

A before-and-after dashboard where you can compare a misleading first pass with results based on checked data.
A validated operating-spend view with 15 cleaned rows. You can explain each correction. You can confirm a $13,450 total plus 7.4 average processing days.
A source-backed decision package that traces three recommended actions from internal evidence to external citations. You can retrieve the saved brief from a fresh project chat.
Secret Mission: A weekly monitor that checks official pages for changes to the ChatGPT Plus capabilities used by your workbench.

Are there any prerequisites?

You need an active ChatGPT Plus subscription, a Windows device, internet access, and permission to install the ChatGPT desktop app. Use only the provided synthetic dataset. Keep confidential employer data out of this project.

Before We Start

Before any hands-on work begins, commit to a safe Operations Intelligence Workbench that uses only synthetic data. The reusable ChatGPT Project will keep raw analysis, cleaned evidence, and external research separate so its validated KPIs, cited recommendations, and decision brief remain inspectable.

Set Up a Safe Windows Workspace

Your workbench plan already protects the boundary between each evidence stage. This step gives that plan a safe starting point on Windows.

First, the ChatGPT desktop app needs to be ready for local-file work through your ChatGPT Plus account.

You will also review how new chats are handled. The supplied synthetic data keeps employer information outside the workflow.

In this step, get ready to:

Prepare the ChatGPT desktop app for your active Plus account.
Choose your personal data-control preference.
Create the untouched synthetic operations dataset.

Install the ChatGPT desktop app

The app may already be installed on your Windows device. A quick search identifies the correct setup path without creating a duplicate installation.

Press the Windows key to open Windows search.
Type ChatGPT and press Enter.

✔️ ChatGPT opens and I am signed in

Your desktop setup is ready. The app is open under your active ChatGPT Plus account.

ⓧ The app opens at sign-in

The app is installed. It still needs your Plus account before you can use the project features in later steps.

Complete the sign-in flow with the account that has ChatGPT Plus.
Return to the desktop app after authorization completes.

You should see the main ChatGPT conversation interface.

ⓧ ChatGPT is not installed

Open the official ChatGPT quickstart in your browser.
Select Download ChatGPT.
Run the downloaded installer from your browser's downloads panel.
Complete the installation prompts.
Press the Windows key to open Windows search.
Type ChatGPT and press Enter.
Sign in with the account that has ChatGPT Plus.

That setup hurdle is cleared. Your Plus workspace is now open for local-file work.

Still Can't Open the App?

Confirm that the installer finished before searching for ChatGPT again.
Use the official quickstart link above if the download was interrupted.
Ask for help with your specific setup issue: Help me install and open the ChatGPT desktop app on Windows.

Review your data controls

This privacy choice deserves a deliberate pause. It determines whether new conversations from your personal workspace can be used to improve OpenAI models.

Chat history stays available when the setting is off. New conversations are excluded from model training in that state.

Open the account menu in the ChatGPT desktop app.
Select Settings.
Select Data controls.
Review Improve the model for everyone.
Choose the position that matches your privacy preference.
Select Done.

Why Keep Using Synthetic Data?

Your privacy preference controls how new conversations may be used. Synthetic practice data also keeps confidential employer information out of the workbench.

You will use the supplied fictional dataset throughout this project.

Create the synthetic dataset

A CSV stores a table as plain text. This practice file contains intentional irregularities that must remain untouched for the later audit.

Start a new regular Chat from the ChatGPT desktop app.
Paste the table request into the message composer using this text:

Create an interactive table from the CSV below. Preserve every value and every row exactly. Do not clean, normalize, infer, or correct anything. Let me download the resulting table as CSV.

Why Preserve Every Value?

This instruction establishes the file as raw evidence. Every later cleaning decision can be compared with the untouched source.

