AI integration services that plug AI into the ERP and CRM you run
We connect an AI tool to your business system, so it can read the records a job needs and save drafts for a person to approve. The tool can be Claude, ChatGPT, your vendor's AI or an AI agent we build, a program that does one job for your team. The system can be an ERP, the software you run orders, stock and accounts in, such as NetSuite, SAP, Odoo or a custom ERP. Or it can be a CRM, where you keep customers and deals, such as Salesforce or HubSpot.
Illustration with sample data.
What AI integration usually delivers
A tested connection between an AI tool and your ERP or CRM. It usually covers the route in, a login made for the AI, and a step where a person approves what it writes.
It suits a team that already knows what it wants AI to do. Two examples: let Claude look up open orders in NetSuite, or let an agent save draft sales orders in Odoo. The usual order is a copy of your system first, then a first release that only reads. The scope of each one is agreed per project.
Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs. We have built AI agents in HubSpot, and in Salesforce we have built AI agents and done data work. Our data and cloud work also covers AWS, Azure, Google Cloud and private servers (a VPS, a virtual private server you rent).
Which AI service is for you
Our AI services differ by what you walk away with. Pick the card that matches what you want at the end.
You get a connection. An AI tool you choose gets its own limited login to your ERP or CRM, and a person approves what it writes.
You get a written plan and no software. Pick it when you are not sure which job to give to AI, or whether your vendor's AI already does it.
You get one back-office job done by software, from the email or PDF that starts it to a draft record in your ERP.
You get a custom AI agent for one job, built when a workflow tool or the vendor's own AI cannot do it.
Also see AI development services and agentic AI development, or every service we offer on the services page.
The route into each system, and its limits
Every vendor documents its own way in for outside software. These are our readings of those documents, with the date we checked. A linked system name opens our page on that system.
Four terms come up again and again in this table. An API, short for application programming interface, is the door a system opens for other software. A REST API is one that other programs call over the web. MCP, short for Model Context Protocol, is a standard way for an AI tool to call a system's functions. OAuth 2.0 is a standard way to give software a sign-in of its own.
Scroll the table sideways to see every column.
| System | Route for the AI | What we plan around | Checked |
|---|---|---|---|
| NetSuite | REST web services with OAuth 2.0, or Oracle's AI Connector Service for MCP tools | All integrations share one limit on requests at once. For contracts from June 2020 the base is 5 on Standard, 15 on Premium and 20 on Enterprise and Ultimate, plus 10 for each SuiteCloud Plus license. From release 2027.1, no new integration may use token-based sign-in. | 15 Nov 2026 |
| SAP Business One | The Service Layer, SAP's web API, plus an MCP server added in feature package 2608 | With approval procedures switched on, an order sent through the Service Layer is saved as a draft, and the call answers with error -2028. An integration has to expect that answer. | 15 Nov 2026 |
| SAP S/4HANA Cloud | SAP's APIs, called through a communication arrangement with a communication user of its own | The user is set up in SAP's Maintain Communication Users app and tied to the outside system. | 16 Nov 2026 |
| Odoo 20 | The JSON-2 API, or Odoo's own MCP server | The JSON-2 API needs Odoo's Custom plan, and its API keys last three months at most. The MCP server shows five lookup tools by default, such as Search and Read group, and hides the rest until someone switches them on. | 15 Nov 2026 |
| ERPNext | The REST API of Frappe, the framework under ERPNext | Frappe gives every record type a REST API with no extra code, so the route exists before we write any. | 15 Nov 2026 |
| Custom ERPs | A database login with a role of its own, or an MCP server built for the system | The MCP standard makes authorization optional, so on a server built for one ERP it has to be added on purpose. | 15 Nov 2026 |
| Sage | Sage 200: a REST API with OAuth 2.0. Sage Intacct: REST APIs or the Intacct MCP server. | Sage lists the Intacct MCP server as read-only. On Sage 200 Professional, the customer or its support provider turns the API on. | 16 Nov 2026 |
| QuickBooks Online | Intuit's REST API with OAuth 2.0 | 500 requests a minute for each company, and 10 a second for each company and app. | 16 Nov 2026 |
| Salesforce CRM | An integration user on the Minimum Access, API Only profile, signed in through an external client app with OAuth 2.0 | An Enterprise org gets 100,000 calls a day plus 1,000 for each Salesforce user license. Salesforce suggests one integration user for every integration. | 16 Nov 2026 |
| HubSpot CRM | The remote HubSpot MCP server over OAuth, or the API with a Service Key for links between systems | For apps built for one account, 100 to 190 requests every 10 seconds by tier, and a daily limit shared by every app in the account. New legacy private apps stopped on 26 October 2026 for older accounts. | 16 Nov 2026 |
How AI is used in an ERP or CRM
Three jobs AI can do with your records, what each one needs from integration, and how you check what it did.
