Forecasting Marketing Performance: Simple Models That Beat Gut Feel

Why Marketing Forecasts Are Terrible (And How to Fix That)

Most marketing forecasts are either gut feel dressed up in a spreadsheet ("we'll grow 30% next year") or so complex that nobody believes or uses them. Both are useless. Simple, documented, assumption-explicit models outperform both.

Our marketing analytics team builds forecasting models that are simple enough to understand, rigorous enough to trust, and updated regularly to remain relevant.

Model 1: Trend Extrapolation (The Starting Point)

The simplest approach: project forward based on recent trend.

1. Pull last 12 months of the metric you're forecasting (monthly revenue, leads, sessions)
2. Calculate month-over-month growth rate for each month
3. Average the last 3–6 months of growth rate
4. Apply that rate to forecast future months

Example: If your MoM lead growth has averaged 4.2% for the last 4 months, forecast next month as current month × 1.042.

Limitation: Assumes past trend continues. Doesn't account for seasonality or planned initiatives. But it's honest and fast — a useful sanity check against more sophisticated models.

Model 2: Seasonality-Adjusted Forecast

Better for businesses with clear seasonal patterns.

1. Pull 2+ years of monthly data
2. Calculate each month's "seasonal index" = that month's value / annual average
3. Average seasonal indices across years to get stable seasonality pattern
4. Apply trend rate to deseasonalized baseline, then apply seasonal index to re-seasonalize

In BigQuery, this can be done with SQL window functions across historical data. In a spreadsheet, it requires a few calculated columns but is manageable.

The result: your January forecast accounts for January historically being 30% below average for your category, while your November forecast reflects holiday uplift.

Model 3: Driver-Based Forecasting

The most useful model for marketing planning. Define the drivers that produce outcomes:

Monthly leads =
  (Paid traffic × Paid conversion rate)
+ (Organic traffic × Organic conversion rate)
+ (Email sends × Email conversion rate)

Forecast each driver separately with assumptions, then compound them. When the overall forecast is off, you can trace it to the specific driver that deviated.

This model also becomes a planning tool: "If we want 20% more leads, which driver(s) do we grow, and what does that cost?"

Model 4: Bottom-Up Budget Forecast

Work from planned spend to expected output:

• Planned Google Ads spend: £20,000/month → at current £28 CPA → 714 leads
• Planned Meta spend: £8,000/month → at current £45 CPA → 178 leads
• Organic (assumed flat): 150 leads/month
• Total leads forecast: 1,042/month

Adjust for expected efficiency changes (seasonality, audience saturation, creative fatigue).

Communicating Forecasts With Ranges

Single-point forecasts ("we'll get 1,000 leads in March") create false precision. Present ranges:

• Conservative: Assumes current efficiency degrades 15%
• Base: Current efficiency maintained
• Optimistic: Planned initiatives deliver expected improvement

Ranges are more honest and set appropriate expectations for variance in actual results.

Connect your historical GA4 and ad platform data to BigQuery to build forecasting models on real data. Our analytics team builds these models. Contact us for marketing forecasting model development.