---
title: "Fintech AI Customer Support Case Study: GiveCard 95%"
description: "GiveCard, a fintech disbursements platform, resolves 95% of its monthly Zendesk tickets with AI at 90% CSAT, saving ~20 hours a month. Here's how."
url: https://myaskai.com/blog/my-askai-givecard-case-study
---
# Fintech AI Customer Support Case Study: GiveCard 95%

GiveCard resolves around 95% of its Zendesk tickets each month at 90% AI CSAT saving the team roughly 20 hours every month.

GiveCard moves money to people who need it, fast. So the questions landing in their inbox carry weight: a card that won't work, a balance that looks wrong, a payment that hasn't arrived. The people asking are often in a tight spot, and the team answering them is small.

After a burst of growth, that volume started to outrun what a handful of people could keep up with by hand. Instead of hiring their way out, GiveCard put our AI agent at the front of their Zendesk inbox. It now resolves about 95% of incoming tickets at 90% CSAT, and it got there with no help center connector and no order-data feed behind it.

Here's how it came together.

## GiveCard AI customer service results on Zendesk at a glance

| Fact | GiveCard |
| --- | --- |
| Website | [GiveCard](https://www.givecard.com/) |
| Industry | Fintech, prepaid cards and fund disbursement (B2C cardholder support) |
| Helpdesk | Zendesk (Support Inbox / Tickets) |
| Support volume | ~255 tickets/month |
| AI resolution rate | 95% |
| AI CSAT | 90% |
| Time saved | ~20 hours/month |
| Key features used | Custom Answers, Self-Learning, Communication Guidance, Context & Clarification Guidance |
| Knowledge sources | Uploaded internal docs, Custom Answers library, Self-Learning auto-drafted articles (no help center connector) |
| Go-live mode | Direct replies to cardholders inside Zendesk |

## What does GiveCard do?

GiveCard calls itself "the simplest way to disburse funds at scale." They issue prepaid cards and run bank transfers so that governments, nonprofits, businesses and social-service agencies can get money to the people they serve. Those people are often the ones the banking system forgets: disaster survivors, research participants, benefit-program recipients.

GiveCard, the simplest way to disburse funds at scale

The reach is bigger than you'd guess for a young company. [GiveCard](https://www.givecard.com/) has issued hundreds of thousands of cards, works with 100+ government agencies, and runs in every US state.

## Which helpdesk does GiveCard use?

GiveCard runs support on [Zendesk](https://myaskai.com/ai-agent-integration/zendesk-tickets). As they scaled, they pulled their email support onto Zendesk, and that general inbox is where our agent now sits, working as a flat layer at the front of every ticket and clearing the routine questions before a human ever opens them.

The fit is a practical one. Most of GiveCard's contact comes in by email and phone, because their cardholders aren't living in a chat widget. Put our agent at the front of the email queue and the everyday questions get an instant answer, which leaves the team free for the cases that need a person.

## How did GiveCard train their AI customer service agent?

This is where GiveCard's setup goes off the usual script. Most teams start by plugging in an existing help center and letting the AI answer from those articles. GiveCard didn't have a deep public help center to lean on, so the agent had to be taught from scratch.

The training came down to three things:

- **Uploaded files.** They fed the agent their internal docs: how each program works, what cardholders can and can't do, the policy behind every kind of card and transfer.
- **A library of Custom Answers.** For the questions where the wording has to be exact (eligibility, balances, what a card can be used for), they wrote Custom Answers so the agent gives the precise, approved reply every time. It's the boring-but-effective option, and on money questions boring is exactly what you want.
- **Self-Learning.** Whenever a human answered something the AI couldn't, our [Self-Learning](https://myaskai.com/features/self-learning) feature drafted that answer into reusable knowledge, so the agent resolved the next identical question on its own.

None of this needed a [pre-built knowledge base](https://myaskai.com/features/knowledge). GiveCard built the agent's knowledge from the ground up, and Self-Learning kept extending it on its own.

GiveCard's three-part AI training stack: uploaded internal files, a Custom Answers library, and Self-Learning that drafts new knowledge from human replies.

## When did GiveCard decide to turn on 'direct replies' to customers?

Our agent replies directly to cardholders inside Zendesk, screening the routine email before a human sees it. The 90% CSAT sitting next to the 95% resolution rate is the tell that those replies are landing: customers are getting answered, and they're rating the answers well.

For a team handling money for vulnerable people, that's a real act of trust. It means our agent is the first voice a cardholder hears, on the questions where a wrong answer would sting.

## What was the biggest thing GiveCard did to improve their AI agent's resolution?

[Self-Learning](https://myaskai.com/blog/what-is-self-learning-ai) is the engine behind the 95%.

Because GiveCard started with no help center to connect, the agent's knowledge had to be built up over time. The loop is a simple one: a customer asks something the agent can't yet answer, a human replies, and our Self-Learning feature turns that reply into a drafted article. Next time the same question shows up, the agent handles it alone.

That compounding adds up fast. Those auto-drafted articles have already powered over 300 ticket responses for GiveCard (fun fact: that's knowledge that didn't exist the day the agent went live, all written off the back of real conversations). Around it, the Custom Answers library keeps the money-and-eligibility wording exact.

[Play video](https://www.youtube.com/watch?v=RT_6jQ_Iwpo)

Put together, that's how a knowledge-led setup with no connected help center still cleared 95%. For context, that sits well above the roughly 70% [median resolution rate](https://myaskai.com/blog/ai-resolution-rate-benchmarks) we see across the field of AI support deployments. I've been on enough rollout calls to know how rare a number like that is on a setup this lean.