Ask in plain words, such as which orders ship late this week. NetSuite's AI Connector Service already lets Claude or ChatGPT read and update records under your role, in a chat. This one may need no build.
An emailed PDF order becomes a draft sales order. This needs a route that can write, a login limited to that record type, and a rule that saves drafts only.
Oracle says its connector's tools can't call external APIs. A job that reads your CRM and writes to NetSuite by itself, with no one in the chat, needs an integration of its own.
NetSuite keeps the AI connector's execution log for 21 days in production, and its system notes record each field change. We plan around what your system keeps.
Your records stay in the ERP, and the AI works through it. Oracle's page on NetSuite's own agents says they “don't replace NetSuite records, reports, approval workflows, or your review of results.”
AI CRM integration: Salesforce and HubSpot
An AI CRM is the CRM you already run, with AI built in or plugged in. Our team has built AI agents in both of these. Each vendor sells its own AI and also opens a door for outside agents.
Salesforce
Salesforce sells Agentforce, priced by use with Flex Credits or Conversations, or by user license. An outside agent signs in as an integration user. Salesforce suggests one such user for each integration, so each sees only its own share of data. That user cannot open Salesforce's screens, only its API. Read 16 November 2026.
HubSpot
HubSpot sells Breeze, and its agents use HubSpot Credits each month. An outside agent can use the remote HubSpot MCP server to read, write and update contacts, companies, deals and tickets, as a user would. HubSpot's own connectors for Claude, ChatGPT, Copilot and Gemini run on that server. Read 16 November 2026.
Linking the CRM to the ERP can be the job itself. Two examples: a won deal in HubSpot becomes a draft sales order in NetSuite, or a quote in Salesforce picks up stock levels from the ERP. More on each CRM: AI agents for Salesforce and AI agents for HubSpot.
AI implementation services, from first read to daily use
Four steps, in the usual order. The AI reads before it writes, and it writes drafts before anything posts. What each step covers is agreed per project.
Pick the job and the route
One job in one system. We read the vendor's documents for your version and plan, then count the requests the job will make at its busiest hour.
Build on a test copy
A sandbox, which is a copy of your system, with a login made for the AI. Anthropic advises “extensive testing in sandboxed environments” for agents.
Read only, then drafts
The first release usually only reads. Writes come next, saved as drafts that a person approves in the ERP's own screens.
Watch it run
The first weeks of live use are checked against the system's own record of what the AI did. Fixes follow from what that record shows.
Read first, write with approval
We asked four AI assistants how to connect an agent to an ERP safely. Their order matched the one we usually build in.
4 of 4 assistants said start read-only. On 14 November 2026 we asked ChatGPT, Claude, Gemini and Perplexity whether it is safe to connect an AI agent to an ERP (ChatGPT and Gemini in their apps, Claude and Perplexity through their APIs). All four told the buyer to begin with read-only access and to have a person approve high-impact changes. Gemini also allowed automatic approval of small purchase orders under a set amount.
Our saved answers, 14 Nov 2026OWASP names an AI app that “fails to independently verify and approve high-impact actions” as one cause of excessive agency, its risk LLM06:2025.
OWASP Gen AI Security Project, read 15 Nov 2026NetSuite's system notes record who changed a field, when, and through which interface, with the old and new values. Oracle says they “can't be edited by any user, script or app.”
NetSuite help, System Notes Overview, read 15 Nov 2026How much AI integration costs
Published ranges, with the date each was read. They describe the market, not our prices.
A Zapier workflow joining two cloud apps costs $5,000 to $8,000. One that reaches an older ERP through a custom API costs $15,000 to $25,000.
Layer3 Labs, AI consulting fees guide, updated 27 Oct 2026, read 15 Nov 2026The same guide puts the build at 4 to 8 hours for two cloud tools joined in Zapier, and 40 to 80 hours for an older ERP or custom database reached through a REST API.