Paste the synthetic CSV below the request using this data:

record_id,date,department,vendor,category,amount_usd,processing_days,status
OP-001,2026-09-01,Engineering,CloudHost,Cloud,850,3,Complete
OP-002,2026-09-02,Research,DataSource,Subscription,1200,5,Complete
OP-003,2026-09-03,Operations,OfficePro,Supplies,320,2,Complete
OP-004,2026-09-04,Finance,LedgerWorks,Software,600,4,Complete
OP-005,2026-09-05,Engineering,CloudHost,Cloud,920,6,Completed
OP-006,2026-09-06,,TravelCo,Travel,780,9,Pending
OP-007,2026-09-07,Research,DataSource,Subscription,1200 USD,5,Complete
OP-008,2026-09-08,Operations,OfficePro,Supplies,-250,1,Complete
OP-009,2026-09-09,Finance,AuditAssist,Consulting,1500,12,Pending
OP-010,2026-09-10,Engineering,DevTools,Software,450,3,Complete
OP-011,2026-09-11,Research,LabMart,Equipment,2400,15,Complete
OP-012,2026-09-12,Operations,ShipFast,Logistics,700,7,Complete
OP-013,2026-09-13,Finance,LedgerWorks,Software,600,4,Complete
OP-013,2026-09-13,Finance,LedgerWorks,Software,600,4,Complete
OP-014,2026-09-14,Engineering,CloudHost,Cloud,980,30,Complete
OP-015,2026-09-15,Research,DataSource,Subscription,1200,5,complete

What Does This Data Contain?

The table contains fictional operations records with deliberate data-quality problems. The exact values give every learner the same raw starting point.

Keeping the irregularities intact makes the later dashboard test meaningful.

Send the message.

You should see an interactive table containing every supplied row. The irregular values should remain untouched.

Use the table download control in the top-right corner.
Save the downloaded file locally as operations_sample.csv.

Good work. You now have the untouched local evidence that drives the rest of the workbench.

Can't Download the Table?

Confirm that ChatGPT generated an interactive table from the supplied CSV.
Repeat the request in a new regular Chat if any value was changed.
Ask for help with the table output: Help me generate and download an unchanged CSV table in ChatGPT.

Before the final check, what file dimensions do you expect ChatGPT to report?

Start another new regular Chat.
Use the file attachment control in the message composer to attach operations_sample.csv.
Paste the verification question using this text:

How many data rows and columns are in this file?

What Does This Check Prove?

The question checks the saved file's shape before any analysis begins. A matching result confirms that the expected records and fields reached the local CSV.

Send the verification question.

You should see 16 data rows and 8 columns. The eight columns are record_id, date, department, vendor, category, amount_usd, processing_days, and status.

Seeing Different Dimensions?

Confirm that the attached filename is operations_sample.csv.
Check that ChatGPT counted data rows separately from the header row.
Regenerate the table from the exact request above if the file has fewer records or fields.
Ask for help comparing the file shape: Help me find why my CSV row or column count differs.

Your Windows workspace is ready with a deliberate privacy choice and an untouched synthetic dataset. Next, you will place that evidence inside a reusable project and test a fast dashboard against it.

Build an Unchecked Dashboard

You have the safety groundwork in place. Your synthetic dataset is saved as operations_sample.csv.

A polished dashboard can still mislead when its source contains unchecked defects. This step deliberately asks ChatGPT to analyze the raw file so you can see that gap before trusting its output.

In this step, get ready to:

Create an Operations Intelligence Workbench project governed by cautious analyst instructions.
Establish operations_sample.csv as the project's confirmed raw source.
Generate an unchecked dashboard that exposes questionable results.

Create the workbench project

A ChatGPT Project keeps related chats, files, and instructions together. Project instructions give every conversation in this workbench the same analytical guardrails.

In the ChatGPT desktop app from earlier, select New project in the sidebar.
Enter Operations Intelligence Workbench as the project name.
Submit the project name to create the project.

You should see Operations Intelligence Workbench in the sidebar.

Select the project's more options menu shown as (•••).
Select Project settings.
Set the project's analyst behavior by adding this instruction:

You are a cautious operations analyst. Separate observed data from assumptions. Never silently clean data. Show calculations, flag data-quality problems, and cite external claims.