## How do GiveCard customize their AI agent setup to work for their business?

The agent feels like GiveCard's own, and that comes down to how they've tuned it with [Guidance](https://myaskai.com/features/guidance).

GiveCard tunes its AI agent with two Guidance types: Communication Guidance for a clear, calm tone, and Context & Clarification Guidance to gather the right details before answering.

### Keeping replies clear and on-brand

They use Communication Guidance to control how the agent talks (and getting that tone right matters when the subject is someone's money). A lot of their cardholders are unbanked or underbanked, and many are reaching out under stress, so replies have to be plain, calm and jargon-free. The Guidance rules hold that tone steady, which counts for a lot.

### Asking the right question before answering

The trickier problem is that the same question can have different answers depending on the program a cardholder is on. Whether a card works at an ATM, say, depends on that program's rules. They use Context & Clarification Guidance so the agent gathers the right details before it answers, instead of guessing and getting it wrong (which is a big part of why the resolution rate holds up across so many programs).

## What impact is GiveCard's AI customer service agent having now?

The scores on the doors, last 30 days:

- **95% AI resolution rate**, well above the ~70% field median for AI support.
- **~20 hours saved per month**
- **90% AI CSAT** across AI-handled tickets.
- **300+ ticket responses** powered by knowledge Self-Learning drafted on its own.

It’s allowing GiveCard to run support with 1 support agent for every 30,000 cardholders, that's real leverage. GiveCard promises round-the-clock human support, and our agent is how a team that size keeps that promise: it soaks up the routine volume so the humans stay on the cases that need them, and nobody has to hire ahead of the growth.

GiveCard results over the last 30 days: 95% AI resolution rate, 90% AI CSAT, and ~20 hours saved per month.

## Where do GiveCard go from here?

GiveCard's roadmap is mostly about reaching the questions the agent can't yet resolve on its own, the ones that need live data.

The clearest next step is connecting our [User Data API](https://myaskai.com/features/user-data) so the agent can look up a cardholder's account and recent transactions. Transaction questions (a declined card, a charge that looks wrong) are GiveCard's single most common ticket type, and right now they still need a human to pull up the card ID and check.

Give the agent secure access to that data and a big slice of those tickets becomes self-serve too. From there, [Tasks](https://myaskai.com/features/tasks-and-tools) would let the agent act rather than only answer: activating or replacing a card, checking a balance.

They're also exploring a multi-agent setup, using Zendesk's routing to give different programs their own tailored agent. And [Handover & Escalation Guidance](https://myaskai.com/features/chatbot-to-human-handoff) (the one Guidance type they haven't switched on yet) is a natural next addition as the agent takes on more sensitive cases.

If you'd like to see more customer stories like GiveCard's, [browse all of our case studies](https://myaskai.com/blog/tags/case-study-posts), or take a look at [our pricing](https://myaskai.com/pricing) to get started.

- Case Study ### [How Honeygain achieves 90% AI resolution, saving 507 hours each month](https://myaskai.com/blog/my-askai-honeygain-case-study) Honeygain hit a 90% Zendesk AI resolution rate, holding 78% AI CSAT across ~3,400 monthly tickets and saving around 507 hours a month. Here's how. 3 Jun 2026  6 min read
- Case Study ### [Sofar Sounds: How My AskAI hits 85% CSAT by escalating most tickets to humans](https://myaskai.com/blog/my-askai-sofar-sounds-case-study-2026) Sofar Sounds runs ~750 monthly Zendesk tickets through AI triage. 85% CSAT, 16 hours saved, and most tickets deliberately escalated to humans. 25 May 2026  12 min read
- Case Study ### [How a creator-payments platform runs AI copilot inside Zendesk](https://myaskai.com/blog/creator-platform-ai-customer-support-case-study) A creator-monetization platform runs My AskAI as a Zendesk copilot, drafting answers on ~1,400 tickets a month with a human on every single reply. 3 Jun 2026  7 min read
- Case Study ### [How Apartment List achieves 76% AI resolution, saving 101 hrs each month](https://myaskai.com/blog/my-askai-apartment-list-case-study-2026) Apartment List resolves 76% of ~1,582 monthly Zendesk tickets with AI — 97% AI CSAT and saving ~101 hours every month. Here's how. 15 May 2026  8 min read
- Moving to AI First Support ### [What Is a Good AI Resolution Rate? Benchmarks From 195 Real Deployments](https://myaskai.com/blog/ai-resolution-rate-benchmarks) Everyone asks "what's a good AI resolution rate?" and gets a hand-waved number. We pulled real data from 195 deployments across 38 vendors. Here's the truth. 2 Jun 2026  15 min read
- Moving to AI First Support ### [6 Best AI Customer Service Agents for Banks & Neobanks (2026)](https://myaskai.com/blog/best-ai-customer-service-banks-2026) AI customer service for banks means KYC chases, frozen cards and account-closure threats. Here are 6 AI agents that handle it without faking compliance. 19 Jun 2026  31 min read

## Start using AI customer service in your business today

Set up an AI agent on your helpdesk in minutes. Pay only for the tickets it resolves.

[Create AI customer service agent   (opens in a new tab)](https://myaskai.com/create-account)

[Book a demo   (opens in a new tab)](https://myaskai.com/book-a-demo)

My AskAI