ChatGPT Dots for Business: Sales Follow-up and Admin Workflows

A broad instruction such as manage our sales pipeline leaves too many decisions unstated. The agent must infer which leads matter, what a useful update looks like, when contact is appropriate and who should act. A narrow instruction gives the team something it can evaluate. Start with research briefs for next week's discovery calls. The input is an approved list of calls. The output is a short brief with source links, known customer context, questions to ask and missing information. The finish line is a brief the salesperson can review before the appointment. Separate research from external action. The pilot can produce a draft follow-up without being responsible for sending it. The salesperson approves wording and commitments. This preserves a review point while you learn whether the research and drafting are reliable. Write down limits in plain language: use the supplied call list; research public company information; do not infer revenue from company size; identify uncertainty; save drafts to the agreed location; ask the owner when a record is ambiguous. Each instruction has a business purpose. Use the workflow audit if you are choosing between several responsibilities. A recurring job with clear evidence is easier to improve than a problem nobody owns.

Consider a consultancy receiving enquiries through a form and email. Before discovery calls, someone reads the enquiry, looks up the business, checks previous notes and assembles questions. After the call, they prepare a follow-up and update the CRM. The repeated work is gathering context and preparing the draft. Give the dot an approved list of prospects with company name, website, contact owner and call date. Where authorised connections are available, it can use relevant records. Otherwise, start with a prepared input file. Do not assume a particular CRM connector or write action exists because the product supports connected apps generally. Ask for the same brief each time: the enquiry in the customer's words, public business context, source links, possible fit with your offer and questions requiring confirmation. Keep facts and interpretations separate. A proposed need is a discovery question until the customer confirms it. After the call, supply approved notes and request a follow-up draft reflecting agreed commitments. The salesperson checks it before sending through the normal process. Record the final message in the CRM and maintain a named next action. The gain comes from a complete, timely handoff. If a brief cannot be matched to the right CRM record, the workflow still needs work. Revenue automation connects those steps rather than stopping at the draft.

A task brief specifies the input, output and decision rules. Adapt this example after verifying account capabilities: Goal: prepare research briefs for the approved discovery-call list. Sources: use supplied enquiry notes, authorised customer records and public company pages. Link the source for each factual claim. Mark information that could not be checked. Output: one brief per prospect containing the company identifier, enquiry summary, verified context, questions for the call, missing details and responsible salesperson. Actions: prepare drafts and place them in the agreed review location. External messages and changes to customer commitments require the salesperson's review. Exceptions: flag ambiguous company names, unavailable access and conflicting dates. Ask the owner instead of silently choosing an identity or date. Success: the salesperson can identify the right prospect, verify evidence and use the brief before the call. Give the task a complete enquiry, an incomplete enquiry and a misleading company name. Explain what an acceptable brief looks like for each. Reusing a brief creates consistency; inspect actual outputs to see whether the agent follows it.

The difficult cases are often the source of the original problem. A lead uses a personal email, a company has changed its name, or the contact is already a customer. Treating every record as a fresh prospect creates more work. Use identifiers the business trusts. Match the input to a CRM record or an approved company website before research. When a match is uncertain, preserve the ambiguity for review. Two similar names should not become one confident brief. Create a visible place for blocked work. Each exception should say what could not be established, which evidence was checked and who needs to resolve it. The owner should be able to supply missing information without reconstructing the agent's conversation. Review early outputs with the person who normally does the job. They recognise details that matter for a sales conversation. A correct company summary can still miss the enquiry's purpose. Decide how a draft becomes an owned action. A summary in an agent conversation is not necessarily a task in the sales process. It needs a destination, owner and due date the team sees.

Assess Dots for ongoing research and context-heavy preparation. Fixed routing, validation and record transfer often fit an existing automation tool. Combine them when each has a clear responsibility. A form submission can trigger fixed rules to assign an owner and record the enquiry. An AI step prepares a reviewable research brief. The owner approves the next action. This avoids making a general agent responsible for every field and decision. If you use n8n, Make or Zapier, compare a specific workflow. Ask which platform can access sources, preserve identifiers, report failures and support team ownership. Our automation tool comparison provides context. Measure the current process before replacing it. If record routing works reliably, retain it while testing research quality. That produces a smaller experiment and a clearer explanation if results change. A cloud computer does not establish the reliability of the complete business workflow. Assess the sequence from input to reviewed result, including missing records and the person receiving the output.

Suppose a team prepares 20 briefs each week at 12 minutes each. Starting effort is 4 hours. If reviewing a draft takes 5 minutes per brief, routine review takes 100 minutes. The possible reduction before exceptions and maintenance is 140 minutes. This is an illustrative estimate, not a Dots benchmark. Record actual preparation, review and correction time. Add time spent resolving identity matches, reconnecting apps and finding missing outputs. Compare accepted briefs with the normal process. Count factual corrections, missing sources, incorrect customer matches and drafts needing major rewrites. Track whether the brief arrived before the call. A late result has little value even when its content is good. Check plan usage for ongoing cost. Tasks managed through Work or Codex can consume those products' allowances. A conversation allowance does not guarantee unlimited deeper work. At the review date, extend, revise or stop the pilot based on better preparation, lower effort or fewer missed follow-ups.

Can Dots automatically send sales follow-ups? Actions depend on apps, permissions and controls. Start by preparing a draft for the salesperson to approve. Verify the specific sending capability before designing unattended work. Is Dots available in Australia? OpenAI's current Pro rollout excludes the EEA, United Kingdom and Switzerland. Australia is outside those exclusions, but account eligibility and gradual rollout apply. Business access includes Business Premium and eligible Enterprise workspaces. Will Dots replace our CRM? A dot is an agent; the CRM holds customer records, ownership and pipeline state. Keep a clear source for those records and assess the agent within an owned process.

Product facts checked on 1 October 2026: Meet dots and Control your dot. The examples are 13labs' analysis, not reported product benchmarks. Start with the workflow audit, or discuss your sales workflow. We can specify the first responsibility, connect the tools and document the team handover.