What Does This Instruction Change?

The opening sentence gives ChatGPT a cautious operations role.
The next two sentences force a visible distinction between evidence and assumptions.
The final sentence requires transparent calculations, explicit warnings, and citations for external claims.

Leave Project settings after the full instruction appears.
Confirm the sidebar still shows Operations Intelligence Workbench.

Your workbench now has a persistent analytical guardrail. Every project chat starts from the same cautious standard.

Project Instruction Missing?

Return to Project settings to compare the saved instruction with the prompt above.
Check that the instruction sits inside Operations Intelligence Workbench.
Help me check why my ChatGPT Project instruction is missing.

Confirm the raw project source

The saved CSV is your raw evidence. Confirming its structure inside the project proves that ChatGPT can access the intended file before any analysis begins.

Select the project's file upload control.
Choose operations_sample.csv from the folder where you saved it earlier.

You should see operations_sample.csv listed among the project sources.

Start a new chat inside Operations Intelligence Workbench.
Verify the raw file structure by sending this prompt:

Confirm the filename, data row count, column count, and column names in operations_sample.csv before analyzing it.

What Does This Check Prove?

The filename confirms that the chat is using the intended project source.
The dimensions reveal whether every raw record and field survived the upload.
The column list establishes the available evidence before calculations begin.

You should see operations_sample.csv with 16 data rows and 8 columns.

The listed columns should be record_id, date, department, vendor, category, amount_usd, processing_days, and status.

File Details Do Not Match?

Confirm that the uploaded filename is operations_sample.csv.
Remove any altered copy from the current chat before selecting the unmodified file from earlier.
Help me diagnose why my uploaded CSV dimensions do not match.

Generate the unchecked dashboard

This first dashboard uses the uploaded evidence exactly as it stands. Its polished presentation gives you something concrete to question.

Before you send the dashboard request, do you expect every KPI to be dependable when the file is used exactly as uploaded?

Generate the unchecked dashboard by sending this prompt:

Create a quick executive dashboard from operations_sample.csv. Show total amount, average processing days, spending by department, and a bar chart. Do not clean or transform the data first. State any calculation problem you encounter.

What Does This Prompt Test?

The request asks for the headline metrics an executive dashboard commonly summarizes.
The no-cleaning constraint preserves every raw defect during calculation.
The final sentence makes calculation problems visible instead of allowing silent corrections.

You should see a KPI summary with total amount, average processing days, and spending by department. You should also see a department bar chart.

The response should flag the mixed amount_usd formatting, the duplicated OP-013, or both. That visible uncertainty is the intended result.

You have caught the important gap: polished output can still rest on questionable evidence.

Review the displayed total amount for a missing, inconsistent, or inflated result.
Review the average processing days for any stated limitation.
Review the department spending chart for effects from the raw file.
Keep this first-pass dashboard in the project chat as raw-analysis evidence.

Dashboard or Warning Missing?

Confirm that the dashboard uses operations_sample.csv from the project sources.
Check that your prompt includes the instruction to state calculation problems.
Help me understand why my unchecked dashboard lacks a chart or data-quality warning.

Your first-pass dashboard is visible, but its evidence is still untrusted. Next, you will make each data-quality decision explicit before relying on the metrics.

Audit and Clean the Data

Your first-pass dashboard showed how quickly unchecked data can produce a polished result. Its questionable metrics cannot support an operational decision yet.

A data-quality audit records each defect before any value changes. You will apply a narrow cleaning policy to the raw CSV so every corrected metric has a traceable explanation.

In this step, get ready to:

Document every defect in the raw dataset.
Apply only the specified cleaning rules.
Save the validated dataset as reusable project evidence.

Audit the raw dataset

The audit separates observed defects from cleaning decisions. This preserves operations_sample.csv as your untouched source of truth.

Document the raw-data defects by sending this prompt in the project chat that contains your first-pass dashboard:

Audit this dataset before analysis. List duplicate IDs and duplicate rows, missing values, inconsistent status labels, nonnumeric amount values, negative amounts, and processing-day outliers. Do not change anything yet. Show the affected record IDs.

