How to Handle Subscription (Recharge) Questions with AI in Gorgias
If you sell on subscription, a huge slice of your ticket volume is the same handful of questions asked over and over: skip my next box, swap the flavour, when's my next charge, pause me for a month, cancel please. Each one is quick for an agent who has the customer's Recharge subscription open in front of them — and painfully slow when they have to go dig it out. This guide shows how to answer those subscription questions faster in Gorgias, first with the native Recharge integration and macros, then with an AI agent that can read the subscription data on the ticket and act on it — and, just as importantly, where you should put guardrails so the AI never touches money without a human in the loop.
What the Gorgias–Recharge integration gives you first
Before any AI enters the picture, the foundation is the native connection between Gorgias and Recharge. Once it's installed, per the Gorgias Recharge documentation, your agents see the customer's Recharge subscriptions right in the ticket sidebar — active subscriptions, order history, and the all-important next charge date — without leaving the conversation.
More than a read-only widget, the integration lets agents take real subscription actions from inside Gorgias:
- Skip the next scheduled order (there's a literal Skip button in the Subscriptions section).
- Cancel or reactivate a subscription.
- Refund a charge, partially or in full.
- Update the customer's shipping address and see upcoming charge details.
To set it up, go to Settings → App Store → All Apps, search for Recharge, click Connect App, and authenticate your Shopify store. One caveat worth knowing up front: refunds only process cleanly for orders that went through Recharge checkout — orders fulfilled at Shopify checkout have to be cancelled on the Shopify side. That distinction matters the moment you start letting automation near refunds.
If you want the deeper tour of everything that widget can show, we cover it in the Gorgias customer sidebar explained.
Layer 1: macros that read live Recharge data
The fastest non-AI win is a set of macros with dynamic Recharge variables. Instead of a static "here's how subscriptions work" template, a good subscription macro pulls the customer's actual data — their next charge date, the price of their last subscription, the product on their plan — straight into the reply. An agent picks the macro, glances at the sidebar to confirm, and sends.
This is genuinely useful and you should build it. A "your next charge" macro and a "here's how to skip" macro will shave the routine subscription ticket from four or five minutes down to two or three. But notice the shape of the work: a human still reads the ticket, picks the right macro, verifies the data, and clicks send on every single one. Macros speed up the typing; they don't remove the human from the loop.
Layer 2: native Gorgias Automate and AI Agent — and where it stops
Gorgias's own automation stack goes further. Automate handles deflection with self-service flows and article-backed answers, and the AI Agent can read a ticket, reason over your help content, and reply autonomously on the questions it's confident about. For subscription FAQs that don't require touching the account — "how do I skip a box?", "when will my subscription renew?" answered generically — this is a real deflection win, and we walk through the whole stack in the complete Gorgias Automate AI guide.
Two honest limits are worth naming, though.
The billing model rewards full deflection, not partial help. Gorgias prices the AI Agent per resolved interaction — roughly $0.90 on annual billing and $1.00 monthly or on Starter, billed only when the AI fully resolves a conversation on its own without a human touching it (within about a 72-hour window), per Gorgias's own pricing breakdown. Overages on the Support + Shopping Assistant plans run $1.50 per interaction. And the AI Agent is a separately-priced add-on to your helpdesk plan — the two are billed separately — so a fully-automated ticket can carry both a helpdesk fee and the AI resolution fee. None of that is hidden; it's just a model where you pay for outcomes, and the price is fixed per resolution regardless of how much or how little work the AI actually did.
Reading and acting on live subscription data is the harder part. Deflecting a generic FAQ is one thing. Confidently answering "when's my next charge and can you skip it?" means reading this customer's Recharge record on the ticket and, on request, performing the skip. That's the line between deflection and true resolution — and it's where a dedicated AI agent layer earns its place.
Here's a way to think about the cost models side by side:
| Per-resolution (native AI Agent) | Per-action (AI agent layer) | |
|---|---|---|
| Billing unit | One fully-resolved conversation | One AI action (a lookup, a reply, a tool call) |
| You pay when | AI closes the ticket unaided within ~72h | The agent performs an action |
| Cost of a deflection | Flat per-resolution fee | Sum of the actions it took |
| Best fit | High-volume, generic FAQ deflection | Data-aware answers + gated write actions |
Neither model is "right" — they optimise for different things. Per-resolution is clean for pure deflection; per-action is transparent when the work is looking things up and taking gated steps. Credits on an agent layer are consumed per AI action, never per resolution — see the pricing page for how that's structured.
Layer 3: an AI agent that reads the subscription and acts — with guardrails
This is where Macha fits. Macha is an AI agent layer that runs on top of the Gorgias you already use — the connector is live, and Macha reads and writes the same tickets your team works. It does not replace Gorgias, Recharge, or your sidebar; it sits on top of them. (For the mechanics of wiring it up, see connecting Gorgias to Macha for automated ticket replies.)
