When Claude can read your CRM: build a follow-up list, not a robot salesperson

The CRM is where many small teams lose pace. Not because HubSpot, Pipedrive or any other CRM is bad. Usually the opposite. The problem is that the history sits one step away from the conversation where someone is trying to think.

A customer opened an email but never replied. A quote is close to its deadline. A support issue has been solved three times in three different ways. In practice, someone clicks between contacts, deals, tickets, notes and email activity just to answer: "who should we follow up with, and what should we say?"

That is why the HubSpot connector for Claude is worth watching. It does not make the CRM magic. It does something more useful: it lets AI read CRM context where the work is happening, inside a conversation where a human can ask for analysis, prioritization, draft language and a sensible next action.

Source: HubSpot: You can now connect HubSpot context to Claude

What is actually new with HubSpot and Claude

HubSpot describes its Claude connector as a way to bring CRM context into Claude for personalized insights, visualizations and action back in HubSpot. It works on Claude web, desktop and mobile, and HubSpot says it is available to HubSpot customers across all tiers who also have a paid Anthropic subscription.

For smaller teams, the useful part is not another integration logo. It is that Claude can query CRM objects such as contacts, companies, deals, tickets, users, products, subscriptions, and lists based on the permissions the user already has. HubSpot's product page describes standard access as read-based. HubSpot's knowledge base, last updated on July 3, 2026, also documents admin-granted permissions where Claude can create and update CRM records or log activities, tasks and notes.

That is where the first practical decision sits. Do not start with "AI should sell for us". Start with "AI should help us see which follow-ups deserve a human minute today".

Source: HubSpot Knowledge Base: Set up and use the HubSpot connector for Claude

CRM AI does not make chat the boss

CRM AI is an AI assistant that can use customer relationship data, such as contacts, deals, support tickets and historical activities, to answer questions or suggest next steps. That sounds big. In daily work, it is fairly concrete.

A restaurant can ask which catering requests received a quote but no reply. A consultant can ask for the five deals where the next step is unclear. A school can review which course contacts have waited longest for a response. A small ecommerce team can find tickets where the customer already got an answer but still seems unsatisfied.

That work is often too important to hand over to a fully automatic bot, but too repetitive for humans to rebuild from scratch every week. That is the gap this connector can fill well.

Claude's connector directory lists the HubSpot connector as a sales and marketing connector with read capability. It describes the use case as querying CRM data such as contacts, deals, companies, and tickets directly inside Claude, based on user permissions. That is already enough if the first goal is a better prioritized work list.

Source: Claude Connector Directory: HubSpot Connector

The first workflow: today's follow-up list

Pick one recurring moment, not the whole sales process. Monday morning works. Friday after lunch works too. The point is to turn CRM data into a short list that someone will actually read and act on.

Build the workflow like this:

1. Choose the view: open only the pipelines, lists, or tickets that belong to today's work. Examples: open deals closing in the next 30 days, high-priority support tickets, or contacts who opened a campaign email but did not book a meeting.
2. Ask Claude to read the pattern: let the AI find records where the next human step looks unclear, late, or valuable.
3. Require source references: every suggestion should point to the contact, deal, ticket, activity or note in HubSpot that supports the conclusion.
4. Do not write automatically at first: let the AI draft a follow-up, but do not send email, update stages or create tasks until someone has reviewed the list.
5. Keep a short run log: date, query, suggested records, reviewer, and actions taken.

This is not defensive. It is just a good way to work. A human still decides whether the tone is right, whether the deal deserves a call, and whether the customer's situation has changed outside the CRM.

Copy-paste prompt: turn CRM history into the next step

Use this prompt in Claude when the HubSpot connector is enabled. Replace the bracketed parts.

You are my CRM assistant for [company/team]. Use the HubSpot connector to help me create a reviewed follow-up list, not to send or update anything automatically.

Scope:
- Look at [pipeline/list/ticket view/campaign] for [date or date range].
- Focus on records where a human follow-up could make a difference today.
- Use only information available in HubSpot and say when something is missing.

Create a list of no more than 10 records. For each record, include:
1. Contact/company/deal/ticket.
2. Why it belongs on the list now.
3. Which HubSpot source you used: field, note, email activity, call, ticket or deal stage.
4. Recommended next human step.
5. A short email draft or call opener if useful.
6. Uncertainties or checks I should make before acting.

Rules:
- Do not create or update anything in HubSpot unless I explicitly ask after review.
- If the data is thin, write what is missing instead of guessing.
- Prioritize quality over quantity.

The useful thing about this prompt is that it does not try to turn the CRM into an oracle. It turns the CRM into a workspace that is easier to read.

Three follow-ups that fit smaller teams

Quote going cold. Ask Claude to find open deals where a quote has been sent, an email has been opened or a meeting has happened, but the next activity is missing. The result should be a list explaining why the follow-up makes sense, not an automatic sales email.

Support issue that should become knowledge. Ask Claude to compare tickets where the same problem appears more than once. The next step might be a short reply template, an internal instruction or a knowledge-base article.

Course or school contact that needs clarity. Ask Claude to find parents, participants or organizations where the latest contact shows unanswered questions, booking uncertainty or pending material. The reviewer can then send a human reply in the right tone.

All three are practical. None of them require AI to own the customer relationship. That is the point.

Permissions without turning this into a security manual

HubSpot's knowledge base says a Super Admin or a user with App Marketplace permissions must enable the connector the first time. It also says create and update actions can be recorded in HubSpot's audit log, and that custom Sensitive Data properties are not available through the connector.

For a smaller team, a simple model is enough:

• Start with read access when the goal is analysis and prioritization.
• Add write access only when you have a reviewed workflow and know exactly which objects AI may create or update.
• Use existing HubSpot user permissions instead of one shared super account.
• Require human approval before AI creates tasks, updates deal stages, logs activities or writes customer-facing notes.
• Keep a run log so you can see what AI suggested and what the human actually did.

If you later connect this to more tools, put API keys in environment variables or a secret manager, keep permissions scoped, and redact unnecessary personal details from prompts and drafts. But start where the value is visible: a list of the next best human follow-ups.

What a good result looks like

After 30 minutes, the team should be able to point to three things:

• A better prioritized list than the usual "who shouted loudest?" order.
• Clear HubSpot sources for every suggested follow-up.
• A decision about what AI may do next time: read, recommend, create a task, or update after approval.

That is the right level of seriousness. Not a huge CRM project. Not another strategy workshop. Just a concrete way to let AI use the customer history you already paid to collect.

For Hammer, this naturally fits Tool Forge: connect the right systems, build a bounded routine, and keep humans reviewing where the relationship actually matters. If the team first needs to choose which follow-ups are worth standardizing, Mindset Forge is the starting point. If the routine should become repeatable and measurable, Skill Forge is the next step.

That is how CRM AI becomes useful. Not by pretending to be a salesperson. By making the next human follow-up easier to find.