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
- Crypto-tax AI customer service results on Intercom at a glance
- What does the company do?
- Which helpdesk does the company 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?
- Setting the communication style
- Knowing when to clarify and when to escalate
- What impact is their AI customer service agent having now?
- Where do they go from here?
Created time
Jun 3, 2026 09:23 AM
Title length (<60)
Author
Last optimised
Sep 15, 2026
Ecomm?
Tags
Image
Publish date
Jun 3, 2026
Video
Slug
crypto-tax-ai-customer-support-case-study
Featured
Featured
Type
Case study
Ready to Publish
Ready to Publish
A crypto-tax software company resolves 53% of ~951 monthly Intercom tickets with AI, holding 82% AI CSAT and saving ~42 hours every month.
This company sells crypto-tax software, and crypto-tax support is a deep, seasonal kind of work. The right answer to "how is this taxed?" changes depending on which country a user files in, and one wrong figure on a tax report is the sort of mistake that loses trust fast.
For a lean team, that adds up to a lot of repetitive, high-stakes questions. So they put our AI to work inside Intercom, and they now resolve 53% of their support tickets with it (roughly 951 a month), while holding an 82% AI CSAT and handing back about 42 hours of agent time every month.
Here's how it came together.
Crypto-tax AI customer service results on Intercom at a glance
Fact | A crypto-tax software company |
Industry | Fintech, crypto-tax software and portfolio tracking (B2C) |
Helpdesk | Intercom (Messenger) |
Support volume | ~951 tickets/month |
AI resolution rate | 53% (around 504 tickets a month) |
AI CSAT | 82% |
Time saved | ~42 hours/month |
Key features used | Self-Learning, Custom Answers, Guidance across all three types |
Knowledge sources | Website sync, an internal Notion workspace via our Notion connector, Custom Answers |
Go-live mode | A setup phase tuning escalation behavior first, then direct replies in Intercom |
What does the company do?
The product is crypto-tax software. It connects to a user's exchanges and wallets, pulls in their transaction history, sorts it, and turns the lot into tax reports ready to file.
The user base is large and international, and every new country, exchange, and transaction type the product supports is another set of edge cases a user might ask about. It is a pattern we see across crypto companies running AI support, where a lean team, a large international user base and answers that depend on where the user is all arrive together.
Which helpdesk does the company use?
They run their support through Intercom, and that's where our AI sits. It answers customers directly in the Intercom Messenger through our Intercom integration, working from the same knowledge the human team relies on.
There was no previous AI tool to rip out. Intercom is their support platform, and we're the AI layer added on top of it, so the rollout was about training and tuning an agent inside an inbox the team already lived in (no migration, no new system to learn).
How did they train their AI customer service agent?
The team gave the agent its knowledge from three sources, in order of leverage.
First, website sync pulled in their public site and help content (the front-line answers that cover most common questions). Second, our Notion connector wired in their internal Notion workspace.
That second source matters more than it sounds for a tax product. The nuanced, country-by-country answers (how an airdrop is taxed in Germany versus Australia, say) live in internal docs, well away from the public marketing site. Connecting Notion put that depth in front of the AI without anyone having to rewrite it as help articles first.
On top of those two, they added a small set of Custom Answers (exact, controlled replies for the handful of questions where the wording has to be just so).
The upshot is that the agent works from the same internal knowledge a trained agent would, so it can handle the country-specific questions that make crypto-tax support hard.
When did they decide to turn on 'direct replies' to customers?
The agent answers customers directly in Intercom today, so the 53% resolution rate and 82% CSAT are measured on the live replies customers actually received.
Getting there was deliberate rather than instant. Before trusting the AI with direct replies, the team spent time in a setup phase getting the escalation behavior right (making sure the agent stopped cleanly once a conversation was handed to a human, instead of carrying on replying).
Luckily, once that handover behavior was solid, flipping to direct replies was an easy call.
What was the biggest thing they did to improve their AI agent's resolution?
The single biggest lever has been Self-Learning. When the agent hits a question its current knowledge can't answer, Self-Learning drafts a new help article to cover it, and once that article exists, the next identical question gets answered automatically.
And the effect compounds. In the last 30 days, articles the AI wrote itself through Self-Learning resolved around 100 tickets (fun fact: that's roughly one in five of every ticket the AI handled).
For a product where the same import quirks and country-specific questions come up again and again, that loop turns each new edge case into a one-time problem. It's the kind of steady climb we see across other Intercom rollouts too.
It works because of the knowledge underneath it. With the internal Notion workspace and the website already connected, Self-Learning has rich material to draw on when it writes a new answer, so the drafted articles are accurate and hold up in front of customers.
How do they customize their AI agent setup to work for their business?
Most of the "make it ours" work happens through Guidance, which they have set up across all three types.
Setting the communication style
Communication Guidance controls how the agent writes (the tone and structure it uses when it explains something as sensitive as a user's capital gains or walks them through a report). It keeps the AI sounding like the brand, in its own voice.
Knowing when to clarify and when to escalate
Context & Clarification Guidance tells the agent to ask for the missing detail before answering (which exchange, which country, which tax year). That directly fixes one of their most repetitive patterns: users who need a country-specific answer but haven't said which country they're in.
Handover & Escalation Guidance covers the other side, which is knowing when not to answer. Account-specific issues, anything that edges toward tax advice, and other sensitive cases get routed to a human. This is the behavior the team tuned most carefully before letting the AI reply directly.
What impact is their AI customer service agent having now?
The numbers from the last 30 days:
- 53% AI resolution rate across ~951 monthly Intercom tickets (around 504 handled by the AI without ever reaching a human).
- ~42 hours saved per month, at about five minutes per resolved ticket.
- 82% AI CSAT on the answers customers actually received.
- ~100 tickets a month resolved by articles Self-Learning drafted on its own.
For a lean team facing deep, seasonal, country-specific questions, that's half the inbox handled before a person ever steps in. (Not bad for an agent that was answering its first ticket a few months ago.)
Where do they go from here?
The next move is from answering to doing. They're building their first Task, an API-triggered tax-report resend, so when a user asks "can you resend my report?", the AI can trigger it directly. It's exactly the kind of repetitive, mechanical job that suits an agentic workflow.
Alongside that, they're working on passing user attributes from Intercom (country, subscription status, payment plan) straight into each conversation. Once the AI knows who it's talking to, it can stop asking users to repeat their context and tailor answers automatically.
They're also planning to lean on Inspect to audit where live answers are coming from, and to keep deepening Self-Learning as their knowledge base grows. We love to see this direction. More of the routine work goes to the AI, so the human team can spend its time on the tricky tax questions.
If you'd like to see more customer stories like this one, browse all our case studies, or take a look at our pricing.
