Skip to main content
Deep Dive June 9, 2026 · 7 min read

How Profit Forecast Works: The Math Behind the Wave

AxonRow's profit forecast isn't a straight line guess. It's built on a moving-average baseline, weekly growth trend, and day-of-week seasonality — here's how each piece works and why it matters for your planning.

Why Forecast Profit, Not Revenue

Most sellers track revenue forecasts. Revenue is easy to project — it goes up and to the right (hopefully). But revenue tells you nothing about what you keep. A $50K revenue week with 40% hidden fees and rising ad costs could leave you with $5K or $15K in actual profit — and those two outcomes require completely different inventory and cash flow decisions.

AxonRow’s Profit Forecast predicts your net contribution profit — the number that hits your bank after TikTok takes its 15+ fee types, after COGS, after ad spend. This is the number that determines whether you can afford to reorder inventory next week.

The Four Engines Behind the Curve

The forecast line you see is not a simple trend extrapolation. It is produced by four independent mathematical components working in concert.

1. Time Series Decomposition

Every day’s profit is the result of multiple forces acting simultaneously. The system decomposes your historical profit into independent components:

Daily Profit = Level × Trend × Seasonality × Noise

Each maps to something concrete in your dashboard:

  • Level — your baseline daily profit stripped of all fluctuation. This is your “AVG Daily Profit (30D)” number.
  • Trend — the directional momentum. If your Weekly Growth Rate shows +8.3%, the model compounds this forward.
  • Seasonality — the predictable weekly rhythm. Your Tuesday spike and Thursday dip are not random — they repeat.
  • Noise — the unpredictable remainder (sudden refund waves, viral spikes). The model acknowledges this exists but does not try to predict it.

2. The 7-Day Moving Average Baseline

To weave these components into a continuous forecast curve, the algorithm anchors on a 7-day moving average of your recent settled daily profit.

The key insight: a one-week baseline smooths daily noise while staying responsive. A simple 7-day moving average weights the last seven days equally, so a single anomalous day (a refund wave, a viral spike) moves the baseline by at most one-seventh. Meanwhile, the weekly window rolls forward every day — if your store broke through in the last week, the baseline catches that momentum shift within seven days, not 30.

Three components run in parallel on top of that baseline:

  • One measures where your profit is right now (the moving average level)
  • One measures where it’s going (the week-over-week growth rate)
  • One measures the weekly rhythm (day-of-week seasonality)

3. Day-of-Week Seasonality

This is why the forecast line waves up and down instead of drawing a boring straight line.

TikTok Shop buyer behavior follows strong weekly patterns. Your “Day-of-Week Profit Pattern” chart shows the multiplier for each day relative to the daily average:

  • A Tuesday multiplier of 1.15 means Tuesdays typically generate 15% more profit than average
  • A Thursday multiplier of 0.87 means Thursdays typically run 13% below average

When the model projects forward, it applies these multipliers to each future day. The result: if the trend says your baseline is growing at 8.3%/week, and Tuesday has a 1.15x seasonal factor, the forecast for next Tuesday reflects both the upward trend AND the Tuesday boost.

This is why experienced sellers see the forecast and immediately recognize their own store’s rhythm in it.

4. Confidence Scoring

Not all forecasts are created equal. The system evaluates its own reliability based on the statistical properties of your data:

What drives confidence up:

  • 28+ consecutive days of order data (minimum 3-4 complete weekly cycles)
  • Consistent seasonal patterns (low variance in day-of-week coefficients)
  • Stable growth trajectory (trend not whipsawing)

What triggers a confidence downgrade:

  • Fewer than 21 days of history — the model cannot reliably separate signal from noise
  • High residual variance — too much unexplained randomness
  • Structural breaks — a sudden business change (new product launch, price change) that invalidates historical patterns

When you see CONFIDENCE: low (Insufficient history), the algorithm is being honest: it has produced a forecast, but the sample size is below the minimum needed for the seasonal coefficients to be statistically significant. Use it as a directional estimate, not a planning certainty.

How to Use This for Decisions

The forecast is not a crystal ball. It is a structured extrapolation of your current trajectory and rhythm. Here is where it adds value:

Cash flow planning: If the 14-day forecast shows $24K in cumulative profit but you have a $20K inventory reorder due in 10 days, you have tight but manageable timing. If it shows $15K, you need to accelerate collections or defer the order.

Ad budget calibration: If Tuesday and Wednesday are your high-profit days (and the model confirms this pattern is stable), front-loading ad spend on Monday/Tuesday to capture that natural demand wave compounds the effect.

Anomaly detection: When actual profit diverges significantly from the forecast, that is a signal worth investigating. Did a fee increase? Did a top SKU go out of stock? Did a competitor undercut your price? The forecast serves as an expected baseline against which reality is measured.

The Honest Limitations

  • The model assumes the near future resembles the recent past. A Black Friday, a TikTok algorithm change, or a competitor price war will not be predicted.
  • Seasonal patterns require at least 3 weeks of data to be reliable. New stores should treat the forecast as directional only.
  • The model does not account for planned actions (upcoming product launches, scheduled promotions). It sees what has happened, not what you intend to do.

What Makes This Different from a Spreadsheet

You could build a projection in Google Sheets with a simple growth rate extrapolation. What you cannot easily replicate:

  1. Automatic seasonal decomposition — separating trend from weekly rhythm requires iterative aggregation, not a formula drag
  2. A rolling baseline — the 7-day moving average recalibrates daily, so a recent momentum shift shows up within a week, not after a 30-day lag
  3. Self-aware confidence — a spreadsheet will never tell you when its own output is unreliable
  4. Continuous recalibration — every new day of settled data refines the baseline, growth rate, and day-of-week coefficients automatically

The forecast updates daily as new settlement data flows in. No manual refresh needed.