What does this prompt do?

The opening instruction tells ChatGPT to inspect the raw dataset before calculating new metrics.
The requested checks cover duplicate records, missing fields, inconsistent labels, invalid amount formatting, refunds, and unusually long processing times.
The final instruction keeps the audit traceable by requiring the affected record IDs.

Compare the audit with the affected records in operations_sample.csv.

You should see OP-013 identified as a duplicate ID and an exact duplicate row. OP-006 should have a blank department value.

The audit should flag 1200 USD on OP-007 as nonnumeric. It should preserve -250 on OP-008 as a negative amount that needs an explicit interpretation.

The status variants should include Completed on OP-005 plus complete on OP-015. The 30 processing days on OP-014 should appear as an outlier.

Good catch. You now have a complete defect log without altering the raw evidence.

Missing an expected defect?

Confirm that the chat is analyzing operations_sample.csv instead of a generated table from the first dashboard.
Ask ChatGPT to display the affected rows if the audit names a defect without its record ID.
Use this for focused help: Help me compare my ChatGPT data-quality audit with the known defects in operations_sample.csv.

Apply the documented cleaning rules

A cleaning policy turns each audit finding into a visible decision. The policy preserves the refund and the processing-time outlier because both contain operational information.

Before you send the cleaning request, what row count do you expect after one exact duplicate is removed?

Create the cleaned table by sending this prompt in the same project chat:

Create a cleaned table using these rules: remove one exact duplicate OP-013 row; set the blank department to Unassigned; parse 1200 USD as numeric 1200; preserve -250 and add record_type as Refund for negative amounts and Expense otherwise; normalize Completed and complete to Complete; keep 30 processing days and add review_flag as Investigate where processing_days is greater than 20, blank otherwise. Then calculate total amount, average processing days, and spending by department. Show a department bar chart. Do not make any other changes.

What does this cleaning policy do?

The duplicate rule removes one repeated OP-013 row while retaining the original record.
The missing-value rule assigns Unassigned to OP-006 so the amount remains visible in department reporting.
The amount rules convert 1200 USD to a numeric value while preserving -250 as a refund.
The review rule retains the 30 processing days on OP-014 while marking the record for investigation.

Inspect the generated table against the documented cleaning rules.

The table should contain one OP-013 row. OP-006 should show Unassigned as its department.

The amount for OP-007 should be numeric 1200. OP-008 should retain -250 with Refund in record_type.

Every nonnegative row should show Expense in record_type. The status variants should be normalized to Complete.

The review_flag value for OP-014 should be Investigate. Every other review_flag value should be blank.

Compare the generated KPI summary with the validated results.

You should see 15 rows. The total amount should be $13,450.

The average processing time should be 7.4 days. The department totals should be Research $6,000, Engineering $3,200, Finance $2,700, Operations $770, and Unassigned $780.

Inspect the department bar chart against the validated department totals.

You should see Research as the largest bar. Operations should be the smallest bar.

Seeing different totals?

Check that only one exact OP-013 row was removed.
Confirm that -250 remains in the total as a refund.
Request a corrected table if ChatGPT changed any value outside the documented rules.
Use this for focused help: Help me find why my cleaned operations dataset does not match the expected KPIs.

Save the cleaned evidence

The cleaned table becomes a separate project source so future analysis can use validated evidence. The raw file remains available for comparison.

Use the table download control to download the final cleaned table.
Save the downloaded file as operations_cleaned.csv.
Upload operations_cleaned.csv with the project file control.
Check the project sources list.

You should see operations_sample.csv plus operations_cleaned.csv listed as separate sources.

Do not see the cleaned file?

Check your Downloads folder for operations_cleaned.csv.
Confirm that the uploaded filename ends with .csv.
Use this for focused help: Help me add my downloaded operations_cleaned.csv file to my ChatGPT Project sources.

