🎧 Prefer to listen?
If you run a business alone, you know the feeling. You wake up to 14 customer messages. Some are simple — “what’s your return policy?” Some are urgent — “I paid but haven’t received anything.” Some are just… weird. And every single one needs a response, because ignoring customers is how businesses die.
I used to spend two hours every morning just replying to messages. Across email, Instagram DMs, website chat, and the occasional unhinged contact form submission. It was the part of running a business I hated most — not because I don’t care about customers, but because answering the same five questions 30 times a week isn’t why I started this.
So I set up AI to handle it. Not perfectly. Not magically. But well enough that I got my mornings back. Here’s exactly what I did, what worked, and what I’d do differently.
The problem nobody warns you about
When you’re a team of one, customer support doesn’t scale linearly with revenue. It scales faster. Ten customers means maybe one message a day. A hundred customers means fifteen messages, and half of them come in at 11pm when you’re trying to have a life. By the time you hit five hundred customers, you’re spending more time answering questions than actually building your product.
The traditional answer is “hire a VA” or “use a helpdesk.” But VAs cost $500–2000/month and need training. Helpdesks like Zendesk start at $55/agent/month and take weeks to configure. If you’re not at $10k MRR yet, that math doesn’t work.
AI changes the equation. Not by replacing human support — by handling the 70% of messages that are repetitive, predictable, and honestly kind of boring.
What I actually set up
I tested three approaches over a month, going from simplest to most involved.
Approach 1: ChatGPT as a drafting assistant (Day 1–7)
The lowest-friction option. I set up a custom GPT trained on my FAQ, product descriptions, and past customer emails. When a message came in, I’d paste it into the GPT, get a draft reply, tweak it, and send it.
What worked: Response quality was genuinely good. The GPT matched my tone, cited the right policies, and caught details I sometimes forgot. My reply time dropped from 15 minutes per message to about 3.
What didn’t work: I was still the bottleneck. Every message still required me to open ChatGPT, paste, review, copy, and send. It saved time on writing, but not on the actual workflow.
If you’re just getting started, this is where I’d begin. It costs $20/month (ChatGPT Plus) and requires zero setup beyond writing your FAQ. If you want to go deeper on building AI workflows, check out my automation pipeline for how I chain tools together.
Approach 2: Crisp with AI auto-responses (Day 8–21)
Crisp is a customer messaging platform with a built-in AI chatbot. Free tier includes two seats, a knowledge base, and basic chatbot flows. I connected it to my website, wrote a 20-article knowledge base, and turned on their AI assistant.
What worked: The AI handled about 60% of incoming chat messages without any involvement from me. Order status, return policy, shipping times — it answered them instantly, 24/7. Customers got faster responses than I ever gave them manually. I got a dashboard where the remaining 40% of conversations waited for me, organized by urgency.
What didn’t work: Instagram DMs and email weren’t covered. Crisp’s AI is great for website chat, but customers message you everywhere. I still had to check four different inboxes. Also, the AI occasionally gave confidently wrong answers about edge cases — I had to review its responses weekly and update the knowledge base.
Crisp’s free tier is genuinely useful. The Pro plan ($25/month per workspace) adds AI-powered replies, a shared inbox for email, and better analytics. For a solopreneur, it’s the sweet spot between “free but limited” and “enterprise pricing.”
I compared Crisp to other automation tools in Make vs Zapier: Which One Is Actually Easier — the same integration-first thinking applies to support tools.
Approach 3: Intercom with Fin (Day 22–30)
Intercom is the enterprise-grade option, but their AI agent Fin is available on the $29/month starter plan. Fin reads your knowledge base and past conversations, then resolves support threads autonomously — not just suggesting answers, but actually closing conversations.
What worked: Fin resolved 74% of support conversations without human intervention. It handled multi-step issues (“I want to return item X but exchange it for item Y”) better than any tool I tested. The resolution reports showed exactly what Fin answered, so I could spot and fix errors. It also integrates with email, so messages from all channels funnel into one inbox.
What didn’t work: The $29/month starter plan includes only 10 Fin resolutions. After that, it’s $0.99 per resolution. If you get 200 support messages a month and Fin handles 70%, that’s ~$140/month on top of the base plan. For a solo business doing $5k/month, that’s fine. For one doing $1k/month, it’s a stretch.
I also noticed Fin sometimes escalated conversations it could have solved — the AI was being cautious, which is better than being wrong, but it meant I still got pinged more than expected.
If you want to build your own AI chatbot instead of using a platform, How to Build Your Own AI Chatbot in 30 Minutes walks through the process step by step.
What I’d recommend if you’re starting today
If you’re a solopreneur getting less than 50 customer messages per week:
Start with Crisp’s free tier. Set up a knowledge base (even 10 articles covers most questions). Turn on the AI chatbot for your website. This alone handles 50–60% of messages.
Use ChatGPT or Claude for email and DM replies. Keep a custom GPT trained on your tone and policies. Paste messages in, get drafts out, send with minor edits. This handles another 20%.
Upgrade to Intercom when you’re ready. Once you’re consistently getting 100+ messages/week and losing sleep over response times, Fin is worth the cost.
The tools I actually use every day include a mix of these — see The Tools I Actually Use Every Day for the full stack.
Mistakes I made (so you don’t have to)
I tried to automate everything at once. Don’t. Start with one channel (website chat or email), get it working well, then expand. I tried setting up AI across email, chat, Instagram, and contact forms simultaneously and spent more time debugging integrations than actually saving time. For more on this, The Mistakes I Made So You Don’t Have To covers the pattern.
I didn’t write a proper knowledge base first. AI tools are only as good as the information you feed them. I rushed through my FAQ, and the AI gave vague or wrong answers for the first week. Spend two hours writing clear, detailed answers to your top 20 questions before turning on any AI.
I forgot to tell customers they were talking to AI. This is a trust thing. I added a simple line — “This response was assisted by AI. If you need to talk to a human, just say so.” — and complaints about “robot responses” dropped to zero. People don’t mind AI. They mind not knowing.
For more on connecting tools together without code, How to Build Your First AI Workflow for Your Online Business covers the setup process. And Automate Client Follow-Ups Without Code shows how to handle the follow-up sequence after the initial support interaction.
The bottom line
AI customer support isn’t about replacing yourself. It’s about not being the bottleneck between your customers and the answers they need. Start with a knowledge base and Crisp’s free chatbot. Add ChatGPT for email replies. Upgrade when the volume demands it. You’ll get your mornings back — and your customers will get faster answers than you ever gave them manually.
Want help picking the right tool for your specific situation? Check out the AI Tool Advisor — I built it to match solopreneurs with the right tools based on what they actually need.