To choose an AI CRM, look past the list of AI features and ask five things: where the record comes from, how much the AI decides alone, whether its answers show their sources, how the price changes as the AI does more, and whether your data can leave. The twelve questions below cover those five, with what a good answer sounds like.
Ask them in a demo, or better, in a trial with your own mailbox connected. Most can be answered in the first hour. How to use AI in your CRM covers what each use of the AI should look like once you have chosen.
1. Where does the record come from?
This is the question that separates an AI CRM from a CRM with AI features. If people still type the record, every AI feature works from whatever they remembered to log.
A good answer: it connects to your mailbox, calendar and calls, creates contacts and accounts from the people you talk to, and files each conversation on them, rather than only enriching the records you add yourself.
2. How far back does it read when you connect?
A CRM that starts empty takes months to become useful. One that reads your history starts with the relationships you already have.
A good answer: months or years of mail and meetings, read back when you connect, with a choice of how far.
3. How does it match conversations to records?
Automatic capture is only as good as its matching. Get it wrong and you have a CRM full of duplicates, filled in faster than ever.
A good answer: people are matched by email address and companies by web domain, a thread two teammates both received is kept once, and likely duplicates are flagged for a person to merge rather than merged blind.
4. Which fields does it write itself, and which does it suggest?
Facts, like who was on a call, are safe to write automatically. Judgements, like a deal’s stage or close date, are somebody’s call, and a CRM that changes them silently produces a forecast nobody believes.
A good answer: a clear list of what it writes and what it suggests, and an admin who can change that field by field.
5. Who presses send?
Some AI CRMs send sequences on their own. Others draft every message for a person to send. Neither is wrong, but you should know which you are buying, because a careless email goes out under your name.
A good answer: a plain statement of what it sends by itself and what it only drafts, with sending held to set hours and daily limits, and stopped by a reply, a bounce or an unsubscribe.
6. Can I see why the AI did something?
An AI you can’t check is an AI your team stops trusting, and then stops using.
A good answer: every suggested change comes with its reason, every answer cites the call, meeting or thread it came from, and every field keeps a history of what changed it and when.
7. Does it count, or estimate?
Ask a question about your pipeline that you already know the answer to: how many deals closed last quarter, and which ones slipped. A language model summarising whatever fitted in its context will get totals wrong with confidence.
A good answer: it counts in the database, lists the records behind the number, and can say what the pipeline looked like on a date in the past.
8. Does it record calls, or work with the recorder you have?
Calls are where most of what matters in a deal is said. A CRM that can’t read them is missing the most important part of the record.
A good answer: a notetaker that joins your video calls once each person turns it on, and a way to bring in transcripts from the recorder you already use.
9. How does the price change as the AI does more?
AI CRMs charge per seat, per workspace, in credits for AI work, or per AI agent. Credits and per-agent pricing mean the bill follows how much the AI does, which is hard to predict and teaches a team to ration it.
Ask what happens when credits run out, whether unused ones roll over, and what a busy month would cost. Agentic CRM goes through the four ways of charging.
A good answer: you can work out next year’s bill from how many people you plan to hire.
10. How long does setup take, and who does it?
Some CRMs are platforms you configure, often with an admin or an implementation partner, and some charge for onboarding. That is right for a large sales organisation and wrong for a team that needs the CRM this week.
A good answer: minutes to connect, standard objects and stages ready to use, fields you can add as your process takes shape, and no required onboarding fee.
11. What happens to my data?
An AI CRM reads your email and calls, so this question deserves a straight answer rather than a badge.
A good answer: encrypted in transit and at rest, never used to train models, a choice over whether email bodies are read at all, and a plain account of where compliance audits stand, including the ones still in progress.
12. Can I get my data out, and can other AI tools get in?
You may outgrow the CRM you start with, and your team will want to use it from the assistants they already have. Using your CRM from Claude or ChatGPT covers what to check.
A good answer: every view exports as CSV or Excel, every record can be read and written through an API, and an MCP server lets assistants like Claude and ChatGPT connect by signing in.
How Migaru answers them
- The record: Gmail or Outlook with its calendar, Slack, and a notetaker. Contacts and accounts are created from the people you email and meet, and each conversation is filed on them.
- History: up to 24 months of mail and calendar when you connect, or 12, 6 or none.
- Matching: people by email address, companies by web domain with subdomains folded in, and likely duplicates flagged for you to merge.
- Fields: summaries and dates are written for you, and a deal’s stage, value and close date are suggested by default. An admin sets each field to fill itself, suggest, or be left alone.
- Sending: follow-ups and sequence steps are drafts, and an approved email goes to your own drafts for you to send. Only Outreach campaigns you launch on Growth send by themselves, within their hours and limits, stopping for anyone who answers, bounces or unsubscribes.
- Sources: answers cite the record, meeting, thread or moment in a call they came from, and every field keeps its last 25 changes with what made each one.
- Counting: answers are counted from your records, the pipeline can be rebuilt as it stood on a past date, and harder questions run code over up to 2,000 records of each kind. On Pro, it searches the web too, linking every page it used.
- Calls: a notetaker on every plan for Google Meet, Zoom and Teams, and transcripts from any other recorder as subtitles, text, Word or PDF files.
- Price: fixed, per user on Starter and one price for the workspace on Pro and Growth, with no credits. See pricing.
- Setup: a few minutes. Starter and Pro start from a card, with a 14-day free trial and no onboarding fee.
- Your data: encrypted in transit and at rest, and not used to train models. Email bodies are read only if you turn it on, in memory, and never stored. The SOC 2 Type II audit and HIPAA work are in progress. The trust center has the detail.
- Getting out, and in: any view exports as CSV, Excel or JSON, every record can be read and written through the API, and Claude and ChatGPT connect to its MCP server by signing in.
Which AI CRM fits which team
The answers sort CRMs more usefully than their feature lists do. From our comparisons, which each say where the other CRM is the better choice:
- Marketing, sales and service on one platform, starting from free CRM tools: HubSpot.
- Territories, quotes and approvals, with an admin or partner to build them out: Salesforce.
- Your own data model and workflows, with a free plan for a team of up to three: Attio.
- Sequences sent for you by email and on LinkedIn, and a success manager to set you up from Pro: Lightfield.
- A CRM that updates the pipeline on its own, with a free plan and unlimited seats: Clarify.
- A team that sells over email and calls, wants the record kept for it and a person to press send, at a fixed price: Migaru.
The best CRM for startups sets eleven of them side by side, by how each charges and what a small team needs.
Frequently asked questions
What is an AI CRM?
A CRM that uses AI to do work people used to do by hand: capturing conversations, updating records, summarising accounts, scoring deals, drafting emails and answering questions about the pipeline. How much of that it does, and how much it decides without asking, varies widely between products.
What is the best AI CRM for a small business?
One whose record keeps itself from the tools you already use, at a price you can predict. For a small team that sells over email and calls, look for capture from Gmail or Outlook and from calls, drafts you approve before they go, and a fixed price rather than credits. If you need a free plan, HubSpot, Attio, Day.ai and Clarify have one.
Are AI CRMs worth it?
They are worth it when the AI keeps the record, because a CRM nobody updates is worth little however good its features are. They are worth less when the AI is a set of features on top of a record your team still has to type.
How much does an AI CRM cost?
From free plans to tens of dollars a seat or hundreds of dollars a month for a workspace, often with AI use billed in credits on top. Compare how each bill grows with your headcount and with how much you use the AI, not only the starting price.
Do AI CRMs train models on my data?
Ask each vendor directly and read their terms, since practice varies. Migaru doesn’t use customer data to train models.