The way an agent reads customer and account data is through custom tools: you point the agent at an API endpoint, and it can call that endpoint mid-conversation to fetch or update data. The screenshot below is Macha's real Custom Tools screen from our demo org — a GET "Get Order Shipping Status" tool alongside Postmark email actions. There's no subscription-lookup tool wired up in this particular demo org, so treat this as the honest, closest-match view of the surface: the exact place where you'd add a Recharge "get subscription" GET tool and a gated "skip next order" action.
With tools like these attached to a Recharge endpoint, the subscription playbook looks like this:
- Identify the customer from the ticket (email or order) and call a read-only "get subscription" tool.
- Answer the read-only questions autonomously — next charge date, what's on the plan, when the next box ships. No money moves, so the agent can just reply.
- Handle skip and swap as gated write actions. A skip changes when a charge happens; a swap changes what ships. The agent drafts the action and either executes it under a tight policy or hands a one-click confirmation to a human.
- Never let the agent move money unattended. Cancellations that trigger refunds, and refunds themselves, stay behind a human confirmation — always.
Guardrails on money actions — the part you can't skip
Automating subscription support is safe right up until the AI can move money, and then it becomes the single most important thing to get right. A few rules that hold up in production:
- Split reads from writes. Anything read-only (next charge, plan details, ship date) is fair game for autonomous answers. Anything that changes state (skip, swap) is a gated action. Anything financial (cancel-with-refund, refund) is human-confirmed, full stop.
- Respect the Recharge/Shopify refund boundary. Because refunds only process for Recharge-checkout orders, and fulfilled-at-checkout orders must be cancelled in Shopify, a naive "just refund it" automation will fail on exactly the tickets that are already sensitive. Keep a human on refunds.
- Log every action on the ticket. When an agent skips an order, that skip should be visible in the Gorgias thread so a human can audit and reverse it. Actions that don't leave a trail are actions you can't trust.
- Set a confidence floor and a clean handoff. If the agent isn't sure which subscription the customer means, or the request is ambiguous ("cancel everything"), it should hand to a human rather than guess.
This split — autonomous on reads, gated on writes, human on money — is what lets you automate the bulk of subscription volume without ever waking up to a batch of unauthorised refunds. The broader pattern of designing agents this way is covered in AI agents for customer service, and the tool-building mechanics live in custom tools.
Where each layer belongs
The clean division of labour: keep Recharge as the system of record for subscriptions, keep Gorgias as the helpdesk and the sidebar your agents trust, use macros and native Automate for generic deflection, and add an AI agent layer for the data-aware answers and gated actions the other layers can't do. Each one does what it's best at, and none of them replaces the other. For a fuller picture of how the agent layer reasons, the Gorgias AI Agent explained is a good next read.
FAQ
Can AI answer "when is my next charge?" from Recharge data in Gorgias? Yes. That's a read-only question, so it's the safest thing to automate. With a custom tool pointed at your Recharge subscription endpoint, an AI agent can read the customer's next charge date from the ticket context and answer it autonomously — no human needed and no money moved.
Can the AI skip or swap a subscription automatically? It can, but it shouldn't do so blindly. Skips and swaps change state, so the right pattern is a gated write action: the agent prepares the skip or swap and either executes it under a tight policy or presents a one-click confirmation to a human. That keeps speed high without handing state changes over unattended.
Should AI handle cancellations and refunds? Keep those human-confirmed. Cancellations often trigger refunds, and refunds only process cleanly for Recharge-checkout orders (fulfilled-at-checkout orders must be cancelled in Shopify). Money actions are exactly where an unattended AI causes the most damage, so gate them behind a person.
Do I have to replace Gorgias or Recharge to add AI? No. Macha is an AI agent layer that runs on top of your existing Gorgias and Recharge setup via a live connector. It reads and writes the same tickets your team works and uses the same Recharge data — it doesn't replace your helpdesk, your subscription platform, or your sidebar.
How is an AI agent layer billed compared with Gorgias's AI Agent? Gorgias's native AI Agent is billed per fully-resolved interaction (about $0.90 annual / $1.00 monthly) as a separate add-on to your helpdesk plan. An agent layer like Macha bills per AI action — the individual lookups, replies, and tool calls the agent performs — which is more transparent when the work is data-aware answers and gated actions rather than pure FAQ deflection. See /pricing for the structure.
Ready to answer subscription questions straight from the Gorgias ticket — without ever letting the AI move money on its own? Start a free trial of Macha and connect it to your Gorgias in minutes.
Add AI agents to your Gorgias
Macha resolves tickets end to end on Gorgias — no migration, no code.
Zendesk
Freshdesk
Gorgias
Front
Shopify
Stripe
Slack
Notion
Google Workspace
Confluence

