AI Sales Automation with Zapier, HubSpot and AI Agents

Sales and solutions work is full of small, repeated tasks: entering a new contact from a business card, researching an account before a call, finding a free slot, writing an agenda, logging the deal, drafting the follow-up. None of them is hard. Together they eat hours every week, and the ones that get skipped are the ones that lose deals.

This article breaks down five Zapier zaps Charley built for his own solutions-architect workflow, working alongside account executives. They connect Slack, Google Drive, Google Forms, Google Calendar, Google Sheets, Gmail, HubSpot and OpenAI, and use Zapier's AI agents for research. It follows the video walkthrough. The same design rules apply when we automate construction systems, so they're worth spelling out.

The building blocks

Zap 1: business card to HubSpot lead

At a conference, you want a new contact in the CRM before you forget the conversation.

1. Post the card. A photo of the business card goes into a Slack channel.
2. Store it. The zap saves the image to Google Drive.
3. Read it. Nanonets OCR extracts the text: name, title, email, phone.
4. Find the company. A code step takes the domain from the email address, and the zap searches HubSpot for a company with that domain, creating it only if none exists.
5. Find the contact. Formatter splits the name into first and last; the zap searches HubSpot for the contact and creates it only if missing.
6. Research. An AI agent looks up the person and the company: what the company does and what it is likely to care about.
7. Write it back. The research goes into a HubSpot note, and the contact is added to a lead list for the account executive or a business development rep to follow up.

Zap 2: discovery call prep and scheduling

Account executives fill in one Google Form for every request. For a discovery call they enter the company, the point of contact, the call time and any notes.

1. Filter to discovery requests only.
2. Find or create the company and contact in HubSpot, the same way as zap 1.
3. Research the organization with an agent, using what the account executive wrote, and save the findings as a HubSpot note.
4. Work out the time. A code step turns the requested start into a start and end time in the right time zone.
5. Check the calendar. If the slot is taken, the zap emails the account executive to pick another time. If it's free, it creates the calendar invite with a generated agenda attached.

Zap 3: demo intake, transcript and agenda

When the account executive brings in a solutions architect for a demo, the intake asks for more: the meeting link and what the team knows about the prospect's processes and software.

1. Filter to demo requests.
2. Pull context from HubSpot and research the company with an agent, focused on what the account executive entered.
3. Transcribe the discovery call. In the video the recording is an audio file in Google Drive, transcribed with OpenAI. A Zoom or Gong recording works the same way.
4. Analyze the call. An agent combines the research, the CRM record and the transcript into a discovery analysis.
5. Check the calendar and either ask the account executive to reschedule or continue.
6. Write the agenda. AI by Zapier turns the analysis into a demo agenda targeted at what the prospect said they need, and attaches it to the calendar invite.
7. Open a deal. A scheduled demo is when an opportunity becomes worth tracking, so the zap creates a HubSpot deal.
8. Notify. An email tells the solutions architect a demo has been booked.

Zap 4: demo follow-up and deal update

After the demo, the solutions architect submits the follow-up intake: expected deal size, expected close date, notes, and any links or resources the prospect asked for.

1. Filter to follow-up submissions.
2. Transcribe the demo from its recording.
3. Gather context from HubSpot (company and deal) and a Google Sheet.
4. Draft the follow-up with an agent.
5. Clean the amount. The account executive typed a dollar sign and a comma; HubSpot's amount field expects a plain number, so Formatter and a code step strip both.
6. Update the deal with the amount, the close date and the next step: send the contract.

Small data fixes like step 5 are where most automations break. Expect them, and put a cleanup step in front of every write to the CRM.

Zap 5: info requests with a human approval

Some prospects have questions before they sign. The account executive pastes the question into the info-request form.

1. Create a HubSpot task right away, so the question can't fall through the cracks even if the rest of the zap fails.
2. Research the answer. An agent searches Google, the CRM, and Zapier's documentation, help articles and community.
3. Draft the email with AI by Zapier, then generate a subject line in a separate step.
4. Ask a person. An approval step sends the draft for review. In the video the reviewer rejects it with "I need it to be more technical", and the zap loops back through the agent to produce a better draft.
5. Save as a draft. The approved email lands in Gmail as a draft. A person still presses send.

Five rules that carry over to any automation

These are the same rules we follow when we connect Procore, accounting and scheduling systems:

1. Capture the input once. One form with a type field beats five forms. Filters route each submission.
2. Look up before you create. Search by a stable key (email domain, email address) before creating a record, or the CRM fills with duplicates.
3. Write results back to the system of record. Research that lives in a chat window is lost. Research saved as a HubSpot note is there for the next person.
4. Keep a person in the loop for anything a customer sees. Agents draft, people approve and send. Our guide to governing AI agents covers how to decide where approvals belong.
5. Clean data at the boundary. Strip currency symbols, normalize names and convert time zones before writing to another system, not after it rejects the value.

When Zapier fits, and when to use something else

Zapier is quick to build in, has a very large catalog of app connections, and is a good fit for workflows like these where each step is a standard action on a SaaS tool. It gets harder when:

• You need to self-host or keep data in your own environment. n8n can be self-hosted; see our n8n construction nodes for a construction example.
• The workflow is really a data pipeline. Monthly reporting with history and quality checks belongs in a data platform, not a chain of zaps. See replacing the monthly Excel report.
• An agent needs many tools across systems with per-user permissions. That calls for MCP servers with governance; see what MCP is and why it matters for construction.

Where to go next

• Watch the walkthrough: AI sales automations with Zapier.
• Workflow automation for construction systems: custom n8n nodes for Procore, Autodesk, HCSS and Power BI.
• Put guardrails on agents: governing AI agents in construction.
• Automate your own handoffs: if your team re-types the same information between forms, CRM, calendar and email, tell us what you want to automate.