# How to use AI in your CRM: 10 uses, and what a good version of each looks like

Migaru, Oct 31, 2026 · Guides

> Ten ways to use AI in a CRM, from capturing calls and email to scoring accounts and drafting follow-ups, with what a good version of each looks like.

AI in a CRM is most useful in ten places: capturing conversations, keeping fields current, turning calls into notes and tasks, preparing meetings, drafting follow-ups, scoring accounts, answering questions about the pipeline, researching accounts, finding new ones, and letting other AI tools work with the record. In each, a good version shows its sources, suggests what is a judgement rather than deciding it, and works from a record that keeps itself current.

Below, each use: what it does, and what a good version looks like.

## 1. Capture conversations into the record

The use everything else depends on. AI reads the mailbox, calendar and calls, files each conversation on the contacts and accounts it involves, and creates the ones that don’t exist yet, so the record stops depending on what someone remembered to log.

**What good looks like:** people matched by email address and companies by web domain, months of history read back when you connect, and likely duplicates flagged for a person to merge rather than merged blind.

## 2. Keep fields current

With conversations captured, AI can keep the record’s fields in step with them: summaries, last-contact dates, a deal’s next steps, a company’s industry or headcount.

**What good looks like:** facts written for you, judgements like a stage or a close date suggested with the reason, an admin who decides field by field which is which, and a history on every field of what changed it.

## 3. Turn calls into notes and tasks

A notetaker records the call, and AI turns the transcript into a summary, and the promises made on it into tasks: what you owe, and what to chase.

**What good looks like:** the transcript filed on the meeting and its account, each task linked to the conversation it came from, and a way to bring in transcripts from the recorder you already use.

## 4. Prepare meetings

Before a call, AI writes a brief from the account, the people attending, recent threads, open deals and open tasks.

**What good looks like:** a brief written before you ask, in time to read it, drawn from your own record rather than the open web, and honest when a first call has too little on file to brief.

## 5. Draft follow-ups and outreach

AI drafts the email after a meeting, the follow-up when a promise comes due, and each step of a sequence, from what the record says about that person.

**What good looks like:** written in the voice of whoever sends it, from what was actually promised, and waiting for a person to edit and send, at least until the team trusts the drafts. Anything that sends by itself should stay inside set hours and limits, and stop for a reply, a bounce or an unsubscribe.

## 6. Score accounts and leads

A score ranks who to call next: accounts with momentum, deals at risk, leads that fit.

**What good looks like:** a score you can argue with, listing every point and the fact behind it. A score of 87 that nobody can explain gets ignored. Not every score needs a language model: points for a booked meeting, an open deal or silence past your follow-up time are easier to trust than a number from a black box.

## 7. Answer questions about the pipeline

Ask which deals slipped this quarter, which accounts have gone quiet, or how win rates compare by industry, and get an answer instead of a report to build.

**What good looks like:** numbers counted in the database, not estimated by a model from whatever fitted in its context, with the records behind each number listed, and answers about the pipeline as it stood on a date in the past. Every answer should cite the record, call or thread it came from.

## 8. Research and enrich accounts

AI reads the web and data providers to fill in what a company does, its size, funding and hiring, and a contact’s role.

**What good looks like:** every page it used linked, what it found offered as suggestions rather than written over what you have, and no hunting for anyone’s personal email address or phone number.

## 9. Find new accounts

Describe your ideal customer, or point at your best accounts, and AI proposes companies that match.

**What good looks like:** proposals that skip companies already in the CRM, each saying why it was proposed, and none becoming an account until a person approves it.

## 10. Work with the AI tools you already use

Your team already asks Claude and ChatGPT for help. Over MCP, the Model Context Protocol, those assistants can read the CRM, and change it if you allow, from a conversation there.

**What good looks like:** connecting by signing in rather than pasting a key, an admin who decides whether an assistant may only read or may change records too, and every change it makes shown in the record’s history. [Using your CRM from Claude or ChatGPT](https://www.migaru.ai/blog/mcp-crm) walks through it.

## How to start using AI in your CRM

The order matters more than the tool. AI that drafts emails from a stale record writes confident, wrong emails.

- **Start with capture.** Connect mailboxes, calendars and calls, and let the CRM read back your history. Check that contacts and accounts land where they should.
- **Turn on suggestions, and review them weekly.** Accept or dismiss what the AI proposes for a few weeks. It shows you which fields to let it write and which to keep.
- **Write down what it should know.** Your ideal customer, how you qualify a deal, how your team writes. AI follows written instructions better than it guesses them.
- **Then drafts.** Follow-ups and sequence steps, approved before they go.
- **Then automations and agents.** Rules for predictable work, agents for the rest. [CRM automation](https://www.migaru.ai/blog/crm-automation) covers where each fits.
- **Last, sending.** Only once you have read enough drafts to trust them.

