Autopilot moves Cleo from analysis to action. She can now map out your long-term goals and help you take action to reach them, all while providing deeply personalized insights into your daily spending.
Last July, Cleo 3.0 introduced agentic architecture, advanced reasoning, individualized memory, and real-time voice. Autopilot continues our trajectory toward increasingly powerful AI financial assistance. With this update, Cleo’s able to understand your money better than ever, then act on what she knows.
How Autopilot works
Each Autopilot component builds on the last:
Roadmap: Cleo creates a long-term plan based on a big-picture goal.
Daily Plan: Cleo keeps you on track with personalized daily insights and recommendations.
Actions: Today, Cleo recommends financial actions; tomorrow, she executes them for you.
As your financial state changes, Cleo recalibrates, so her guidance is always relevant to your actual position.
It all happens within the same conversational interface users already know, with no separate configuration required. You tell Cleo what you want to accomplish, and Autopilot coordinates the underlying systems to carry it out.
Insights
Insights is where Cleo learns who you are financially. When you enable Autopilot, Cleo works through years of your transaction history to build a model of how you manage your money, including income patterns, spending distribution, recurring expenses, and shifts over time.
This baseline serves two purposes: Downstream components depend on it for planning, and Cleo draws on it to surface insights about your spending and habits. By the time Autopilot starts generating a plan, Cleo already has a structured view of how money flows through your accounts and which constraints will matter most for you.
Roadmap
Next, Roadmap translates that baseline into a long-term plan. Based on your current financial state and your big-picture goal, Cleo charts a path forward and breaks it down into concrete milestones, spread out over three to six months.
At launch, Autopilot supports the goal of establishing positive cashflow through spending less than you earn. To achieve that goal, Cleo generates a Roadmap shaped by your income, fixed expenses, and typical spending patterns. If any of those inputs shifts—for example, your income changes, or you want to adjust your timeline—your Roadmap recalculates automatically.
Throughout 2026, we’ll expand Autopilot to support additional goals, like building an emergency fund, paying down debt, or saving for a major purchase. Setting up a goal is as simple as chatting with Cleo; she’ll ask follow-up questions to understand your situation and build a plan accordingly. Each new goal can use the same underlying planning framework, meaning that Cleo can scale to new objectives without needing to treat each addition as a separate feature.
Daily Plan
Whereas your Roadmap sets the destination, your Daily Plan keeps you on course. Each day, Cleo generates a snapshot of where you stand and what’s ahead. That includes a summary of recent transaction activity, a safe-to-spend amount for that month, and specific steps to correct course if you’re drifting off track.
These figures are dynamically generated for each individual, factoring in predicted expenses, upcoming income, and progress toward Roadmap milestones. When something unexpected happens, your plan reflects it immediately, rather than waiting for a new month or budget cycle. Because Cleo’s underlying models are continuously learning and adapting, her guidance is always up to date and personalized.
Actions
Actions enable Cleo to move from advice to execution, always with your consent. Each action ties directly to your Roadmap goal and is surfaced through the Daily Plan, so you can see why Cleo is recommending it and what outcome it supports. Once you confirm the action, Cleo handles orchestration behind the scenes and updates the plan based on the result.
At launch, Cleo can move money into savings, help prevent overdrafts with cash advances, and set spending limits at specific merchants. And throughout 2026, we’ll add actions that expand what she can do, centering around interventions tied to new Roadmap goals and quality-of-life improvements like negotiating bills and finding cheaper insurance. Every action is evaluated based on whether it measurably improves progress toward your plan, so Cleo is always learning what works best for you.
Technical foundations
Given what Cleo knows about a user’s finances, Autopilot determines what should happen next, then helps the user carry that decision through to execution. It’s built on four core capabilities: a multi-agent architecture, an advanced transaction enrichment layer, adaptive planning models, and continuous learning from observed outcomes.
Agent architecture: The right tool for each task
Financial guidance naturally decomposes into distinct tasks, such as transaction understanding, long-term planning, daily prioritization, and action execution. Each has different latency, reliability, and context requirements.
Cleo addresses this by routing work to specialized subagents scoped to a specific function. Agents operate within constrained contexts (e.g., reasoning or generating insights) and exchange structured outputs, rather than freeform text.
This keeps responsibilities explicit and reduces interference across unrelated tasks. And new actions can be added without retraining the system or layering on prompt complexity elsewhere. That modularity lets us expand Autopilot’s capabilities while keeping its behavior consistent.
Transaction enrichment: Turning raw data into real understanding
An AI financial assistant can accurately interpret users’ financial behavior only if it understands what their transactions actually mean. To that end, we built a transaction enrichment pipeline using large language models trained on high-quality labeled data, which enables Cleo to classify transactions with more nuance and accuracy.
Miscategorizing merchants and misclassifying transactions obscures user intent and makes forward-looking analysis unreliable. To address this, we trained Cleo to better understand real-world income and expenses, including recognizing recurring transactions, improving income stream identification, and separating essential from discretionary spending.
That means Cleo doesn’t just know that a transaction is a bill; she knows it’s an electricity payment, categorized under utilities, and that it’s a recurring essential expense. And downstream components in Autopilot consume this enriched representation, rather than raw transaction data. Because Cleo knows the context of this essential expense, she can use it to engage with users when it’s most relevant—for example, to help calculate how much they can safely spend without missing their bills, or to remind them that a payment’s coming up before it hits their account.
Adaptive planning: Guidance that updates when your finances do
Roadmap and Daily Plan are generated from planning models that update continuously as new information arrives. The system accounts for predicted expenses, income timing, and goal progress, recalculating as conditions change.
For example, say a user whose goal is positive cashflow incurs an unexpected medical bill midway through the month. Autopilot classifies the transaction as a non-recurring essential expense, updates their projected end-of-month balance, and recalculates what’s remaining in their budget.
Their Daily Plan then adjusts with a new safe-to-spend figure and corrective steps to stay on track. And because Autopilot updates in real time to reflect the user’s financial state, they don’t need to manually intervene.
Continuous learning: A system that improves through use
Finally, Autopilot improves the more it’s used. The system interprets how users respond to Cleo’s guidance and whether recommended actions lead to measurable progress, then feeds that information back into future planning.
At the system level, we use aggregated user outcomes to refine prioritization, planning heuristics, and action selection across the product. This ensures that improvements to Cleo are grounded in real financial results, like goal adherence and successful execution, rather than just optimizing for engagement metrics.
At the individual level, we’re also planning to teach Cleo to adapt to patterns in how a specific user engages with her. If someone consistently acts on a certain type of recommendation, and that decision helps them make progress against their Roadmap goal, Cleo should learn that that’s a useful lever for them and tailor her guidance accordingly.
Access
Autopilot is rolling out via waitlist to select users. To get access as it becomes available, sign up for the waitlist. We’ll continue expanding Autopilot access over the course of the next months as we add new functionalities and geographies.
What’s next
Autopilot marks a shift in how Cleo operates, from interpreting past financial decisions to carrying out new ones. Cleo can now reason over a user’s financial state, understand their big-picture goals, and give them a personalized path to achieve them.
What we’re launching today is the foundation for a system that can manage increasingly complex financial objectives over time. Throughout 2026, we’ll expand the range of Roadmap outcomes Autopilot can support to serve users with a wide range of financial goals. In parallel, we’ll introduce new actions that let Cleo intervene more directly and effectively, while keeping users in control of what Cleo can do on their behalf.
We’re looking forward to seeing how Autopilot’s deeper financial intelligence helps people change their relationships with money. And we’re excited to keep building genuinely autonomous financial AI that adapts continuously and consistently improves users’ lives.
