← Back to Go2 AI Workflow Discovery Pilot
Publishing Desk Discovery • Social media operations example

Elena Cruz is not “just posting.” She is running a publishing desk.

The original version missed this. This role is a braid of content scheduling, community management, approvals, ecommerce context, and whatever “special” request spills into the queue. The authentic story is not “she uses a lot of tools.” The story is that the tools keep breaking the desk into fragments, so the operator has to reassemble the post packet, the reply packet, and the business signal by hand.

Two-week discovery, three-month backstory

The window is still a two-week operator discovery. The difference here is that the page now shows the surrounding 90-day role shift, because the March work only makes sense when you see how sharply the desk moved into content and community work.

Biggest problem

Context does not travel. The system keeps forcing the operator to jump between Sprout, Instagram, Meta, ClickUp, inboxes, files, and sticky notes to finish what is really one unit of work.

14
Day Window
10
Worked Days
79.3h
Session Hours
6,739
Discovery Events
2,142
Alt-Tab / Copy / Paste Bins
20.7h
Recoverable Hrs / Week

90-Day Context

This is the layer the first version was missing. The observed March window is not random. Over the prior 90 days, the role visibly shifted away from order/admin-heavy work and into a true social desk: publishing, community, and always-on spillover.

Role Shift
Januarymore mixed ops load

Admin Support 18.24h • Special Tasks 17.22h • Order Fulfillment 15.59h • Content Scheduling 12.03h • Community Management 5.25h

Februarycommunity starts to rise

Special Tasks 20.34h • Order Fulfillment 14.73h • Community Management 13.98h • Admin Support 9.95h • Content Scheduling 8.98h

Marchpublishing desk goes live

Special Tasks 61.83h • Community Management 25.40h • Content Scheduling 23.78h • Admin Support 7.97h

Special tasks Content scheduling Community management Admin support Order fulfillment
What Changed
March is where the role becomes legible. The prior months show a broader ops/support mix. March shows the actual social-media desk: community up, content up, special-task spill exploding around the desk, admin down. That is the reason to discover this as a publishing/community role instead of a generic admin role.
The real opportunity This operator is already close to content, customer voice, approvals, and demand signal. The value is not just saving her time. It is turning that desk into a cleaner machine so the company gets faster posts, faster replies, and cleaner signal back into merch, paid, and support.

Desk Map

These are the real machines inside the role. The first draft hid this under generic findings. The actual story is three overlapping desks that keep borrowing memory from each other.

Machine 1

Publishing Machine

Posts are being assembled as a shuttle between scheduling, assets, notes, and native-channel checks.

Sprout Calendar252 events • 3.04h in the window
128 jumps out to files
Windows Explorer91 returns back into the calendar • 46 returns into Instagram
97 jumps into side-memory
Sticky Notes72 returns into Sprout calendar • 34 returns into Instagram
then back to channel surface
Instagram149 scheduling events • 2.67h
Machine 2

Community Machine

Replies are live, fragmented, and partially visible in multiple queues instead of one clean response desk.

Sprout Smart Inbox54 events • 1.57h
queue splits
Meta Business Suite92 events • 3.01h
native check
Instagram25 community events • 0.52h
needs routing
ClickUp / inbox / support follow-upCommunity Management still carries 224 Alt+Tab bins because outcomes are not landing cleanly.
Machine 3

Spillover Machine

“Special Tasks” is the place unfinished systems go to die, and then come back as interruptions.

ClickUp projects + stakeholder threads3.12h around one stakeholder view • 1.32h in projects list
77 jumps out
Explorer / task switching68 returns into ClickUp projects list
detours into ad hoc work
Google Forms / Shopify / Ads / inboxactual titles include Bold Upsell, Shopify orders, Google Ads negative keywords, mail follow-ups

Daily Shape

This operator is not drifting. The day is structurally late, steady, and full. The inconsistency is what kind of desk wins each day.

Day Strips
Mar 12
6.7h
Mar 13
8.0h
Mar 16
7.8h
Mar 17
8.1h
Mar 18
7.9h
Mar 19
8.0h
Mar 20
7.9h
Mar 23
6.2h

The day strips make the role feel real in a way the first version did not. March 17 is mostly community. March 20 is mostly spillover. March 23 looks like a true publishing day.

