I'm Lilach Bullock, and I provide AI integration services for small and mid-sized businesses: I connect AI to the tools you already run, like HubSpot, your inbox and your forms, through Make.com, Zapier and n8n. Done this way it beats copy and paste, because data sent through OpenAI's API hasn't been used to train its models since March 1, 2023.
Most businesses I talk to already use AI. Someone pastes an enquiry into ChatGPT, copies the answer back into an email, and does the same thing forty times a week. That's AI use, but it isn't integration. Integration is when the AI step sits inside the workflow itself, so the enquiry arrives, gets read, gets sorted and gets a drafted reply without anyone copying anything.
That's the work I do. I've been in marketing since 2003, I built, scaled and sold a digital marketing agency, and I've partnered with Twitter, Dropbox, IBM, monday.com and Greenpeace. These days I build the AI workflows that take the repetitive work off a team's plate. Here's what AI integration services should cover, what the data rules look like on OpenAI's own pages, and how a project with me runs.
AI integration means connecting an AI model to the systems a business already runs, such as its CRM, inbox or website forms, so the AI does one named job inside an existing workflow. It suits small and mid-sized businesses with a repetitive, high-volume task and someone willing to own the result.
What AI integration looks like in a small business
Forget the idea of a big new platform. In a small business, AI integration usually means adding one smart step to a process you already run. The tools stay the same. What changes is that the boring middle of the job stops needing a person.
These are the integrations I'm asked for most, because they save hours from the first week:
- Enquiry triage: a website form or inbox message is read by AI, tagged by type and urgency, logged in your CRM and given a drafted reply for a human to check.
- Call notes to follow-ups: notes or a transcript go in, a follow-up email and a CRM update come out.
- Content in your voice: a short brief becomes a first draft for a newsletter, blog post or social post, written to your own style guide.
- Knowledge tools: your company's documents sit behind a chat interface, so the team gets answers from your own material instead of guessing.
- Reporting: numbers from several tools land in one summary every week, written in plain English.
Why the API route matters for your data
This is the part most buyers don't ask about, and it's the part I'd want answered first. When staff paste client details into whatever AI account they signed up for themselves, you've no say over how that data is handled. When the AI step runs through an integration that you set up, on an account your business owns, the rules are written down and you can check them.
OpenAI publishes those rules. Its data controls page says that data sent to the OpenAI API hasn't been used to train or improve its models since March 1, 2023, unless you opt in. It also says abuse monitoring logs are kept for up to 30 days by default, unless the law requires longer. Its enterprise privacy page covers ChatGPT Business, ChatGPT Enterprise and the API Platform, and here's what it commits to:
| What you're asking | What OpenAI's own pages say |
|---|---|
| Is my API data used to train models? | No, not since March 1, 2023, unless you opt in |
| How long are API logs kept? | Abuse monitoring logs up to 30 days by default, longer only if the law requires it |
| Business plans and training | Business data isn't used for training by default |
| Who owns the inputs and outputs? | You do, where the law allows |
| Access and sign-in | SAML SSO, and you choose which internal sources are connected |
| Security | A completed SOC 2 audit, AES-256 encryption at rest and TLS 1.2+ in transit |
What I build and connect
My integration work sits inside a wider set of services, so you don't need a different person for each piece. I build in the tools most small businesses already pay for, and I'll tell you when something you're paying for can go.
- HubSpot implementation and AI integration: a full HubSpot build or rebuild, with the automation, workflows and AI steps wired in.
- Workflow automation: Make.com, Zapier and n8n integrations that connect the tools you use, so the repetitive steps run themselves.
- AI agents and custom tools: assistants your team uses every day, content agents that draft to your voice, and knowledge tools built on your own expertise.
- Training: your team learns how the integration works and how to check its output, so you aren't dependent on me forever.
How an integration project runs
Every project starts small on purpose. One workflow that works beats five that half work, and it gives you something real to judge me on before you commit to more.
- A free call: you tell me where the hours go, and I ask the questions that show where AI pays back first. If it isn't a fit, I'll say so before we hang up.
- Map the workflow: we write down every step, every tool it touches and every piece of data that moves, including what should never leave your systems.
- Choose the route: an API connection, a business AI plan or a feature inside a tool you already pay for, based on cost, data and how much control you need.
- Build with a human check: anything that goes to a client gets a person's approval until you trust it.
- Hand over and train: your team learns what it does, how to check it and what to do when a step fails.
- Review: we check the hours saved, keep what works and drop what doesn't.
How to test an AI integration before it goes live, and after
Testing is where the risk gets caught. The US National Institute of Standards and Technology (NIST) says in its AI Risk Management Framework, released in January 2023, that AI systems should be tested before their deployment and regularly while in operation. NIST describes the framework as intended for voluntary use, so treat it as a yardstick rather than a rule. It's a sensible one for a business of any size.
Before switch-on, work through four checks:
- Run real examples in a copy: take a batch of recent enquiries, call notes or documents and push them through a test version that can't email anyone or change a live record.