Before you validate the saved file, do you expect its metrics to match the cleaned table exactly?

Validate the uploaded file by sending this prompt in the project chat:

Validate operations_cleaned.csv against the documented cleaning rules. Report the row count, total amount, average processing days, and spending by department. Create a department bar chart from the validated department totals. List every transformation applied and confirm whether any other values changed.

What does this validation check?

The filename directs ChatGPT to the saved cleaned source instead of the raw dataset.
The requested metrics test whether the downloaded file preserved the validated calculations.
The transformation list makes every change auditable.
The final confirmation checks that no undocumented cleaning slipped into the file.

You should again see 15 rows with a total amount of $13,450. The average processing time should be 7.4 days.

The department totals should remain Research $6,000, Engineering $3,200, Finance $2,700, Operations $770, and Unassigned $780.

The new department bar chart should reflect those same totals. This confirms that the reusable file matches your documented cleaning policy.

Does the saved file fail validation?

Confirm that ChatGPT analyzed operations_cleaned.csv.
Download the corrected table again if the saved file contains an earlier result.
Use this for focused help: Help me compare operations_cleaned.csv with the documented cleaning policy and validated KPIs.

That is the evidence layer locked down. Next, you will connect these validated internal results to current external research.

Research External Evidence

Your cleaned dataset now explains what happened inside the fictional organization. It cannot show whether a recommendation reflects current external guidance.

In this step, you'll use Deep Research to add current external evidence to the workbench. You'll keep that evidence separate from the cleaned internal data.

In this step, get ready to:

Run a Deep Research task using operations_cleaned.csv as internal evidence.
Inspect three recommendations that separate internal evidence from external evidence.
Verify two citations before accepting the report.

Start a Deep Research chat

Deep Research can combine an uploaded file with current public web sources. Its proposed research plan gives you a chance to check the evidence strategy before the research proceeds.

Return to the Operations Intelligence Workbench project from earlier.
Start a new chat inside the project.
Select Deep research from the tools menu (+).
Send the research request by pasting this prompt:

Using operations_cleaned.csv as the internal evidence, research current best practices for reducing software and cloud operating spend and shortening purchase-approval cycle time in small software and research organizations. Prioritize primary and authoritative sources. Recommend three actions. For each action, separate our internal evidence from external evidence, cite the external source, state assumptions, and explain how we would measure success. Do not invent a benchmark when no authoritative one exists.

What does this request do?

The request treats operations_cleaned.csv as internal evidence.
It directs the web research toward primary and authoritative sources.
It requires three actions with separate internal evidence and external evidence.
It blocks unsupported benchmarks from entering the recommendations.

Review the proposed research plan before the research begins.
Confirm the plan uses operations_cleaned.csv as uploaded file evidence.
Confirm the plan searches the public web for current external evidence.
Request a plan change if either evidence stream is missing.
Allow the research to proceed after the plan reflects both evidence streams.

You'll see the proposed plan identify the uploaded file and the external research scope. This proves that the task has both evidence streams before it starts building recommendations.

Don't see the right research plan?

Confirm that you selected Deep research before sending the request.
Refer to operations_cleaned.csv by its exact filename inside the request.
Help me fix a Deep Research plan that is missing my uploaded file or external web research.

Inspect the completed report

A completed Deep Research report provides a table of contents with citations or source links. Its sources-used section makes the external evidence traceable.

Open the completed report in the project chat.
Confirm that the report has a table of contents.
Confirm that the report contains citations or source links.
Confirm that the report has a sources-used section.
Count the three recommended actions.
Check that each action separates internal evidence from external evidence.
Check that each action states its assumptions.
Check how each action measures success.

You should see three actions for reducing software and cloud operating spend or shortening purchase-approval cycle time. Each action should expose the evidence behind it.

Is the report missing a required section?

Ask Deep Research to add the missing section without changing supported findings.
Ask Deep Research to label internal evidence and external evidence separately for each action.
Help me revise a Deep Research report that is missing required evidence sections.

