Mike HeapMike is an experienced Product Manager who focuses on all the “non-development” areas of My AskAI, from finance and customer success to product design, copywriting, testing and more.
Table of Contents
- Subscription platform AI customer service results on Intercom at a glance
- What does the platform do?
- Which helpdesk does the platform use?
- How did they train their AI customer service agent?
- When did they decide to turn on 'direct replies' to customers?
- What was the biggest thing they did to improve their AI agent's resolution?
- How do they customize their AI agent setup to work for their business?
- Guidance across all three types
- Routing the right tickets to the AI
- Reading non-English conversations on handover
- What impact is their AI customer service agent having now?
- Where do they go from here?
Created time
May 29, 2026 09:21 AM
Title length (<60)
Author
Last optimised
Sep 15, 2026
Ecomm?
Tags
Image
Publish date
May 29, 2026
Video
Slug
subscription-platform-ai-customer-support-case-study
Featured
Featured
Type
Case study
Ready to Publish
Ready to Publish
A Shopify subscription platform resolves 63% of around 2,200 monthly Intercom tickets with AI, holding 77% AI CSAT and saving roughly 705 hours every month.
This company sells subscription software to Shopify merchants, so their tickets are B2B: the people writing in run subscription programs for a living, and their questions are about configuration, billing behavior and migrations rather than a single order. A product like that throws off a steady stream of setup and "how do I" tickets, and a small team can't keep answering them by hand without falling behind.
Running our AI agent inside Intercom changed the math. They now resolve nearly two in three of those tickets automatically, customer satisfaction sits at 77%, and around 705 hours of agent time comes back every month.
Here's how it came together.
Subscription platform AI customer service results on Intercom at a glance
Fact | A Shopify subscription platform |
Industry | SaaS, subscription management for Shopify DTC brands (B2B) |
Helpdesk | Intercom |
Support volume | ~2,200 tickets/month |
AI resolution rate | 63% (roughly 1,386 tickets a month) |
AI CSAT | 77% |
Time saved | ~705 hours/month |
Key features used | Self-Learning, Guidance across all three types, an Intercom triage workflow, Live Translation, Train on Historic Tickets |
Knowledge sources | Intercom help articles via the Knowledge Base connector, website sync, historic Intercom tickets |
Previously evaluated | Intercom Fin (ruled out at $0.99 per resolution) |
Go-live mode | Direct replies from day one, with handover guidance and triage as the guardrails |
What does the platform do?
The product is a subscription-management platform built for Shopify. It gives DTC brands the machinery a subscription program needs: customer self-service, payment recovery, retention flows, and reporting.
Their customers are subscription brands selling direct, and a good share of new customers arrive from another subscription tool, so migration tickets are a steady part of the support load: a migration is a configuration project with a deadline and live revenue attached.
It's a lean team selling to operators rather than consumers. Each ticket is a merchant with a store running, so the questions are detailed and the stakes on any given one are higher than a consumer enquiry.
Which helpdesk does the platform use?
They run support in Intercom, and our AI agent sits right inside it.
They didn't land there by default. Before choosing us, they looked at Intercom's own Fin and turned it down on price. Fin charges $0.99 for every resolution, so the bill climbs with each ticket the AI closes (and it only climbs faster as the agent gets better at its job).
At their volume, the roughly 1,386 tickets the AI resolves each month would run to around $1,370 on Fin's per-resolution pricing alone. We price differently. The cost is flat, so the bill stays predictable while the resolution rate climbs.
Most of what lifts a resolution rate is the customer's own work: the knowledge they connect, the guidance they write, the triage they set up. Per-resolution pricing charges you more for the improvements you made yourself. Our rate does not climb as the agent gets better, so the cost per resolved ticket falls as it improves. Put your own volume into the Intercom Fin ROI calculator to see the gap at your numbers.
You'll find us in the Intercom App Store as "My AskAI: AI Customer-Service-Agent", installed straight into the Intercom workspace.
How did they train their AI customer service agent?
The agent is only ever as good as what it knows, so the team fed it three sources in order.
First, they connected their Intercom help articles through the Intercom Knowledge Base connector, giving the agent their existing documentation as a foundation.