- Hunt for the wrong answers: include the awkward cases, like a complaint, a message in another language, a half-empty form or a client with two accounts. Fix each miss with a rule, a better prompt or a human check.
- Name the reviewer: one person signs off the test results and owns the workflow once it's live. If no name goes on it, it isn't ready.
- Break it on purpose: switch off one connection and see whether anyone hears about it.
Testing doesn't end at go-live. Put a monthly check in the diary and read a sample of real outputs each time. Re-test straight away when a connected tool updates, a prompt is edited or the work itself changes shape. NIST also describes governance as a cross-cutting function, infused through the rest of the framework, so the named owner keeps looking.
Integration, automation or an agent?
These words get used as if they mean the same thing. They don't. Automation runs the same steps every time. Integration adds an AI step that reads, sorts or writes inside that flow. An agent decides its own next step, which is powerful and also harder to keep safe. Most small businesses need the first two long before they need the third.
If you think you need something closer to an assistant that makes decisions, my page on building custom AI agents explains when that's worth it. And if your team is technical enough to build workflows yourselves, my guide to doing automation yourself or hiring a consultant will help you decide.
Who I'm not the right fit for
Plenty of the firms selling AI integration services are software houses that build AI into products their clients sell. If you want AI features inside your own app for your own customers, you need a development team, and I'm not it.
I integrate AI into how your business runs: your sales, marketing, admin and customer service. You don't need to be technical to work with me. If you can use email, you can use what I build. For the bigger picture of rolling AI out across a business, read what an AI implementation consultant for small business does.
When to wait before integrating AI
Some workflows aren't ready for an AI step. Hold off if any of these sound familiar:
- The process changes every week: if the steps, forms or people involved keep shifting, settle the process by hand first. An integration fixes a process in place, so it needs a process worth fixing.
- Nobody owns it: a workflow that belongs to everyone breaks quietly. Give it one named owner before anything is built.
- The data is messy: duplicate contacts, empty fields and three spellings of the same client will flow straight into whatever the AI produces. Clean the records it will read first.
Questions to ask an AI integration partner
Take these six into any sales call. A partner who knows the work answers each one in plain words.
What connects to what?
You need to know every system the AI step can read from or write to. A good answer is a simple list or diagram of each tool, each connection and which way the data moves.
Where does the data go?
Client details can pass through several services on the way. Expect every service named, which AI account handles the data, and what's logged or kept.
Who owns the accounts and keys?
If the builder holds the logins, you can be locked out of your own workflow. The accounts and API keys should be in your business's name, with builder access you can remove.
How will we know when it fails?
A silent failure can lose enquiries for days before anyone spots it. Ask for an alert to a named person and a log of what ran and what didn't.
What gets documented?
Without notes, nobody else can fix or change the workflow. You want a written handover covering each step, each prompt and how to pause it.
What happens when a tool changes?
Connected apps and AI models update, sometimes without warning. Look for a plan for re-testing after changes and a named person responsible for it.
My advice before you pay for AI integration services
Pick one workflow with real volume before you talk to anyone, ideally something your team does dozens of times a week. Ask every provider three questions: where does our data go, who owns the accounts and logins (it should be you), and what happens when a step fails at two in the morning. If the answers are vague, keep looking. And check the current terms of any AI provider before you connect it, because these pages get updated.
If you'd like me to look at your workflows, book a free 30-minute call through my work with me page. You'll leave knowing which integration would pay back first.
Frequently asked questions
What are AI integration services?
AI integration services connect AI to the tools a business already uses, such as a CRM, inbox or forms, so AI steps like sorting enquiries or drafting replies run inside existing workflows. Lilach Bullock builds them in HubSpot, Make.com, Zapier and n8n.
Is my data used to train AI if I integrate through the OpenAI API?
According to OpenAI's data controls page, data sent to the OpenAI API hasn't been used to train or improve its models since March 1, 2023, unless you opt in. Abuse monitoring logs are kept for up to 30 days by default.
Do I need to be technical to use AI integrations?
No. Lilach Bullock builds the integration and trains the team on how to use and check it. If you can use email, you can use the outputs.
What should a small business integrate AI into first?
Start with one high-volume workflow, such as enquiry triage or turning call notes into follow-up emails, where the hours saved are easy to measure.
What is the difference between AI integration and AI automation?
Automation runs the same fixed steps every time, such as copying a form entry into a CRM. AI integration adds a step inside that flow where an AI model reads, sorts or writes, like tagging an enquiry by urgency or drafting the reply.
How do I test that an AI integration is safe to switch on?
Run it on a batch of real past examples in a copy that can't contact anyone, and look hardest at the cases it gets wrong. Have one named person sign off the results. NIST's voluntary AI Risk Management Framework says AI systems should be tested before deployment and regularly while in operation.
Who should own the accounts and keys for an AI integration?
Your business should. The AI accounts, automation platform logins and API keys belong in your company's name, and the person who builds the workflow gets access you can switch off. That way the workflow stays yours if you change provider.
How do I start an AI integration project with Lilach Bullock?
Book a free 30-minute call through the work with me page. Lilach will ask where the hours go and tell you which integration would pay back first.