Sales & CRM

How to set up your first crm with AI

Sales & CRM3 AI tools7 steps6 friction points

Want an agent to handle this on your live data? See the Starch version →

Setting up your first CRM means deciding what to track, where to track it, and how to keep it from becoming shelfware by month two. Most operators land here because a spreadsheet stopped being enough — deals are slipping, follow-ups are getting missed, and nobody can answer 'where does this contact stand?' without digging through three tabs and a Gmail thread. It's not a glamorous problem, but it's a real one that compounds fast.

The reason AI feels like the right tool here is that a lot of CRM setup work is definitional and structural — figuring out what fields to include, what pipeline stages make sense for your sales motion, what data you actually need versus what a generic CRM vendor assumes you need. Those are design decisions, and AI is genuinely good at helping you think through them quickly. It's faster to describe your process to Claude and get a schema back than to read HubSpot's documentation.

ChatGPT, Claude, and Gemini can all meaningfully help with the upfront design work: drafting a contact and deal schema, mapping your pipeline stages, writing import templates, even cleaning up a messy CSV export from a previous tool. Where they hit limits is in the execution layer — none of them connect to your inbox, none of them keep a contact record updated, and none of them remember what you built last week.

Skip the manual work → See the Starch version

Sales & CRM3 AI tools7 steps6 friction points

AI walkthrough

How to do it with AI today

A practical walkthrough using ChatGPT, Claude, and other off-the-shelf LLMs — what they're good at, what you'll have to do by hand.

Tools that work for this

ClaudeChatGPTGemini

Step-by-step

1 Open Claude or ChatGPT and describe your sales process in plain language — how leads come in, what stages a deal moves through, and what information you actually reference before a call. Ask the model to draft a CRM schema: contact fields, company fields, deal fields, and pipeline stages.

2 Take the schema output and ask the model to generate a CSV template with the right column headers. This becomes your import file if you're moving contacts from a spreadsheet or exporting from a previous tool.

3 If you're migrating from an existing CRM (or a messy spreadsheet), export your data and paste a sample of 10-15 rows into the chat. Ask the model to map your existing columns to your new schema and flag missing or inconsistent data.

4 Ask the model to write a short 'CRM hygiene checklist' for your team — what fields are required at each stage, what triggers a stage change, when a deal should be marked lost. This becomes your operating playbook.

5 Use the model to draft email templates for each pipeline stage: first outreach, follow-up after a call, check-in after 30 days of silence. Ask for variations so you're not sending the same copy every time.

6 Manually paste the clean CSV into your chosen CRM tool (Notion, Airtable, a spreadsheet, or a dedicated CRM) and verify that the schema maps correctly. The AI helped you design it; the import is still on you.

7 Set a calendar reminder to re-run this process monthly — because nothing you built in the chat persists, and the model has no visibility into what's changed in your pipeline since the last session.

Prompts you can copy

I run a B2B services business. Deals come in through referrals and LinkedIn outreach, average close time is 6 weeks. Draft a CRM schema for contacts, companies, and deals — include custom fields for referral source, decision-maker status, and next follow-up date.

Here are 15 rows from my existing contact spreadsheet: [paste]. Map these columns to a new schema with fields for: full name, company, title, email, phone, lead source, pipeline stage, last contacted date, and deal value.

Write pipeline stage definitions for a 5-stage sales process: New Lead, Qualified, Proposal Sent, Negotiation, Closed. For each stage, describe what has to be true to enter it and what action moves a deal to the next stage.

Draft three follow-up email templates for a software consulting firm: one for 48 hours after an intro call, one for 2 weeks of no response, and one for a deal that's been stalled in Proposal Sent for 30 days.

I have a list of 200 contacts with inconsistent data — some have full names, some have just first names, company names are formatted differently, and about 30 rows are missing email addresses. Write me a step-by-step cleanup plan I can follow in a spreadsheet.

Reality check

Where this gets hard

The walkthrough above works — until your numbers change, the LLM hallucinates, or you have to re-paste everything next month.

No live connection to your inbox — contact records stay stale the moment you close the chat, and you're re-pasting context every time you want a useful answer about a deal.

Nothing persists between sessions — the schema you designed, the pipeline stages you defined, and the cleanup work you did last week exist only in a chat history you have to dig out and re-explain.

The model can design a CRM but can't populate it — importing, deduplicating, and keeping records current is still entirely manual work on your end.

Output structure drifts — ask Claude for a contact schema today and ask again next week and you'll get something slightly different, which creates inconsistency if you're iterating on the same system.

No cross-source enrichment — the model can't pull in LinkedIn data, Gmail thread history, or company details unless you paste them in yourself, so 'complete' contact records require significant manual assembly.

No automation — follow-up reminders, stale deal alerts, and 'who haven't I contacted in 30 days' queries require you to ask the question manually, every time, with fresh context.

Tired of the friction?

Starch runs the whole workflow on live data — no copy-paste, no hallucinated numbers, no re-prompting next month.

See the Starch version →

Starch alternative

The same workflow on Starch

Starch is an agentic operating system — you describe the CRM you want, and an agent builds it as a persistent app connected to your live email, LinkedIn, and contact data. It runs continuously, not just when you remember to open a chat.

Start with the CRM starter app or describe your pipeline from scratch — Starch builds a schema around your actual sales stages, the fields you care about, and the data you track, not a generic template you spend weeks reconfiguring.

Starch syncs your Gmail data on a schedule, so every contact record shows real thread history. Ask 'who haven't I spoken to in 30 days?' and get a live answer pulled from your actual inbox — not a prompt you re-run manually.

LinkedIn enrichment runs automatically through browser automation — no API needed. Starch keeps contact profiles current without you pasting anything in.

The Sales Agent CRM starter app connects to HubSpot, Apollo.io, and Capsule CRM through Starch's integration catalog if you're already using one of those tools — you get Starch's AI layer on top of data you've already built.

Automations trigger without you — set up a follow-up reminder that fires when a deal hasn't moved stages in two weeks, or a weekly digest of stalled contacts, described in plain English and running on a schedule.

Once the CRM is built, you can extend it in the same way you built it — tell Starch to add a new field, connect a new data source, or build a dashboard showing deal velocity by lead source, without touching a config file.

Get closed-beta access →

Toolkit