From 200 to 800 weekly support tickets handled. Zero new hires. Here’s the AI stack, the build timeline, and the honest ROI math at 60 days.
Clover & Co. sells premium home goods direct to consumer, the kind of brand with a loyal customer base, a strong repeat purchase rate, and a support inbox that was quietly destroying the business from the inside.
At $2.4M in annual revenue with 14 employees, they had two full-time support reps handling roughly 200 tickets per week. That sounds manageable until you factor in what support looked like for a DTC brand in 2025: WISMO emails ("where is my order?"), return requests, damage claims, size exchange questions, and personalized order inquiries, all requiring individualized responses that a canned reply couldn't handle without feeling cold.
The two reps were maxed. Tickets were getting 48-72 hour responses during peak weeks. Customer satisfaction scores were slipping. The founder, Reyna, was seeing it in the data: customers who waited more than 24 hours for a support response had a 34% lower repeat purchase rate than customers who got a same-day reply. That correlation had a dollar value, and it wasn't small.
The obvious answer was hiring a third support rep. Fully loaded cost: approximately $52,000/year. Reyna's question before pulling the trigger: was there a way to solve the capacity problem without permanently adding that cost to the P&L?
“I could see in the data that slow support was costing us repeat customers. The question wasn’t whether to fix it, it was whether the fix had to be a permanent headcount addition.”
Reyna M., Founder, Clover & Co.AI Receptionist was deployed on the customer-facing phone line first. A surprising number of support interactions at Clover & Co. came in by phone, particularly from older demographics who'd purchased as gifts. The AI Receptionist handled all inbound calls: WISMO inquiries using order lookup, return policy questions, and general FAQs. Human reps were only roped in for situations the AI couldn't resolve, damaged shipments requiring photo review, complex exchanges, and escalations.
Copy.ai was layered onto the email support workflow. Rather than automating responses entirely, the team used Copy.ai to generate first-draft responses for their reps. A rep would review the AI draft, edit for accuracy and tone, and send. What previously took 8–12 minutes per ticket now took 2–3 minutes. Same human oversight, 75% less time per interaction.
Week one was AI Receptionist setup, straightforward given they were already on Nextiva. The FAQ and policy knowledge base took the bulk of the time: writing clear answers to the 40 most common support questions in a format the AI could use reliably. This content creation work took two days but paid dividends immediately.
Week two was Copy.ai integration into the email workflow. The team built prompt templates for each support ticket category: WISMO, return requests, damage claims, and general inquiries. Each template was tested against 20 real historical tickets before going live. Rep feedback at the end of week two: "It's like having a first-draft assistant. I still approve everything but I'm not staring at a blank screen anymore."
Weeks three and four were refinement. The AI Receptionist FAQ was expanded based on actual call transcripts. The Copy.ai templates were tuned based on cases where the first draft required significant editing. By week four, average editing time per AI draft was under 90 seconds.
At the 60-day mark, Reyna pulled the numbers. Ticket volume had increased 18% (seasonal) while the team stayed the same size. Total tickets handled per week: 820, up from 200 before the AI stack. The math on that increase: the AI Receptionist was handling approximately 340 calls per week that previously generated email tickets or missed calls. Copy.ai was compressing email ticket handling time by 73% on average.
The 82% autonomous resolution rate from AI Receptionist meant that only 18% of inbound calls required rep involvement, and those calls were shorter because callers had already received context from the AI about their order status or return options.
CSAT scores: up 18 points from the pre-deployment baseline. Response time average: down from 51 hours to 6 hours. Repeat purchase rate correlation to support response time: the trend reversed, customers who'd previously been in the 48-72 hour response bucket were now receiving same-day replies and showing early repeat purchase signals consistent with the faster-response cohort.
ROI at 60 days: the combined tool cost was approximately $420/month. The estimated value of recovered repeat purchase behavior in the first two months, based on cohort analysis, was approximately $13,400. The hire that didn't happen: $52,000 annually. Reyna's conclusion: "We avoided a $52K hire, improved CSAT significantly, and we're seeing the repeat purchase data trend in the right direction. That's 3.2x ROI in 60 days, conservatively."
The two-tool AI stack cost $420/month and replaced the need for a $52,000 annual hire while handling 4x the support volume with better CSAT scores. The key insight: they didn't remove humans from the loop. They used AI to eliminate the parts of support work that didn't require human judgment, drafting, WISMO lookups, FAQ responses, and let the reps focus on the 18% of cases that actually need a person. That's the right deployment model. AI doesn't replace good support. It makes good support affordable at scale.
Reyna flagged one ongoing risk: Copy.ai draft quality degrades when a ticket falls outside the established categories. The AI generates a draft, but it's often off-tone or off-topic enough that the rep effectively rewrites it from scratch. The fix is ongoing template maintenance, reviewing edge cases monthly and adding new prompt templates for emerging ticket categories. This takes about 90 minutes per month and has largely kept the issue contained.
The AI Receptionist occasionally struggles with callers who want to place new orders by phone (a legacy behavior from older customers who distrust online checkout). The current routing sends these callers to a human rep. There's no AI solution for live order entry without deeper system integration. Something to plan for if your customer base includes a significant phone-order segment.
The best AI tools, real case studies, and actionable guides, delivered every Thursday. No noise. Just signal.