BH Technology Book a consultation

Back to Capacity

Department of 10 · Setup 1 of 5

Shared data

Can everyone see the same customer or work situation? Yes, when one record brings the facts together and each source has a named keeper.

One work recordFictitious deal · Sample Co renewal

  • Record IDhubspot deal 48213HubSpot

    One reliable identifier. One deal ID gets one open work item, so a duplicate run cannot quietly create a second job. Without a stable ID, people and tools can work the same deal twice.

  • Ownerrep_cHubSpot

    The person responsible for movement. HubSpot owns customer status, deal stage and owner. If this field is blank, a stalled deal has nobody to ask.

  • StatusProposal sent, no reply after 3 business daysHubSpot

    Where the deal stands. The trigger here is the silence: no reply after 3 business days. If status is out of date, the next step gets prepared for the wrong situation.

  • Next actionFollow-up draft and next task, for rep_c to checkWork queue

    A Postgres queue item for deal 48213 is assigned to rep_c, and Grok Bot prepares the draft and next task. In the sample, rep_c checks facts, tone, pricing and exceptions, corrects one price line and approves on Nov 4 at 09:12. Run 0917 shows no error, and the final status and note file back to HubSpot.

  • Company contextfollowup-playbook v4, pricing-2026-11Drive

    The rules and definitions around the record. Old pricing or empty notes block the draft, so the routine stops instead of guessing. The queue row stores only a link and status. Customer details stay in HubSpot.

Who keeps it true

  • HubSpot: system of recordNamed deal owner

    Owns customer status, deal stage and owner. Reps see their own accounts, and final status and notes return here after approved work.

  • Drive: approved docsSales manager, reviewed monthly

    Holds two approved context docs, followup-playbook v4 and pricing-2026-11. The sales manager owns and reviews both each month.

  • Postgres: work queueBuilders maintain the service

    Tracks work state, approvals and runs by ID. In the department example, 3 builders maintain and test the queue service, and the company owns it.

Shared data means the right accessWho can reach what
  • Reps see approved context and their own accounts.
  • Managers approve templates and exceptions.
  • With row-level security policies configured, each identity reaches only its assigned work.
  • Agent tools hold only scoped keys.

How the work item moves

  1. New
  2. Preparing
  3. Needs review
  4. Approved
  5. Filed

Any item that fails a check moves to Exception for a manager.

The three data layers, one queue row and the full sample dealStored fields, deal timeline, permissions and sources

Shared data

Three layers

  1. HubSpotSystem of record
  2. DriveApproved docs
  3. PostgresWork queue
HubSpot
Owns customer status, deal stage and owner.
Drive
Holds two approved context docs, followup-playbook v4 and pricing-2026-11. Sales manager owns and reviews both monthly.
Supabase Postgres
Tracks work state, approvals and runs by ID.

What one queue row stores

  • Link · hubspot deal 48213
  • Context · playbook v4
  • State · approved
  • Approved by · rep_c · Nov 4 09:12
  • Run · 0917 · error none

Customer details stay in HubSpot. The row holds a link and status only.

One sample deal, start to finish

Fictitious deal and names for illustration.

  • Where the deal livesHubSpot deal 48213 · Sample Co renewal · stage Proposal sent · owner rep_c · no reply after 3 business days
  • Approved contextfollowup-playbook v4 and pricing-2026-11 in Drive · owner sales manager · reviewed monthly
  • AssignmentPostgres queue item for deal 48213 assigned to rep_c · Grok Bot prepares the draft and next task
  • Approvalrep_c checks facts, tone, pricing and exceptions · approved Nov 4 09:12 · one price line corrected
  • Run traceRun 0917 · error none · final status and note filed back to HubSpot
  1. New
  2. Preparing
  3. Needs review
  4. Approved
  5. Filed

Any item that fails a check moves to Exception for a manager.

Shared data means the right access. People and agents reach only the approved records their job needs, and every important action has an owner and a review trail.

Permissions

  • Reps see approved context and their own accounts.
  • Managers approve templates and exceptions.
  • With row-level security policies configured, each identity reaches only its assigned work.
  • Each item carries the minimum fields the job needs.
  • Agent tools hold only scoped keys.

Back to Capacity