OpenAI

Always-on follow-through

OpenAI dots for Always-On Follow-Through

Keeps chasing open work across connected apps until a named person decides.

A dot is an always-on assistant in ChatGPT with its own cloud computer and the apps you choose to connect. It keeps working between conversations, prepares the next packet, and stops for your rules before money moves or messages leave draft.

Owner requests this job can take

Plain requests an owner would actually say. The assistant prepares. A named human decides.

  • Find the last email with the approved quote and show me the original thread.
  • Draft the follow-up in my voice from the samples I already approved.
  • Prepare the send, then wait. Asking for a draft does not authorize sending.
  • Read tomorrow's calendar and flag the visit that still has an open decision.
  • Where CRM write is not connected, prepare a next-action note I can paste with owner and due date.

What KCG configures

  • Eligible ChatGPT access checks and dots setup where your plan and market support it
  • Connected apps limited to the systems your reviewer names, with permissions reviewed before connect
  • Custom rules for ask before action, act if pre-approved, or hand off on money, outbound messages, and credential changes
  • Scheduled checks that surface open sales, service, scheduling, or ops exceptions for a named owner
  • A written handoff so your team knows where Activity View lives and how to pause the dot

What you take away

  • One named dot tied to an eligible ChatGPT account your business controls
  • App connections and custom rules documented for the reviewer
  • A short runbook covering pause and delete
  • A first live workflow scoped to ongoing follow-through, not unsupervised spend

Onboarding

  • Confirm plan eligibility and whether a personal or business-managed account owns the dot
  • Name the reviewer and the irreversible actions that always need approval
  • Connect only the apps required for the first ongoing job
  • Write custom rules for ask / act / hand off, then walk one dry-run end to end
  • Record where Activity View lives and how to pause the dot

Deliverables

  • Setup notes for the named account and connected apps
  • Custom-rule sheet for money movement, outbound messages, and credential changes
  • One taught follow-through workflow with a dry-run transcript
  • Pause and delete runbook for the owner

Connectors by time saved and input / output

Dots can research, draft, and keep scheduled checks running, but permission to read is not permission to change or send. Asking for a draft does not authorize sending it. KCG does not invent equal send or CRM-write ability across every app you name.

Personal email (connected account)

Native connector
Time / rework saved
Stops rebuilding quote history from memory when a buyer pings again.
Input
Connected personal email the reviewer authorizes for the job.
Output
Located thread, quote attachment, and a follow-up draft. Sending waits for approval. At launch a dot does not get its own standalone email address.

Slack or Microsoft Teams (reach)

Native connector
Time / rework saved
Puts the exception packet where the owner already checks messages, instead of another private notebook.
Input
Open work the dot tracked plus the channel or person who should decide.
Output
A reviewable update or handoff in Slack or Teams. Reach is not the same as permission to change records or send customer email.

Plugin ecosystem apps

Scoped integration
Time / rework saved
Cuts tab-hopping when a supported plugin can read the system the job needs.
Input
Only the plugins and permissions the reviewer names for the first workflow.
Output
Research, drafts, or updates allowed by that plugin. Read permission does not include change or send. OpenAI markets a large plugin ecosystem; KCG only claims apps you actually connect and test.

CRM or specialty ops tools without a live write path

Manual export
Time / rework saved
Keeps the next action from dying in chat even when the CRM cannot be written directly.
Input
What happened, who owns it, and when it is due.
Output
A paste-ready next-action note or import sheet. The owner or staff logs it in the CRM you already use.

Worked examples across industries

Each example stays inside capabilities documented by the vendor. The assistant drafts. A named human decides.

Approved quote buried in email

HVAC / home services
Problem
A homeowner replies "send the quote again" and the office burns twenty minutes hunting the approved version.
Input
Connected personal email plus the customer name and approximate date.
Draft
Original approved quote thread surfaced with a short follow-up draft in the owner's voice samples.
Human decision
Reviewer confirms it is the right version, edits the draft, then approves send on the connected account. The dot does not send alone.
Business benefit
Follow-up starts from the real approved quote, not a reconstructed guess.

Missed invoice after delivery

Wholesale / light manufacturing
Problem
Work shipped, but the invoice never left draft because nobody rebuilt the context.
Input
Connected work and email context that shows completed delivery without a sent invoice.
Draft
Prepared invoice draft with open line items and customer reference.
Human decision
Reviewer checks amounts and customer details, then approves send. Money movement stays human.
Business benefit
Billing follow-through stays behind a named person, matching OpenAI guidance to review consequential work.

Open decision before tomorrow's visit

Field service / contracting
Problem
A site visit is on the calendar, but the change-order decision is still open.
Input
Connected calendar and email context for the visit, plus the open thread.
Draft
Short packet: visit time, open items, and a customer reply sitting in draft.
Human decision
Owner picks the scope, edits the reply, and only then approves send. The brief does not auto-cancel the visit.
Business benefit
Scheduling conflicts and open decisions surface before travel, without unsupervised customer messages.

Ops exception without CRM write

Auto repair
Problem
Parts delay and warranty notes live in three places, so the next owner never gets assigned.
Input
Connected inbox or notes plus the job number the shop already uses.
Draft
Exception summary with proposed next action, owner, and due date.
Human decision
Service writer accepts the packet and logs it in the shop CRM, or pastes the manual-export note when write is unavailable.
Business benefit
Ops follow-through continues even when the CRM connector is read-only or missing.

Good fit

  • You have an eligible ChatGPT plan in a supported market and someone who will own review.
  • The first job is ongoing follow-through across email and connected apps, not a one-shot prompt.
  • You can name irreversible actions that must stay behind ask-before-action rules.

Poor fit

  • You need universal availability today; dots are rolling out and are not on every plan or market.
  • You want unsupervised payments, outbound calls the product cannot initiate, or a standalone agent email.
  • You expect a specialist organization pilot or deep company-system integration from a personal dot.

Why it helps

Most shops lose time rebuilding context for work that already happened. A dot can keep working between conversations, pull context from connected apps, and bring you a reviewable packet instead of another chat thread you have to reconstruct.

Pilot measurement

  • Open follow-ups drafted versus follow-ups that still needed human reconstruction
  • Time from "customer asked again" to reviewer decision
  • Share of drafts accepted without major factual rewrite
  • Number of times the reviewer paused or tightened a custom rule

Limits

  • Dots are rolling out to eligible ChatGPT plans and markets. Access is not universal and may take time to appear on an account.
  • Specialist organization pilots and deep company-system integrations are separate from a personal dot. Do not treat early marketing as a finished rollout.
  • Dots can make mistakes. Custom rules and review help, but consequential work still needs a human.
  • Local computer access starts off. Cloud browser sign-ins are separate from your personal sessions.
  • KCG does not connect production accounts during sales demos. Setup happens only after you approve the scope.

Named reviewer before actions

Before a dot can send, purchase, or change an account, a named reviewer on your roster must approve. Default KCG rule: ask before taking action on money movement, outbound messages, and credential changes.

Official sources: OpenAI Help: Getting started with your dot; OpenAI: Introducing dots

Back to virtual assistants