Hour Of Day
14:00
7.09h
15:00
7.69h
16:00
8.30h
17:00
8.09h
18:00
9.34h
19:00
6.45h
20:00
7.46h
21:00
6.09h
This is a late-day desk. The system should be built for a heavy 16:00-21:00 publishing and reply lane. That means queue prep, asset prep, and approval prep should happen before the live desk peaks, not during it.

Receipts, Boundaries, And One Important Non-Problem

This is the anti-bullshit layer. The frontier review was right that the page needed a harder line between what was directly observed, what was inferred, and what was already working well enough that it should not be mislabeled as a failure.

Observed directly
Telemetry receiptWhat was literally seen
6,739 app/web eventsWindow contained sustained work across Sprout, Instagram, Meta Business Suite, ClickUp, inboxes, files, and notes.
79.3 session hours / 10 daysLate, steady, full desk. This was not a “spotty attention” or attendance problem.
1,155 Alt+Tab / 490 copy / 497 paste binsReal context-shuttle load was visible in the populated keystroke days.
Title-level evidence`Calendar | Publishing | Sprout Social`, `Smart Inbox | Sprout Social`, `Meta Business Suite`, `Instagram`, stakeholder threads, Bold Upsell, Google Ads, and inbox titles all appeared in the same operating lane.
Transition chainsExplorer ↔ Chrome, ClickUp ↔ Explorer, and Notes ↔ browser loops were repeated often enough to reconstruct the missing work object.
Inferred carefully

The report does make interpretation moves, but they are tied to direct receipts rather than vibes:

  • Publishing rebuild loop is inferred from Sprout → Explorer → Notes → back-to-Sprout movement, not from generic “content is hard” language.
  • Reply queue split is inferred from Smart Inbox + Meta + Instagram + follow-up surfaces behaving like multiple live desks.
  • Signal loss is an inference from community work sharing a lane with Shopify, Google Ads, inboxes, and follow-up titles.
  • 20.7h/week recoverable is a modeled estimate, not a literal timesheet subtraction. It is the combined recovery case for publishing rebuild, reply split, spillover gating, and signal routing.
What was already working Once the operator was inside the real work surfaces, she could stay there for meaningful blocks. Sprout calendar alone carried 252 events and 3.04h in the discovery window. The problem is not inability to focus. The problem is that the packet is incomplete, so the desk has to keep leaving itself.

Where The Extra Steps Actually Are

This is the first-principles layer the previous version lacked. These are not generic “findings.” These are the real places where the desk is making extra moves because the system is incomplete.

Publishing rebuild loop • 4.4 hrs / week

Sprout calendar is not carrying the full post packet.

The operator keeps leaving the scheduler to find files, notes, and native-channel context. The loop is visible in the transitions, not just the hours.

  • 128 transitions from Sprout calendar into Windows Explorer during Content Scheduling
  • 97 jumps from Windows Explorer into Sticky Notes during Content Scheduling
  • 72 returns from Sticky Notes back into Sprout calendar

What this means: the operator is rebuilding the caption/asset packet instead of reviewing one completed packet.

Reply queue split • 3.6 hrs / week

The community desk is spread across three live surfaces and at least one hidden follow-up surface.

Smart Inbox, Meta Business Suite, Instagram, and follow-up routing are behaving like separate queues instead of one response system.

  • Smart Inbox 1.57h • Meta Business Suite 3.01h • Instagram community surface 0.52h
  • Community Management still carries 224 Alt+Tab bins
  • Explorer still returns into Meta during community work, which means reply context or assets are leaving the desk

What this means: response quality depends too much on what the operator remembers right then.

Spillover lane • 7.1 hrs / week

“Special Tasks” is not a job. It is a queue of missing systems.

March is dominated by special-task work because every broken edge case, stakeholder ask, or adjacent ops request falls into the same body.

  • 61.83h of March time sits under Special Tasks
  • Actual titles include Shopify orders, Bold Upsell, Google Ads negative keywords, Google Forms, stakeholder threads, and inbox follow-up
  • Projects list ↔ Explorer bounce is one of the strongest repeated transition patterns in the whole window

What this means: the desk is subsidizing other unfinished machines, which makes content work feel more chaotic than it actually is.

Signal loss • revenue upside

The operator is sitting on market signal the business is not harvesting cleanly.

By March, this desk is close to customer questions, comment patterns, creator interest, product friction, and ad objections. Without a signal router, all of that stays trapped in labor.