Then they added more of their own content with website sync, pulling in extra pages from their site (for a configuration-heavy product, the docs carry a lot of the detail merchant questions hinge on).
Finally, they trained the agent on their historic Intercom tickets. Past conversations show it how the team actually answers in practice, the texture the help docs leave out, so it starts with real resolutions to learn from on day one.
When did they decide to turn on 'direct replies' to customers?
From day one. They went straight to direct replies, even on a product this configurable, and skipped the weeks in internal-note mode watching the agent draft.
That's a confident call for a product where a wrong answer touches a merchant's live billing. What made it safe was the setup underneath: handover guidance that routes anything high-judgment to a human, a triage workflow controlling which tickets reach the agent at all, and a cold start trained on real help articles, website content, and past tickets. With those guardrails in place, going direct on day one was a deliberate, well-tested default.
What was the biggest thing they did to improve their AI agent's resolution?
The single biggest lever was our Self-Learning, paired with heavy use of Guidance.
Self-Learning lets the agent draft and refine help articles from real conversations, so coverage grows as customers ask new things. Here it's used in roughly 1,000 ticket responses each month. For a product with a long tail of "how do I configure this" questions, that compounding effect is what keeps the resolution rate from stalling once the obvious tickets are handled.
Guidance does the other half of the job. Answering more questions is only part of it; the agent has to answer them the right way. The team wrote a lot of Guidance, so the agent asks for the details a useful answer needs (which store, which subscription flow, what the merchant is seeing) before it commits to a response, and hands off cleanly when something is out of scope.
Put together, the cold start from historic tickets and the compounding from our Self-Learning are how they reached 63% resolution and hold it there (and it keeps widening coverage every week the agent runs).
How do they customize their AI agent setup to work for their business?
A good AI agent should feel like part of the team. They got there with a few deliberate tuning moves (this is usually where the real difference is made).
Guidance across all three types
They configured all three kinds of our Guidance, and they didn't do it lightly. Communication guidance sets the tone, practical and direct, the way an operator mid-task wants a reply written.
Context and clarification guidance makes the agent gather specifics before answering, which matters when the right answer hinges on how one store has its subscriptions set up. Handover and escalation guidance decides what the agent shouldn't attempt (bugs, billing and account questions, anything sensitive), routing those straight to a human.
Routing the right tickets to the AI
Not every ticket should hit the AI first. They set up an Intercom triage workflow that routes the right tickets to the agent, so it handles the tickets it's suited to and the rest goes where it belongs.
Reading non-English conversations on handover
Their merchants aren't all English-speaking (a subscription program looks the same in any market). They use Intercom Live Translation, so when a conversation hands over to a human, the agent can read and respond to non-English threads without anyone losing the thread.
What impact is their AI customer service agent having now?
Here's where it all nets out for us, from the last 30 days:
- 63% AI resolution rate across their Intercom tickets.
- ~2,200 tickets handled per month, with roughly 1,386 of them resolved by the AI.
- ~705 hours saved per month, at around 30 minutes of agent handling time per resolved ticket. A merchant-facing configuration or migration ticket takes far longer to work than a consumer order-status query, which is why the hours add up the way they do.
- 77% AI CSAT from customers rating the AI's replies.
For a lean team supporting merchants, that's the outcome we like to see: a backlog that no longer grows faster than they can handle, and hundreds of hours redirected from repetitive answers to the work only people can do. Other teams have got there too: RecruitCRM turned Fin down on price and now sits at 68% resolution, and a high-volume prop-trading platform runs the same agent on Intercom at 73%.
Where do they go from here?
Their setup already covers knowledge, guidance, and routing well. The next gains come from giving the agent live, account-specific context and the ability to act.
The obvious next step is our User Data API, which would let the agent pull live account and plan details so it can answer questions about a merchant's own setup, the ones the docs can't reach on their own. From there, Tools and Tasks would let the agent take the action itself on a merchant's behalf (a natural fit for a configuration-heavy product).
And Self-Learning keeps working away in the background, so the knowledge base deepens with every conversation the agent has.
If you'd like to see more customer stories like this one, browse all our case studies.
