Automate More of Sales With ChatGPT
Most sales teams spend a large share of every week on work that is not selling: researching prospects, preparing for calls, writing follow-ups, and updating the CRM. The selling happens in the gaps.
Agency Growth builds a ChatGPT-powered sales system that collects the right information, follows your process, creates the required work product, and sends it to the appropriate person for review.
How sales works today
The process now
- A new lead comes in and a rep spends time searching the company website, LinkedIn, and past emails to understand who they are and what they might need.
- Before each call, the rep pulls together notes from the CRM, prior email threads, and any earlier conversations, usually by reading each source one at a time.
- After the call, the rep writes up notes from memory, decides on next steps, and drafts a follow-up email.
- Proposals and quotes get assembled by copying from old documents and adjusting names, scope, and pricing by hand.
- CRM records get updated when the rep has time, which means fields are often incomplete or stale.
- Pipeline reviews depend on each rep summarizing their own deals verbally, because the written record is thin.
Where time is lost
- Prospect research repeated from scratch for every new lead, even when the pattern is the same.
- Pre-call preparation that consists mostly of finding and rereading information the company already has.
- Post-call admin: writing notes, logging activity, and drafting the same categories of follow-up email again and again.
- Proposal assembly, where most of the effort is formatting and adapting boilerplate rather than deciding what to offer.
- Chasing reps for CRM updates and reconstructing deal history before pipeline meetings.
Reps start every call prepared, every follow-up goes out the same day in your voice, and the CRM reflects reality, without adding admin hours. The goal is not to remove judgment from selling. The goal is to eliminate avoidable research, preparation, formatting, summarization, and administrative work around it.
What ChatGPT does, and what stays human
Information it needs
- Your sales process, written down: stages, qualification criteria, and what a good next step looks like at each stage.
- Your positioning and messaging: who you serve, the problems you solve, and the language you use to describe your offer.
- Templates and past examples: your best follow-up emails, proposals, and call summaries, so output matches what good looks like for you.
- Product, pricing, and packaging information the AI is allowed to reference, with clear boundaries on what it must never state without approval.
- Access to context per deal: call notes or transcripts, email threads, and CRM records, provided per conversation or through connected tools where your plan supports them.
What ChatGPT does
- Summarizes prospect research: paste in a company website, a LinkedIn profile excerpt, or an inbound inquiry, and ChatGPT produces a structured brief on who they are and what they likely need.
- Builds pre-call briefs from the materials you give it: prior notes, email threads, and CRM excerpts become a one-page summary with open questions and suggested talking points.
- Turns call notes or transcripts into clean summaries, action items, and CRM-ready field updates in a consistent format you define once.
- Drafts follow-up emails, check-ins, and re-engagement messages in your voice, using custom instructions or a Custom GPT configured with your templates and tone.
- Assembles first-draft proposals and quotes from your approved boilerplate, adjusted to the specific deal, ready for a human to review and finish.
- With connectors available on business plans, ChatGPT may pull context from tools like Google Drive or SharePoint, so reps spend less time hunting for source documents.
What remains human
- Qualification and prioritization: deciding which deals deserve time is judgment, not formatting.
- The conversations themselves: discovery calls, demos, and negotiations stay with your reps.
- Pricing decisions, discounts, and any commercial commitment: the AI drafts, a person approves.
- Relationship moments: a sensitive renewal, a frustrated buyer, or a strategic account gets a human-written message, not an edited draft.
- Exceptions: any deal that does not fit the standard process gets flagged to a person rather than forced through the template.
- Final review of every outbound email and proposal before it is sent.
Systems involved: ChatGPT (a business plan such as Team or Enterprise for shared workspace features and admin controls), Your CRM, such as HubSpot or a similar system of record, Email (Gmail or Outlook), Your calendar and meeting tools, Document storage such as Google Drive or SharePoint, where proposals and templates live, Call recording or transcription tools, if you use them.
From input to finished work product
Example input
A rep pastes in their raw notes from a 30-minute discovery call with a mid-sized distribution company: bullet points about the prospect's current ordering process, two pain points around manual data entry, a mention that the decision involves the operations director, and a rough budget range discussed near the end. The rep asks for a call summary, CRM updates, and a follow-up email.