  • March combines community surfaces with Shopify, Google Ads, inboxes, and follow-up work in the same operator lane
  • This is exactly where buying signals, creator leads, and product objections show up first
  • The current system mostly asks the operator to react, not to route

What this means: the ROI is not only time. It is faster conversion feedback and cleaner demand signal back into the company.

What We’d Install First

Not six repeated recommendation cards. Four actual installs, in order, because this role does not need more commentary. It needs fewer resets.

Install 1

Publishing Desk Pack

Create one packet per post: caption, cutdowns, asset links, reply angle, approval owner, publish slot. The operator edits the pack instead of assembling it live.

Removes: Sprout → Explorer → Notes → Instagram rebuild loop.

Install 2

Community Queue Copilot

Batch comments and DMs into one review flow that drafts replies, flags escalations, and surfaces business signal. The desk becomes review-first instead of reply-from-scratch.

Removes: repeated rebriefing across Smart Inbox, Meta, and Instagram.

Install 3

Signal Router To ClickUp

Buying signals, creator leads, approval blocks, and product issues become tasks or notes instead of dying in inboxes or memory.

Removes: the operator’s job as the manual bridge between audience signal and the rest of the business.

Install 4

Special-Task Gate

Every adjacent ask gets triaged before it lands on the desk. If it is not publishing, community, or routed signal, it needs an owner or a different queue.

Removes: invisible scope creep disguised as “quick asks.”

Monday test

Use it on one real day

Run the next publish day and the next comment batch through the tool, then route only the important signals out. That is enough to prove the value without boiling the ocean.

7-day result

What should change fast

Fewer missing assets, cleaner replies, fewer approval misses, and visible signal leaving the social desk instead of being trapped inside it.

One Concrete Before / After Packet

The frontier critique was right that the page needed one testable work object. This is the object the current system keeps scattering, and the exact thing the install should reunify.

Current state

What one post looks like before the install

  • Brief starts in ClickUp or a stakeholder thread.
  • Asset link gets pulled from Explorer or a side folder.
  • Caption fragments live in Sticky Notes or inbox.
  • Instagram or Meta gets opened for native-surface sanity checking.
  • Approval status still lives in email or memory.
  • Replies and business signal arrive later with no durable route out.
Installed state

What one post should look like after the install

  • Brief: angle, offer, and constraints in one object.
  • Assets: final links and platform-ready variants attached.
  • Caption pack: primary caption, cutdown, story frame, first reply angle.
  • Approval owner: one visible yes/no state, not an inbox hunt.
  • Publish slot: scheduler-ready date, channel, and owner.
  • Signal route: support, creator, buying, or product friction has a next destination.
Why this makes the page more valuable A similar operator can now look at the packet and say, “Yes, that is the missing thing in my day,” instead of only agreeing with the findings in the abstract.

Built Artifacts

This is the part a similar operator would actually steal. One tool they can use now, one admin handoff for the stronger version, and one real connector starter if they want the desk wired into ClickUp.

Use now

Social Community Command Center

Paste real comments, caption drafts, campaign notes, or approval threads. Get reply batches, caption variants, escalations, and a clean system update back.

Open the tool
Send to admin

Install note

If the operator should not touch setup, give the install note to an admin. They can paste it into Codex or Claude Code and wire the helper without making the operator play IT.

Open install note
Power path

ClickUp Social Router MCP

Real starter connector that can create a signal task, add a comment, and set a custom field in ClickUp. This is the bridge between the social desk and the operating system.

Read the connector View server

Sources

This version is intentionally more literal about what it came from: app/web telemetry, session boundaries, task categories, hour-of-day shape, and title-level workflow evidence across both the two-week window and the prior 90 days.

Telemetry surfaces used
SurfaceWhat it proved
App / web eventsWhich tools actually own the day, and how the role shifted over 90 days.
Window titlesWhat the operator was really doing: Sprout calendar, Meta, Instagram, projects lists, Shopify, Google Ads, stakeholder threads.
SessionsSteady late-day desk with 79.3 session hours across 10 worked days.
Keystroke patternsContext shuttle load: 1,155 Alt+Tab bins, 490 copy bins, 497 paste bins in the populated days.
Public-safety note

Identity, company, and client-specific details were changed for public use. The work shape, counts, and role logic are real.

ClickUp API docs Apps Script docs Sheets API docs