AI automation
How to start using AI for customer engagement without annoying your customers
By Rafia Aamer · · 8 min read
Start with one channel and one repetitive job, such as answering order-status questions on your website chat or replying to missed-call texts for a clinic. Write the answers the AI is allowed to give, set clear rules for when it hands the conversation to a person, and measure response time and handoffs for a few weeks before you add a second job. Launching on every channel and topic at once makes the problems harder to find.
What AI customer engagement covers
The phrase covers several tools that do different jobs. Pick the one that matches your problem.
| Tool | What it does | Good first use |
|---|---|---|
| AI chat assistant | Answers questions on your site or in messaging apps from a knowledge base you control | Order status, shipping, returns, opening hours |
| Conversation AI in a CRM | Replies to texts, social messages and form leads, and can book appointments | Missed-call text-back and appointment booking for service businesses |
| Voice AI | Answers or places phone calls with a synthetic voice | After-hours calls for a clinic or a dealership |
| AI in email and SMS | Writes drafts, picks send times and suggests segments | Subject line and copy drafts a person approves |
| Predictive scoring | Estimates who is likely to buy, book or leave | Prioritizing which leads a person calls first |
A rule-based chatbot follows a fixed menu of buttons. An AI assistant built on a large language model reads free text and writes its own replies. The second is more flexible and more likely to say something you did not approve, which is why the setup steps below matter.
Step 1: Find the questions you answer every day
Before choosing a tool, export a month of support emails, chat logs, call notes or DMs. Sort them into topics and count them. For an online store, the top of the list is usually order status, shipping times, returns and sizing. For a dental practice, it is booking, rescheduling, insurance questions and directions.
Pick the topic with the highest volume and the lowest risk. "Where is my order?" is a good first job: the answer comes from your store data, and a wrong answer is easy to correct. "Is this product safe during pregnancy?" is a bad first job, and should always go to a person.
Step 2: Set one goal and measure the starting point
Write down one goal in numbers you can measure from your current tools:
- First response time: how long a customer waits for any reply.
- Resolution without a person: the share of conversations the AI closes on its own.
- Handoff rate: how often the AI passes a conversation to your team.
- Customer satisfaction: a one-question rating after the chat.
- Booked appointments or orders that start in a conversation.
Record where you stand today. Without a baseline, you cannot tell whether the AI helped or simply moved work around.
Step 3: Choose the tool by where your customers already talk
Customers will not move to a new channel for your chatbot. Put the AI where they already contact you.
- Shopify stores. Shopify Inbox handles chat on the store, and help desk apps such as Gorgias and Zendesk add AI replies that can look up orders. Check which features your plan includes.
- Service businesses on GoHighLevel. GoHighLevel includes AI features for replying to conversations and booking appointments inside the same CRM that holds your pipeline and calendar. Our GoHighLevel development service sets these up.
- Email and SMS marketing. Klaviyo has AI features for writing and timing campaigns. Keep a person approving every send at first.
- Something none of these covers. A custom assistant connected to your own systems, such as an inventory database or a booking API, is a software project. Our custom AI agents and chatbots service builds them.
Buy before you build. A tool that already connects to your store or CRM goes live sooner than a custom build and costs less to maintain. Vendors change features and prices often, so confirm both on their sites.
Step 4: Write what the AI is allowed to say
The quality of an AI assistant depends on the material you give it. Write a short knowledge base in plain language: shipping times by region, return rules, warranty terms, opening hours, booking rules, prices you are willing to quote. Remove old pages that contradict current policy, because the AI will repeat them.
Then write the limits:
- Topics it must never answer, such as medical, legal or financial advice.
- Promises it must never make: refunds outside policy, delivery dates you cannot guarantee, discounts.
- The tone you want, with two or three example replies.
- A clear statement at the start of the chat that the customer is talking to an AI assistant.
Step 5: Design the handoff before launch
The fastest way to lose a customer is an AI that will not let them reach a person. Set rules for handing over:
- The customer asks for a person, in any wording.
- The customer is upset, or the conversation loops twice without progress.
- The topic is on the never-answer list.
- The question involves a payment problem, a complaint or a safety issue.
When the AI hands over, the person who picks it up should see the whole conversation, so the customer does not repeat themselves. Outside business hours, the AI should say when someone will reply and collect what your team needs.
Step 6: Run a pilot, read the transcripts, then expand
Launch on one channel, for one topic, and read transcripts every day for the first two weeks. You will find wrong answers, missing knowledge-base entries and handoff rules that fire too late. Fix them as you go. Compare your measures with the baseline after a few weeks.
Add the second job only when the first runs well. A common order for a store: order status, then returns, then product questions. For a clinic: missed-call text-back, then booking, then reminders and follow-ups.
Privacy and consent rules in the US and Canada
Conversations contain personal data, and some contain health or payment details.
- Health businesses in the US need a HIPAA review and a signed business associate agreement with each vendor that handles patient information. Not every chat or AI vendor will sign one, so check before you connect them.
- Text messages in the US fall under the TCPA, and Canadian commercial messages fall under CASL. Get consent before marketing texts and honor opt-outs.
- Canadian customer data is covered by PIPEDA and provincial laws such as Quebec's Law 25. Several US states also have consumer privacy laws.
- AI disclosure. Tell customers when they are talking to an AI. Some jurisdictions require it, and it sets fair expectations either way.
This is not legal advice. Ask a privacy lawyer to review your setup, especially in health care.
When AI is the wrong answer
Skip AI engagement tools for now if you get a handful of messages a day, since a person answers them better and faster. Skip it if your policies are unclear or change weekly, because the AI will give out-of-date answers. And keep people on conversations where trust decides the sale, such as a high-value custom order or a nervous new patient.
What devinsta does here
devinsta sets up AI customer engagement for online stores and service businesses in the US, Canada and the UK. We audit your current conversations, choose tools that connect to your store or CRM, write the knowledge base and handoff rules, and build the automations around them in GoHighLevel, Klaviyo or Zapier. When no tool fits, we build a custom assistant connected to your systems. You own the accounts and the code.
Book a free 30-minute strategy call to talk through where AI would help your customers, or start a project brief.
Common questions
What is the best AI chatbot for a small business?
The best one connects to where your customers already talk and to the data it needs to answer. For a Shopify store, that is usually a help desk app with order lookup. For a service business on GoHighLevel, its built-in AI features may be enough. Compare two or three options on your real questions before you pay.
Will an AI chatbot replace my support team?
It can take repetitive questions off their plate, such as order status and booking. Complaints, complex orders and anything sensitive should still go to a person, and someone needs to review transcripts and update the knowledge base.
Do I have to tell customers they are talking to AI?
Tell them. Some jurisdictions require disclosure, and customers trust a clearly labeled assistant more than one that pretends to be a person. A short line at the start of the chat is enough.
Can I use an AI chatbot in a dental or medical practice?
Yes, for low-risk jobs such as booking, directions and reminders, if the vendor will sign a business associate agreement where HIPAA applies. Keep clinical questions with your staff, and have a privacy lawyer review the setup.