AI integration: AI that works with your tools and data.
I connect ChatGPT, Claude and AI agents to the systems your business runs on, from Shopify and HubSpot to QuickBooks, NetSuite and your own database, so they can answer real questions and take on real work, limited to the data and actions you allow.
- Starts with
- Free 30-minute call
- Pricing
- Fixed quote per workflow
- Connects
- Claude, ChatGPT, any MCP client
- Based in
- Los Angeles, California
(01)What gets connected
Where AI usually earns its place
| System | What the AI does there |
|---|---|
| Store (Shopify, BigCommerce) | Answers about orders, stock and customers; drafts product copy; flags orders that need a person. |
| CRM (HubSpot, Salesforce) | Looks up a contact’s history before a call, logs notes, drafts follow-ups for someone to send. |
| ERP and accounting (NetSuite, QuickBooks) | Answers about invoices, open orders and inventory; turns emailed orders into draft records. |
| Your own database | Plain-English questions over read-only views of your data, with the query shown next to the answer. |
| Email and documents | Reads purchase orders, invoices and forms into structured records, checked before anything is saved. |
| Slack or Teams | Where your team asks the questions, so the assistant lives in the tool people already use. |
(02)Three ways to integrate AI
| Shape | How it works | Best for |
|---|---|---|
| An assistant your team asks | Claude or ChatGPT connected to your tools through MCP or a custom connector. People ask questions and request actions; the assistant works under their permissions. | Questions spread across several systems |
| A pipeline that runs on its own | Documents or events come in, a model reads them, code checks the result against your system, and a draft waits for approval. | Repetitive data entry from emails and PDFs |
| A feature in your own product | Search, summaries or a chat built into your app or portal through a model API, using your data and your rules. | Customers who need answers from your data |
(03)Limits, built in
It can only see and do what you allow.
- 01Read-only first. The assistant starts by reading; each action it may take (create a draft, update a field) is added on purpose.
- 02Scoped to a role. It connects as a user or service account with only the permissions the job needs, never an admin key.
- 03Drafts, not final records. Anything it creates waits for a person to approve until it has earned trust on real work.
- 04Code checks the model. Prices, quantities, customers and duplicates are verified against your system by code, not by the model’s judgment.
- 05Every call logged. What was asked, what the assistant read and what it changed, so mistakes can be traced.
- 06Keys stay on the server. API keys and tokens never reach a browser, a prompt or a shared document.
(04)A first project
- 01A free 30-minute call: the work you want AI to take on, the systems involved, and whether AI is the right tool at all.
- 02One workflow, chosen together. The one with the most hours spent and the clearest right answer usually goes first.
- 03A fixed quote and date for that workflow, with what it will and won’t do written down.
- 04Built against your real data in a test setup, then run beside your team before it touches anything live.
- 05Handover: the code, the prompts, the logs and a short guide are yours. Monthly support if you want it.
(05)Built this way
On my own product: Safe Streets Map.
Safe Streets Map turns California’s public crash records into maps and thousands of pages. I built an MCP server over its data so Claude can answer questions like “which streets in Pasadena have the most serious crashes?” through four read-only tools: find a city’s exact name, a city’s totals, the streets with the most crashes, and the crashes near a point. The assistant can read the records but never change them, and its answers come from the data rather than the model’s memory.
The same site uses Claude to read a bike club’s Instagram post into a draft group-ride listing, which a person checks and approves before it goes live.
Swap crash records for your orders, customers and inventory, and it’s the same build: your team asks in plain English, and the answer comes from your own systems, within the limits you set.
(06)When not to
Honest scope
If a spreadsheet formula, a Zapier step or a built-in feature of your software already does the job, use it. AI is worth building when the input is messy, the volume is real and a person can check the result quickly.
(07)Related
- ServiceMCP server developmentYour systems as tools inside Claude and ChatGPT, read-only first and per person.
- ServiceAI document processingEmailed orders, invoices and forms read into your system and checked before saving.
- ServiceAPI integration servicesThe system-to-system plumbing AI depends on: stores, CRMs, ERPs and 3PLs kept in sync.
- GuideNetSuite AI: what’s built in, and what’s worth adding
(08)Questions
What is AI integration?
Connecting an AI model, such as ChatGPT or Claude, to the software your business already runs on, so it can read your real data and do useful work with it: answer questions about orders, turn emailed documents into records, or draft replies from a customer’s history. Without integration, an assistant only knows what someone pastes into it.
What is MCP, and do I need it?
The Model Context Protocol is an open standard for connecting AI assistants to tools and data. A system that offers an MCP server can be used by any assistant that supports MCP, including Claude and ChatGPT. Several business platforms now publish their own MCP servers; NetSuite’s is its AI Connector Service. Where one exists and fits, I use it; where it doesn’t, I build a small server or a direct API integration.
ChatGPT or Claude?
It depends on the work and on which assistant your team already pays for. Both connect to tools, both read documents well, and for most business workflows the integration matters more than the model. I build so the model can be swapped later without redoing the rest.
Will my data be used to train AI models?
That is set by each provider’s terms for the plan or API you use, so it is one of the first things we check. I set up the integration on a plan whose terms fit your data, and keep what is sent to the model to what the task needs.
Is this the same as Zapier or Make?
Those tools are good for simple triggers, and if one covers your workflow, use it. Custom integration is for work they handle poorly: messy documents, checks against your own rules, actions that need approval, or an assistant that answers questions across several systems.
What does it cost?
It starts with a free 30-minute call. After that each workflow is quoted at a fixed price before work starts, so you know the cost before committing.
Do you work with businesses in Los Angeles?
Yes. I’m based in Los Angeles, California and work with businesses across California and the U.S., remotely or in person in LA.
(Next step)
Have work AI should be doing? Let’s map it out.
Start with a free 30-minute call: what you want connected, automated or found, and the smallest first step worth paying for.