How Clarity Predicts Your Next Bill Amount

Your Netflix subscription is $15.49 every month. Easy to budget for. Your electric bill ranges from $85 to $210 depending on the season. Your cloud storage bill fluctuates with usage. These variable recurring charges are the hardest to plan around — and they're exactly what Clarity's TimesFM-powered predictions are designed for.

The Problem with Variable Recurring Charges

Fixed subscriptions are simple: the same amount, the same date, every month. But many recurring charges aren't fixed:

• Utility bills (electricity, gas, water) vary with usage and season
• Usage-based subscriptions (cloud storage, metered APIs, phone data overages) change monthly
• Insurance premiums that adjust quarterly or annually
• Grocery delivery subscriptions where the base fee is fixed but add-ons vary
• Fuel and commuting costs that follow seasonal price patterns

When Clarity detects these as recurring transactions, it shows them on your recurring bills calendar. But what amount should it display for next month? The average of all past charges? The most recent one? Both approaches can be significantly off.

How TimesFM Predicts the Next Charge

Clarity uses TimesFM 2.5, Google's time-series foundation model, to predict the next charge for each variable recurring transaction. The model receives the sequence of past charge amounts and produces a prediction for the next occurrence, including confidence intervals.

For example, if your electric bill over the past 8 months was:

• June: $195
• July: $210
• August: $205
• September: $165
• October: $130
• November: $110
• December: $125
• January: $140

A simple average gives $147.50. The most recent charge was $140. But TimesFM recognizes the seasonal pattern: bills are high in summer (AC), drop in fall, and start climbing again in winter (heating). For February, it might predict $155 (p50), with a range of $135 - $175 (p10-p90). It extrapolates the upward winter trend rather than anchoring to the average or the last value.

What Patterns the Model Detects

Even with limited data points (recurring charges produce one observation per billing cycle), TimesFM identifies several useful patterns:

• Seasonal cycles. Utility bills follow predictable seasonal curves. TimesFM detects the periodicity and phase of these cycles. Your electric bill in July should be predicted differently than your electric bill in October, even if recent values are similar.
• Trend direction. If a subscription has been gradually increasing ($49, $52, $55, $58), the model recognizes the upward trend and predicts accordingly. This catches slow price creep that you might not notice looking at individual charges.
• Level shifts. If your phone bill jumped from $85 to $120 three months ago (maybe you upgraded your plan), the model recognizes this as a permanent level change and predicts from the new baseline, not the historical average that includes the old plan.
• Stable charges. For charges that barely vary ($49.99, $49.99, $49.99, $52.49, $49.99), the model produces a tight prediction with very narrow confidence bands. It effectively confirms this is a near-fixed charge.

Confidence Ranges for Budgeting

The most practical aspect of the prediction is the confidence range. For each upcoming variable charge, Clarity shows:

• Expected amount (p50): The model's best estimate. Use this as your budget line item.
• Range (p10-p90): The amount will almost certainly fall within this range. Use the upper end (p90) if you want to budget conservatively.

For example, on the recurring bills calendar, your electric bill entry might show: ~$155 ($135 - $175). If you're building a monthly budget and want to be safe, allocate $175 for electricity. If you want the most likely number, use $155.

This is significantly more useful than showing "$147.50" (the average) or "$140" (last month), because it accounts for the direction your bills are trending and gives you a range to plan around.

Where Predictions Appear

TimesFM recurring predictions integrate into Clarity's existing recurring transaction features:

• Recurring bills calendar. Each upcoming variable charge shows the predicted amount instead of the last charged amount. The confidence range appears on hover or tap.
• Recurring sidebar. The sidebar summary of upcoming bills uses predicted totals. Your "bills due this week" total reflects ML predictions for variable charges and exact amounts for fixed ones.
• Budget integration. When a variable recurring charge is assigned to a budget category, the predicted amount contributes to the budget's projected spending. This gives you a more accurate picture of how much of each budget category is already spoken for.
• AI assistant. Ask "how much will my utilities cost next month?" and the assistant sums the TimesFM predictions for all recurring charges tagged as utilities, with the combined confidence range.

How Much History Does It Need?

Recurring charge predictions work with as few as 4 historical data points (4 billing cycles). With fewer than 4, Clarity falls back to showing the last charged amount. The predictions improve with more history:

• 4-6 data points: Basic trend detection. The model can tell if charges are going up or down but can't reliably detect seasonality.
• 7-12 data points: Seasonal pattern detection becomes possible. The model can distinguish a winter heating spike from a permanent increase.
• 13+ data points: Full annual cycle captured. The model has seen at least one complete year and can make strong seasonal predictions.

Catching Billing Surprises Before They Hit

One underappreciated benefit of recurring charge prediction is catching anomalies. If your predicted electric bill is $155 but you get charged $280, that's well outside the p90 bound. Clarity can flag this as unusual, prompting you to check whether it's a meter error, a rate change, or an actual spike in usage.

Similarly, if a "fixed" subscription that's been $29.99 for 18 months suddenly charges $39.99, the tight confidence bands around the predicted amount make it immediately obvious. You might have missed a price increase notification — the model won't miss it.

Variable recurring charges account for 15-30% of most households' monthly expenses. Getting accurate predictions for these charges — instead of guessing or using stale averages — means your budget reflects what you'll actually owe, not what you owed three months ago.