Example output
ChatGPT returns a structured summary: situation, pain points, stakeholders, budget signal, and stage recommendation, formatted to match the CRM fields the company defined during setup. Below it, a follow-up email in the company's voice that thanks the prospect, restates the two pain points in their own words, confirms the agreed next step, and proposes two meeting times. The rep edits one line, updates the CRM, and sends the email the same afternoon.
How we build it
- 1
Map the sales process
We document how deals actually move through your pipeline today: stages, handoffs, the repetitive tasks at each stage, and where reps lose the most time. This map decides what gets automated first.
- 2
Set up the workspace
We configure your ChatGPT business workspace: seats, roles, and the admin and data controls appropriate for sales data. Business plans state that your data is not used to train models by default.
- 3
Load your sales context
Your positioning, qualification criteria, objection responses, templates, and approved product and pricing language get structured into instructions and reference documents the AI works from, so output sounds like your company, not a generic assistant.
- 4
Build the sales assets
We create the reusable assets your team will actually use: a pre-call brief generator, a call summary and CRM update formatter, follow-up email drafters for each common scenario, and a proposal assembler, typically built as Custom GPTs or projects your whole team shares.
- 5
Connect your tools where supported
Where your plan supports it, we set up connectors to sources like Google Drive or SharePoint so the AI can reference current documents. For your CRM, we define the exact formats for updates so pasting them in takes seconds.
- 6
Train the team
Reps learn the handful of workflows that matter, when to use each asset, and the review rules: what they may send after a quick edit and what needs a manager's eyes. Adoption is the difference between a tool and a result.
- 7
Review and refine
After the first weeks of real use, we review outputs with your team, tighten the instructions where drafts miss the mark, and extend the system to the next set of tasks worth automating.
Guardrails before speed
The AI states wrong pricing, terms, or product claims in a draft that goes out unreviewed.
Approved pricing and product language is loaded as the only source the AI may use for those topics, and every outbound draft requires human review before sending. Commercial commitments are never sent without approval.
Sensitive prospect or customer data is handled carelessly.
We use a business workspace with admin controls, define what deal data may be shared with the AI, and set team rules for handling personal information. Business plans state that your data is not used for model training by default.
Outreach starts sounding generic and prospects notice.
Assets are built from your real emails and voice, reps are trained to personalize the parts that matter, and high-stakes messages stay human-written. The AI drafts the routine, not the relationship.
The CRM fills with AI-generated notes that reps never verified.
CRM updates follow a fixed format the rep confirms before logging, and summaries always cite what the prospect actually said, so a wrong inference is easy to catch.
The team uses it for two weeks and drifts back to old habits.
We scope the first phase to the two or three tasks with the most obvious payoff, train against real deals rather than demos, and schedule a refinement review so the system improves instead of decaying.
How we know it is working
- Hours per rep per week spent on research, prep, and admin, measured before and after implementation.
- Time from call ended to follow-up sent.
- Share of calls with a complete, structured summary logged in the CRM.
- Time from qualified opportunity to proposal delivered.
- CRM field completeness on active deals.
- Rep adoption: how many reps use the sales assets in a normal week.
Frequently Asked Questions
Will ChatGPT send emails to my prospects automatically?
Not in this implementation. ChatGPT drafts the email and a rep reviews, edits, and sends it. Keeping a person on the send button is a control, not a limitation: it protects your relationships and your accuracy.
Does this replace our CRM or our sales engagement tools?
No. Your CRM stays the system of record. ChatGPT removes the work around it: turning conversations into clean updates, briefs, and drafts. It complements the tools you already run rather than replacing them.
Is our deal data used to train the model?
On ChatGPT business plans such as Team and Enterprise, your data is not used to train models by default. We also set workspace rules for what deal information may be shared with the AI.
How is this different from reps just using ChatGPT on their own?
Individual use produces inconsistent output and no shared standard. An implementation gives the whole team the same assets, built on your process, voice, and approved language, with review rules and admin controls. Creating an account gives your team access to the platform. It does not teach it how your company operates.
What do we need ready before starting?
Mostly things you already have: examples of your best emails and proposals, your pricing and product information, and a willingness to write down how your sales process works. We handle the structuring.
The goal is not to remove judgment from sales
The goal is to eliminate avoidable research, preparation, formatting, summarization, and administrative work around it.