AWS Builder Center
Savvy: Building an AI-Powered Rewards Assistant with AWS Strands

Savvy: Building an AI-Powered Rewards Assistant with AWS Strands

What if your wallet could tell you the smartest way to pay? We built Savvy to help people maximize cashback, rewards, discounts, and offers, combining deterministic financial logic with an AI agent powered by AWS Strands.

Coding

Savvy: Make Your Payment Worth More

Every payment feels simple.
You choose a card, scan a QR code, enter your details, and move on.
But behind that simple payment can be a surprisingly complicated decision.
One card might offer cashback. Another might give reward points. A bank may have an instant discount. A merchant could have a special offer. And the rewards you've already accumulated might have a completely different value depending on how you redeem them.
So we asked a simple question:
What if your wallet could tell you the smartest way to pay?
That question became Savvy.

The Problem

Most people don't have a clear picture of all the value available across their financial products.
They may own multiple credit cards, but don't know which one is optimal for a particular purchase. They may have thousands of reward points sitting unused. They may miss offers simply because they don't know they exist.
Existing solutions often focus on helping users compare or choose credit cards.
We wanted to focus on something different:
What should I use for this purchase, right now?
For example, if someone is about to spend ₹5,000, the answer shouldn't simply be "use Card A."
It should consider the bigger picture:
  • What cashback can I earn?
  • Are there relevant discounts or offers?
  • How many reward points will I receive?
  • What is the effective cost of the purchase?
  • What rewards do I already have?
  • Is there another payment method that gives me more value?
That is where Savvy starts.

From Card Comparison to Payment Intelligence

Our initial idea was to build a simple credit-card recommendation tool.
But as we explored the problem, we realized that optimizing a payment is about much more than choosing a card.
A user's wallet contains cards, rewards, offers, spending patterns, and different ways to redeem value.
So the product evolved.
Instead of building another card comparison website, we started building a personal rewards intelligence platform.
Savvy brings together:

Purchase Optimization

Enter what you're buying and how much you're spending, and Savvy helps identify the most valuable way to pay.

Wallet Analysis

Understand how your existing cards and rewards fit together and where you may be leaving value unused.

Offers

Discover relevant merchant, bank, and payment offers that can affect the effective cost of a purchase.

Rewards Vault

See the rewards you've accumulated and understand the ways they can potentially be used.

Insights

Understand spending and rewards patterns and identify opportunities to improve how you earn.

AI Assistant

Instead of navigating through multiple screens, users can simply ask questions about their wallet and spending.
This last part became particularly interesting when we started exploring agentic AI.

Why We Used AWS Strands

We didn't want AI to simply generate another conversational answer.
We wanted the assistant to actually work with Savvy's capabilities.
A user might ask:
"I'm spending ₹8,000 on a flight. What's the best way to pay?"
To answer that properly, the system may need information about the user's wallet, relevant rewards, available offers, and the actual purchase optimization result.
This is where we integrated the AWS Strands Agents SDK.
Strands provides the agent layer that can reason about the user's request and determine what information or capabilities it needs to answer the question.
Instead of treating the assistant as a single prompt, we can think of it as an agent that can interact with Savvy's capabilities.
The flow becomes:
User → Strands Agent → Savvy capabilities → Deterministic result → Explanation
The agent can retrieve relevant information, use the appropriate tools, and then turn the resulting information into a conversational response.
But there is one important design decision we made.

AI Should Explain the Numbers, Not Invent Them

Financial recommendations are different from ordinary conversational answers.
If Savvy says:
"You can save ₹350 on this purchase."
we need to know exactly where that number came from.
We therefore kept the financial calculations deterministic.
Savvy's optimization logic remains responsible for calculating things such as cashback, rewards, discounts, and effective purchase cost.
The Strands agent does not make up these values.
Instead:
The deterministic engine calculates.
The agent reasons about what it needs.
The agent explains the result.
This separation gave us a way to combine the flexibility of an AI assistant with predictable financial calculations.

Building the Experience

Once the core idea was clear, we focused on making the product feel simple.
The user shouldn't need to understand reward programs, complicated card benefits, or optimization rules.
They should be able to tell Savvy what they're trying to do and get an understandable answer.
For example:
"I want to spend ₹5,000 at this merchant."
Savvy turns that into a useful recommendation rather than forcing the user to manually compare several cards and offers.
The same idea extends to questions such as:
"Which of my cards is best for travel?"
"How can I get more value from my rewards?"
"Am I missing any important offers?"
The interface is designed around making those questions easier to answer.

What We Learned

One of our biggest lessons from building Savvy was that adding AI isn't automatically the same as building an AI product.
The useful part is deciding where AI should be responsible and where it shouldn't be.
We could have allowed a language model to calculate everything, but that would make the system harder to trust.
Instead, we gave each layer a clear responsibility.
The rules and optimization layer handles the numbers.
The agent handles reasoning and orchestration.
The conversational layer makes the results understandable.
This made the architecture much more natural for our use case.
We also learned that a small question can hide a large product opportunity.
"Which card should I use?"
sounds simple.
But answering it properly can involve the user's wallet, spending behavior, rewards, offers, merchant information, and redemption possibilities.
That realization is what transformed Savvy from a card recommendation idea into a broader payment intelligence platform.

What's Next for Savvy?

Savvy is still at an early stage.
There is a lot we want to explore—from better personalization and richer offer discovery to smarter reward redemption and more useful spending simulations.
Our long-term vision is not simply to help people choose another credit card.
We want Savvy to help people get more value from the financial products they already use.
Before making a payment, you shouldn't have to open five different apps and remember dozens of reward rules.
You should be able to ask one simple question:
"What's the smartest way to pay?"
And that's the problem we're building Savvy to solve.
Savvy — Make Your Payment Worth More.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article