Layer3 Labs, AI consulting fees guide, updated 27 Oct 2026, read 15 Nov 2026Oracle says NetSuite's AI Connector Service “is not a paid feature.” The AI tool on the other end may need a paid plan of its own.
NetSuite help, AI Connector Service FAQ, read 15 Nov 2026Layer3 Labs sells AI consulting itself, so read its ranges as one firm's view of the market. Ask for the running cost as monthly lines: AI usage, hosting, and vendor fees such as Odoo's Custom plan. Our comparison of six published price lists is in what an AI agent costs to build.
Tell us the system and what the AI should touch
Name the system you run and the records you want AI to read or write. The first call is free.
Questions about AI integration
What is an AI integration service?
It is work that connects an AI tool to software you already run, such as an ERP or CRM. The AI can then read the records a job needs and save its work as drafts for a person to approve. Plain ERP integration services join an ERP to other software, such as a CRM. AI integration adds an AI tool to that link. The work covers the route into the system, the login the AI uses, the approval step and the number of requests it may make.
What do we get at the end of an AI integration project?
Usually a tested connection. The AI reaches your system by the route the vendor documents, under a login of its own that sees only what its job needs. What it writes can be saved as a draft for a named person to approve. The running cost is usually set out as monthly lines: AI usage, hosting and any vendor fee. The scope of each connection is agreed per project.
What is the difference between AI integration and AI implementation?
Integration is the connection: the API, the login and the data the AI may touch. Implementation is the whole path to daily use. It adds choosing the job, testing on a copy of your system, training the people who approve and watching the first weeks of live use. We sell both.
How much does AI integration cost?
Layer3 Labs, a US AI agency, puts a Zapier workflow joining two cloud apps at $5,000 to $8,000. A workflow that reaches an older ERP through a custom API costs $15,000 to $25,000 (guide updated 27 October 2026, read 15 November 2026). Layer3 sells this work itself, so read it as one firm's view. Running costs come on top. Our AI agent cost article compares six published price lists.
Is it worth paying for AI integration?
It is when the hours it saves are worth more than the build price plus the monthly running cost. At 200 orders a week and 6 minutes each, that is 20 hours of typing a week. Check the free routes too: Oracle says NetSuite's AI Connector Service “is not a paid feature.” Anthropic's own advice to builders is to find the simplest solution, which “might mean not building agentic systems at all.”
How can AI be used in an ERP?
AI can look up records when you ask in plain words, draft records from emails and PDFs, or move data between the ERP and another system. The first may need no build. NetSuite's AI Connector Service lets Claude or ChatGPT read and update records while a person works with it in a chat. Work that should run by itself, such as drafting an order from each email as it lands, needs an integration of its own. Oracle says the connector's tools can't call external APIs.
What is an AI CRM, and do we need a new CRM to get one?
No new CRM is needed. An AI CRM is a CRM with AI built in or connected to it. Salesforce sells Agentforce and HubSpot sells Breeze inside the CRM you already run. An outside agent can connect too. HubSpot's remote MCP server lets an agent read, write and update records as a user would. Salesforce offers an integration user that can only work through the API. Pick the route that fits the job: the vendor's AI, or an outside agent on a login of its own.
Will AI replace our ERP?
Not in the work we do. The AI reads from the ERP and writes drafts back into it, so the ERP stays the place your orders, stock and accounts live. Oracle says NetSuite's own agents “don't replace NetSuite records, reports, approval workflows, or your review of results.”
Can we connect ChatGPT or Claude to NetSuite ourselves?
Yes. Oracle's AI Connector Service supports Claude Pro and ChatGPT, and Oracle says it is not a paid feature. It will not run if you sign in with the Administrator role, so you set up a role for it. Bring in an integration team when the job needs drafts made from emailed documents, or data from another system.
How many requests can an AI make in NetSuite at once?
It depends on your service tier, and every integration shares the same limit. For contracts from June 2020, Oracle sets the base at 5 requests at once on Standard, 15 on Premium and 20 on Enterprise and Ultimate, plus 10 for each SuiteCloud Plus license. Oracle warns that an AI tool “may generate multiple requests per action,” so count the busiest hour before you build.
Is it safe to connect AI to our ERP?
It can be, when the limits sit in the ERP itself. The AI can sign in with a role that reaches only the records its job needs, and not as an administrator. Oracle's NetSuite AI connector will not run under the Administrator role at all. Writes can be saved as drafts for a person to approve, and the system's own log records each change.
Can AI connect to an ERP that has no API?
