You integrate AI into a CRM or ERP by adding a model that reads and writes to the system through its API, so the intelligence appears inside the tool your team already uses rather than in a separate app. The Model Context Protocol, or MCP, gives the model a standard way to connect to HubSpot, Salesforce, an ERP or a Shopify store. You then set permissions so it can only see and change what it should. The work is connection and permission, not replacement.

What does it mean to put AI inside your CRM or ERP?

It means the intelligence runs where the work already happens. A sales rep opens an account in the CRM and sees a plain summary of the relationship, the last three interactions and a suggested next step, drafted on the record itself. An operations manager looks at a purchase order in the ERP and sees a flag that this supplier has slipped on the last two deliveries. Nobody opened a new tab. The AI read the live data in the system and wrote its output back into the same screen.

This is the difference between in context AI and the more common pattern of a separate assistant sitting outside your tools. When the intelligence lives inside the CRM, it reflects the current state of the account rather than a snapshot someone exported last week. The gain comes from proximity: the model looks at the same records your team is looking at, at the same moment.

How do you integrate AI into a CRM or ERP?

Three parts do the work: an API, a connection standard and a model. The CRM or ERP exposes an API, which is the door the model uses to read records and write updates. MCP is the standard way through that door. And the model, Claude in the setups we build, is what reads the data and produces the summary, the draft or the flag.

MCP matters here because it removes most of the custom plumbing. Before an open standard existed, connecting a model to five systems meant building five bespoke integrations and maintaining all of them. With the Model Context Protocol, an open standard published by Anthropic, you connect each system once and the model reads and acts through that link. The Claude Agent SDK then lets you run these actions on a schedule or in response to an event, so the account summary refreshes overnight and the delivery flag appears the moment the data changes, rather than when someone thinks to check. This sits one layer up from the work of unifying marketing and sales data with AI: unification reconciles the records, integration pushes intelligence back into the tools where people act on them.

Do you have to replace HubSpot or Salesforce to add AI?

No, and you usually should not. Both HubSpot and Salesforce expose APIs built for exactly this, so a model can read records and write updates into the CRM you already run. The AI sits alongside your existing setup rather than replacing it. Migrating CRM to get AI features is a large, disruptive project to solve a problem that a connection solves in a fraction of the time.

The same holds for an ERP or an ecommerce platform. A Shopify store, a NetSuite or Sage instance, a Salesforce org: each has an interface a model can connect to. The honest question is which records hold the decisions worth informing, and whether those systems expose the access to do it safely. Most modern platforms do. Where a legacy ERP does not, that constraint is worth surfacing early, because it changes the plan.

What does AI inside an ERP or ecommerce store actually do?

In a CRM, the common jobs are account summaries, drafted follow ups, and risk flags on deals that have gone quiet. In an ERP, it is supplier and inventory intelligence: spotting a supplier whose reliability is slipping, or a stock line trending toward a shortfall, and writing that note where the buyer will see it. In an ecommerce store, it is lifecycle and merchandising work: segmenting customers by behaviour, drafting the next campaign, or surfacing which products are quietly underperforming against their page traffic.

The pattern is the same across all three. The model reads live data, produces a short piece of useful thinking, and puts it back in the tool. It removes the twenty minutes of assembling context that used to come before a decision. That is also why integration pairs with reporting that explains why a number moved: the dashboard tells leadership what shifted and why, and the in context assistant lets the person who owns the account act on it without leaving their screen.

Why does in context AI beat another dashboard?

Because switching tools is expensive, and most people already do too much of it. Harvard Business Review reported in 2022 that the average digital worker toggles between applications and websites nearly 1,200 times a day. Every one of those switches is a small tax on attention. A separate AI app, however good, adds another destination to that count. Intelligence delivered inside the CRM or ERP removes a switch instead of adding one.

There is a data quality argument too. Amra and Elma's 2025 research on marketing silos found that 47 percent of marketers name data silos as the single biggest barrier to getting useful insight from their data. An assistant that reads directly from the system of record is working from the live source, not a copy that may already disagree with it. The closer the intelligence sits to the real data, the less it drifts from the truth, and the more a rep or a buyer trusts what it tells them. Trust is what decides whether any of this gets used after the launch buzz fades.

Is AI integration with a CRM or ERP safe under UK and EU rules?

It can be, with the setup handled properly. UK data protection sits under the ICO, and the AI processing can run in a UK region such as AWS Bedrock UK South, which keeps customer and supplier records in the country and shortens the conversation with your security team. The connection through MCP is permissioned, so the model reads and writes only the fields you allow, and every action can be logged.

The EU AI Act adds governance duties on top of data protection. Full high-risk enforcement lands on 2 August 2026, and penalties under the Act reach up to 35 million euro or 7 percent of global turnover. A CRM assistant that summarises accounts is usually lower risk than a system making automated decisions about people, but the baseline duties still apply: an AI inventory that records what you have connected, a named owner, documentation of what the model can do, and the access controls above. Get those in place and integration is a controlled change, not an open door. Our practical 2026 framework for implementing AI in a B2B business sets governance out in the order it should happen.

Where should you start?

Start with one system and one job. Pick the tool your team lives in, most often the CRM, and the single task that eats the most time before a decision: assembling account context, drafting the follow up, flagging the deal that has gone cold. Connect the model to that one system, prove the assistant earns its place on real records, then extend to the ERP or the store once the pattern is trusted.

The sequence matters. Integration comes after the data foundation, because an assistant is only as reliable as the records it reads. If you are planning a wider programme, the AI integration with CRM, ERP and ecommerce service page covers how Teylu builds and runs these connections, and the AI marketing and sales unified reporting dashboards page covers the reporting layer that sits alongside them. Done in that order, AI stops being an app your team has to remember to open and becomes a quiet improvement to the tools they already use all day.

FAQ

How do you integrate AI into a CRM or ERP?

You add a model that reads and writes to the system through its API, so the intelligence appears inside the tool your team already uses. The Model Context Protocol, or MCP, gives the model a standard way to connect to HubSpot, Salesforce, an ERP or a Shopify store, and you set permissions so it can only see and change what it should. The work is connection and permission, not replacing the system.

What is in context AI?

In context AI is intelligence that runs inside the tool where the work already happens, rather than in a separate app the user has to switch to. A rep sees a summarised account, a suggested next step or a drafted reply on the CRM record itself. Because the AI reads live data from the system, its output reflects the current state of the account instead of a stale export.

What is MCP and how does it connect AI to business tools?

MCP, the Model Context Protocol, is an open standard from Anthropic that gives an AI model a common way to connect to external systems such as a CRM, an ERP or an ecommerce platform. Instead of building a bespoke integration for every tool, you connect each system once through MCP and the model can then read records and take permitted actions through that link. It reduces integration work and keeps access controlled.

Can you add AI to HubSpot or Salesforce without replacing them?

Yes. Both HubSpot and Salesforce expose APIs that let a model read records and write updates, so you can connect AI to the CRM you already run rather than migrating to a new one. The AI sits alongside the existing setup, summarising accounts, drafting follow ups and flagging risk on the records your team already works in. Replacement is rarely the point; adding intelligence to what you have usually is.

Is AI integration with a CRM or ERP compliant with UK and EU rules?

It can be, with the right setup. UK data protection sits under the ICO, and the processing can run in a UK region such as AWS Bedrock UK South so customer records stay in the country. The EU AI Act adds governance duties, with full high-risk enforcement from 2 August 2026 and penalties up to 35 million euro or 7 percent of global turnover. A CRM assistant is usually lower risk, but you still need an AI inventory, a named owner, documentation and controlled permissions.