Verify sources and revise weak claims

Before you open the first source, do you expect every citation to support the specific claim beside it?

Choose a citation beside one recommended action.
Open that citation.
Compare the source evidence with the nearby claim.
Return to the Deep Research report.
Repeat the source check with a citation beside a different recommendation.

What are you testing?

A nearby citation should substantiate the specific claim it follows. A source that only discusses the broad topic leaves the recommendation unsupported.

Checking the source itself protects the brief from polished advice that lacks usable evidence.

Confirm that both checked sources are primary or authoritative.
Ask Deep Research to revise any recommendation whose source does not substantiate it.
Open any replacement citation after a revision.
Confirm that the replacement source supports the revised claim.
Scan the final report for any unsupported benchmark.

You should finish with at least two checked citations. Any recommendation that lacked support should now be revised.

The final report should retain three actions with distinct evidence lanes, stated assumptions, and measurable success criteria.

Does a citation miss the claim?

Ask Deep Research to revise the recommendation using evidence that directly supports its claim.
Recheck the replacement citation before accepting the revision.
Help me assess whether a citation supports a Deep Research recommendation.

You've closed the trust gap. Three recommendations now connect cleaned internal data to checked external sources.

Next, you'll turn this evidence into a concise decision brief that another decision-maker can inspect and reuse.

Save a Decision Brief

Your ChatGPT Project now holds validated internal evidence. It also holds a source-checked Deep Research report.

A decision-maker needs a concise recommendation they can inspect without rebuilding your workflow. This step turns your evidence into a reusable one-page brief.

In this step, get ready to:

Generate a one-page brief from the cleaned analysis plus the Deep Research report.
Review the brief against the evidence requirements.
Prove the saved brief is reusable from a fresh project chat.

Create the one-page brief

A decision brief compresses detailed analysis into a recommendation someone can act on. Its structure must preserve the boundary between internal results and external claims.

In the open Operations Intelligence Workbench project, start a new project chat for the brief.
Create the brief by pasting this request into the message box:

Create a one-page executive decision brief using the cleaned analysis and the Deep Research report. Include an executive summary, the verified KPIs, three findings, three prioritized actions, evidence for each action, assumptions and limitations, an owner and next checkpoint for each action, and a clear separation between internal data and external research.

What does this request produce?

The executive summary gives a decision-maker the conclusion before the detail.
The evidence sections preserve the boundary between internal results and external claims.
Owners plus checkpoints turn recommendations into accountable next steps.

Submit the request.

You should see a one-page executive brief built from the cleaned analysis plus the research report. The response should contain three findings with three prioritized actions.

Brief missing sections?

Ask ChatGPT to rewrite the response as one page if it produced a long report.
Tell ChatGPT to use the cleaned analysis for internal figures if a KPI is missing.
Tell ChatGPT to use the Deep Research report for external claims if citations are missing.
Help me correct my executive decision brief.

Check the evidence trail

A polished brief can still carry a wrong figure or an unsupported claim. This review keeps the approved version tied to the evidence you already validated.

Confirm the brief opens with an executive summary.
Confirm the verified KPI section shows $13,450 total amount plus 7.4 average processing days.
Confirm the brief contains three findings.
Confirm the brief contains three prioritized actions.
Confirm the brief treats -250 as a Refund without deleting the record.
Confirm OP-014 keeps 30 processing days with an investigation flag.

Why check these details?

The refund and outlier show whether the brief preserved your documented cleaning policy. A decision-ready summary keeps unusual records visible when they affect interpretation.

Confirm each action separates internal evidence from external evidence.
Confirm every external claim has a citation.
Confirm the brief includes assumptions plus limitations.
Confirm each action names an owner.
Confirm each action names a next checkpoint.
Ask ChatGPT to revise every mismatch against this checklist before you continue.

Your approved brief now traces every recommendation back to cleaned internal evidence or cited external research. It is ready to become a reusable source.

Save the brief and test reuse

A response saved to project sources becomes part of the workbench context. A fresh-chat test checks whether the recommendation can be retrieved consistently.

