Download PDF ↓

TAX REFUND PLANNING & AUTOMATION

The Opportunity

For most working households, the tax refund is the largest single deposit of the year. This season the IRS refunded $324.8 billion to 99.1 million households, an average of $3,276 per household. This amount is meaningful to most Americans, and especially to the caring professionals we serve.

Two problems sit inside that moment. First, people leave money unclaimed. The average caring professional is eligible for roughly $5,900 in tax credits, and $12 billion in credits goes unclaimed nationally every year. Second, money that arrives without a plan for its intended use gets absorbed into day-to-day spending. We built our tax product to address both: free filing that maximizes refund value, and a refund planning feature that members use to set goals for their refund before they file.

WHY PRECOMMITMENT

The refund planning feature’s design rests on a well-established behavioral finding called precommitment: people follow through more reliably when they decide how they want to use funds before the money arrives. The most relevant evidence is a randomized controlled trial run by the Center for Advanced Hindsight at Duke University that inspired this feature. In that experiment, moving one question (“What percent would you like to save?”) from after the refund arrived to before it arrived raised savings rates from 17% to 27% and generated over $1 million in savings. Three months later, 85% of that savings was still in place. The timing of the ask drove the result.

How It Works

Members file their taxes in the app for free with maximum-refund and 100% accuracy guarantees. Before they file, they can make a refund plan: they decide how much to allocate to savings, how much to allocate to spending, and how much to keep in checking. A member expecting $3,000 might plan to put $1,200 toward paying off a high-interest credit card, $1,000 into Essential Savings, and $800 toward a laptop. Plans can be adjusted any time before the refund arrives.

Then, our agentic automation layer runs on each member’s behalf, without requiring them to do anything. Software monitors the member’s transactions, recognizes the refund deposit, reads the plan, adjusts based on the actual refund amount and current user context, and moves the money accordingly. Every action is visible and reversible, so the user maintains ultimate control. The member decides once, in a moment of calm, and the agent handles the rest.

This past year, we enabled members who filed outside The Beans to also benefit from the refund planning feature. They make a plan, the agent watches their linked account for the refund, and moves the money when it arrives.

Results — 2026 Season

In Q1 2026, members made plans for $300,000 in refunds and committed $100,000 of it to automated execution. 14% of new app users adopted an agent. This rate is 3× the 5% benchmark for top-tier adoption shared by growth leaders at Chime, SoFi, and Brex.

Filing completion is the other headline number: across two tax seasons, 57% of members who started filing submitted their returns through our platform, roughly four times the industry norm. We attribute this to our high engagement, user trust, and to the support offered by our platform focused on maximizing tax credit utilization.

Our agent layer also grew alongside the feature: from 82 agents to 264 in one quarter, with $750,000 committed annually to automations.

Why This Scales

The Open Problem: Detection

Automation only works if we can see the refund arrive. When members file with The Beans, that more likely, we can accept the refund directly or at least have precise information for the agents to consume. For members who file elsewhere, we depend on reading their bank transactions, and the industry’s tools fall short there. Today, aggregators identify only about half of refunds in transaction data. Our patented models, layered on aggregator feeds, do better, but not by enough to call the problem solved. Better detection converts committed dollars into automated dollars. If you think this is as fun and interesting a challenge as we do, please get in touch. We’ve been deep in transaction data since day one and want to connect with other great minds working on these issues.

What’s Next

Season three launches in the fall ahead of early filers (instead of behind them, as in seasons one and two). The same pre-commitment and automation engine will run year-round on bonuses and windfalls. The goal is unchanged: caring professionals claim the money they are entitled to, and it goes where they intended.