## What to watch out for

- **Answers without sources.** A language model fills a gap with something plausible. If an answer doesn’t say where it came from, check it before you act on it.
- **AI on a stale record.** Every use above works from what the CRM holds. If the record doesn’t keep itself current, AI makes the gaps harder to see, not smaller.
- **Silent changes.** A stage or close date changed without anyone approving it, and no history saying so, is how a forecast stops being believed.
- **Your data.** Ask whether your email and calls are used to train models, and whether email bodies are stored at all.
- **Metered AI.** Credits or per-agent pricing put a cost on every question, which teaches a team to ration the AI. [How to choose an AI CRM](https://www.migaru.ai/blog/how-to-choose-an-ai-crm) has the questions to ask.

## AI in HubSpot, Salesforce and other CRMs

Most CRMs now have AI in some form. From our comparisons, each of which lists its sources:

- [HubSpot](https://www.migaru.ai/compare/migaru-vs-hubspot): its self-updating Smart CRM, from the Fall 2026 release, captures calls, emails and meetings, and Deal Progression, in public beta, proposes updates and follow-ups for one-click approval. Its Meeting Notetaker comes with Sales Hub Professional and Enterprise.
- [Salesforce](https://www.migaru.ai/compare/migaru-vs-salesforce): Einstein Activity Capture syncs email and calendar activity to records, and Agentforce agents draw on Flex Credits, an allowance of which comes with the Core, Advanced and Max editions.
- [Attio](https://www.migaru.ai/compare/migaru-vs-attio): Ask Attio, and a research agent inside workflows, both drawing on credits.
- [Day.ai](https://www.migaru.ai/compare/migaru-vs-day-ai): AI agents you deploy, each running a set number of automated skills, priced per agent.
- [folk](https://www.migaru.ai/compare/migaru-vs-folk): AI assistants for follow-ups, research, recaps and workflows, and fields filled by AI.

[The best CRM for startups](https://www.migaru.ai/compare/best-crm-for-startups) sets eleven of them side by side.

## AI in Migaru

Migaru is built around the first use, so that the other nine work from a current record:

- **Capture:** Gmail or Outlook with its calendar, Slack, and a notetaker for Google Meet, Zoom and Teams, with up to 24 months of mail and calendar read back when you connect.
- **Fields:** summaries and meeting prep written for you, and a deal’s stage, value and close date suggested into one review queue. An admin sets each field to fill itself, suggest, or be left alone.
- **Calls:** transcripts filed on the meeting, summarised and searchable, with promises turned into tasks linked to the conversation. Transcripts from any other recorder come in as subtitles, text, Word or PDF files.
- **Prep:** a brief for each external meeting in the next two days.
- **Drafts:** follow-ups when a task comes due, in the voice of whoever owns it, and on Pro and above, sequence steps, each approved into your own drafts. Only Outreach campaigns you launch, on Growth, send by themselves.
- **Scores:** every account and contact starts at 50 and moves between 1 and 99 with each signal. No model is involved, and every point shows its evidence.
- **Questions:** the agent, in the app, in Slack and on your phone, counts across your records rather than estimating, cites the record, meeting, thread or moment in a call behind each point, and for harder questions runs code over up to 2,000 records of each kind.
- **Research:** on Pro and above, it searches the web, linking every page it used, and researches an account or contact into a sourced note.
- **Prospecting:** on Growth and above, companies that match your ideal customer, or look like accounts you pick, proposed up to ten at a time for review.
- **Other AI tools:** Claude, ChatGPT and other assistants connect to its MCP server by signing in, on every plan.

The agent is on every plan at a fixed price, with no credits. [See pricing](https://www.migaru.ai/pricing), or read [what the AI sales assistant does](https://www.migaru.ai/product/ai-sales-assistant).

## Frequently asked questions

### How is AI used in a CRM?

To capture emails, meetings and calls into the record, keep fields and summaries current, turn calls into notes and tasks, prepare meetings, draft follow-ups and outreach, score accounts, answer questions about the pipeline, research and find accounts, and let assistants like Claude and ChatGPT work with the record.

### How do I start using AI in my CRM?

Start with capture, so the AI works from a current record: connect mailboxes, calendars and calls. Then review the changes it suggests for a few weeks, write down what it should know about your customers and how you sell, turn on drafts you approve, and only then let anything send by itself.

### Can AI update a CRM automatically?

Yes. An AI-native CRM can file conversations, create contacts and accounts, and keep summaries and last-contact dates current by itself. Judgements such as a deal’s stage, value and close date are better suggested for a person to accept, with a history kept of every change.

### Is AI in a CRM accurate?

It is as accurate as the record it reads and the checks around it. Look for answers that cite their source, numbers counted in the database rather than estimated, and changes to judgement fields suggested rather than made silently.

### Will AI in a CRM replace salespeople?

No. It takes over the admin around selling, such as logging calls, updating fields, writing prep and drafting follow-ups, so that people spend more of their time with customers. Decisions about deals, and what a customer reads, are still best left with a person.

---

Canonical page: https://www.migaru.ai/blog/ai-in-crm
Last updated: 2026-10-31