Open the response message menu on the final brief.
Select Add to project sources.

That is the reuse layer in place. Your executive decision brief is now stored with the project evidence.

Start a fresh chat inside the Operations Intelligence Workbench project.

Before you test it, what details should a fresh chat recover if the saved source is working?

Test the saved source by asking this question:

From the saved project sources, what are the three recommended actions and the evidence for each?

What does this check prove?

The question directs ChatGPT to use the saved project sources.
A consistent response shows that the decision brief can be reused without rebuilding the analysis.

You should see the same three recommended actions with supporting evidence from the saved sources. The response should use the cleaned KPIs plus cited research consistently.

That's the workflow made reusable. A new decision-maker can now inspect the same recommendation from the saved project context.

Missing actions or evidence?

Confirm the fresh chat remains inside the Operations Intelligence Workbench project.
Return to the final brief if it was not saved.
Reopen the response message menu.
Select Add to project sources.
Help me verify my saved decision brief.

Secret mission

Schedule a Weekly Capability Monitor

Your Operations Intelligence Workbench depends on ChatGPT Plus capabilities that can change over time. Create a weekly monitor that checks official OpenAI sources for relevant updates without accessing your project files.

Clean Up Your Resources

These resources add no separate charge beyond your existing ChatGPT Plus subscription. Choose how much of the workbench you want to keep.

Resources you used:

The ChatGPT Project named Operations Intelligence Workbench. Deleting it also removes its project chats, instructions, uploaded sources, saved executive decision brief, and Deep Research report.
The weekly ChatGPT capability monitor Scheduled task.
The local operations_sample.csv file.
The local operations_cleaned.csv file.

Keep everything running

No action is needed. Choose this if you want to keep using the workbench or its weekly monitor.

Leave the Operations Intelligence Workbench project available for future analysis.
Keep the weekly capability monitor active for its Monday 9:00 AM checks.
Retain operations_sample.csv as the unchanged synthetic source file.
Retain operations_cleaned.csv as the validated analysis file.

Pause - I'll come back to this later

Pause the weekly monitor to stop its recurring checks. Your ChatGPT Project and both local CSV files remain available.

Return to Scheduled in ChatGPT.
Select the weekly ChatGPT capability monitor.
Use Pause in the task panel.

That gives you a clean stopping point. The project and its sources stay ready for your return.

Delete - I don't want to use this again

Deleting the project is permanent, so this option is for a full reset. Its project chats, instructions, and project-only files cannot be recovered.

Return to Scheduled in ChatGPT.
Select the weekly ChatGPT capability monitor.
Use Delete in the task panel.
Complete the task deletion confirmation.
Switch back to Operations Intelligence Workbench.
Open the project's more options menu.
Select Delete project.
Complete the permanent deletion confirmation.

The online resources are now removed. Your two local CSV files are separate, so they remain on Windows.

Use Windows search to locate operations_sample.csv.
Delete operations_sample.csv from its saved location.
Use Windows search to locate operations_cleaned.csv.
Delete operations_cleaned.csv from its saved location.
Search Windows again for operations_sample.csv.

You should see no remaining copy in the locations you searched.

Search Windows again for operations_cleaned.csv.

You should see no remaining copy in the locations you searched. Your Operations Intelligence Workbench resources are now cleaned up.

Nice Work!

You did it! Your Operations Intelligence Workbench is now a reusable ChatGPT Project that turns synthetic files into evidence a decision-maker can inspect.

You've learned how to:

Turn an untrusted 16-row dataset into a validated 15-row dataset through a documented data-quality audit.
Produce a corrected department-spend chart from the verified total of $13,450, the 7.4-day average processing time, and validated department totals.
Build a reusable executive decision brief that keeps cleaned internal evidence separate from source-checked Deep Research.
Secret Mission: Create a weekly web-only Scheduled task that checks official sources for changes to the ChatGPT Plus capabilities used in your workbench.

Ready to quiz yourself?