Yes, through the database or a server built for it. Through the database, the AI signs in with a role of its own that can read only the tables the job needs. PostgreSQL's built-in pg_read_all_data and SQL Server's db_datareader read every table, so the usual grant is narrower than either. The other route is an MCP server written for your system. The MCP standard makes authorization optional, so it has to be added on purpose.
Which systems do you integrate AI with?
ERPs and CRMs. Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs, and has built AI agents in Salesforce and HubSpot. For each one we read the vendor's documents for the version and plan you run, because the routes differ. Odoo's outside API needs its Custom plan, and NetSuite expects new integrations to use REST web services with OAuth 2.0.
Do you work with Salesforce and HubSpot?
Yes. Our team has built AI agents in both. In Salesforce, the usual route is an integration user with API-only access, signed in through an external client app with OAuth 2.0. In HubSpot, new links between systems use Service Keys. HubSpot stopped new legacy private apps on 26 October 2026 for accounts created before 28 September 2026. We read both vendors' pages on 16 November 2026.
What should we ask an AI integration company?
Ask five things, and ask for the answers in writing. Which route into our system will you use? What can the AI's login see? How many requests will it make at the busiest hour? Where does a person approve its work? What does it cost each month to run? A firm that cannot answer these has not read your system yet.
Sources
Every source behind a figure or a vendor fact on this page, with the day we read it.
- Layer3 Labs, AI Consulting Fees 2026 guide, updated 27 October 2026Read 15 Nov 2026
- Oracle NetSuite help, SOAP Removal Plans FAQRead 15 Nov 2026
- Oracle NetSuite help, Preparing for Token-based Authentication (TBA) End of SupportRead 15 Nov 2026
- Oracle NetSuite help, Concurrency Governance Limits Based on Service Tiers and SuiteCloud Plus LicensesRead 15 Nov 2026
- Oracle NetSuite help, NetSuite AI Connector Service FAQRead 15 Nov 2026
- Oracle NetSuite help, Required Features and PermissionsRead 15 Nov 2026
- Oracle NetSuite help, Connect to the NetSuite AI Connector ServiceRead 15 Nov 2026
- Oracle NetSuite help, Agents OverviewRead 15 Nov 2026
- Oracle NetSuite help, System Notes OverviewRead 15 Nov 2026
- SAP Business One, Service Layer API ReferenceRead 15 Nov 2026
- SAP Business One 10.0 FP 2608, what's newRead 15 Nov 2026
- SAP Help Portal, S/4HANA Cloud Public Edition 2608, Maintain Communication Users, updated 5 November 2026Read 16 Nov 2026
- Odoo 20 documentation, External JSON-2 APIRead 15 Nov 2026
- Odoo 20 documentation, AI MCP serverRead 15 Nov 2026
- Frappe Framework documentation, REST APIRead 15 Nov 2026
- Salesforce Help, Give Integration Users API Only AccessRead 16 Nov 2026
- Salesforce Developers, API Request Limits and AllocationsRead 16 Nov 2026
- Salesforce, AgentforceRead 16 Nov 2026
- HubSpot Developers, remote HubSpot MCP serverRead 16 Nov 2026
- HubSpot Developers, API usage guidelines and limitsRead 16 Nov 2026
- HubSpot Developers changelog, legacy private app creation sunset, 27 August 2026Read 16 Nov 2026
- HubSpot Knowledge Base, Understand Breeze AssistantRead 16 Nov 2026
- Sage Developer, Sage Intacct AI GatewayRead 16 Nov 2026
- Sage Developer, Sage 200 API introductionRead 16 Nov 2026
- Intuit Developer, QuickBooks Online API call limits and throttlesRead 16 Nov 2026
- Intuit Developer, OAuth 2.0 for QuickBooks OnlineRead 16 Nov 2026
- Model Context Protocol specification, AuthorizationRead 15 Nov 2026
- PostgreSQL documentation, Predefined RolesRead 15 Nov 2026
- Microsoft Learn, SQL Server database-level rolesRead 15 Nov 2026
- OWASP Gen AI Security Project, LLM06:2025 Excessive AgencyRead 15 Nov 2026
- Anthropic, Building effective AI agents, 19 December 2024Read 15 Nov 2026
- Our own answers from the ChatGPT and Gemini apps and the Claude and Perplexity APIs to “Is it safe to connect an AI agent to our ERP, and how do we keep a person in control?”Read 14 Nov 2026