Forecasting unit cost · ZopDev University

Outcome

By the end of this lesson, you will be able to project unit cost trends using three methods (extrapolation, driver-based, scenarios), recognize when to worry about trend reversal, and communicate forecast uncertainty honestly.


Prerequisites	M5.3.L1-L3 (denominator, numerator, dashboard) · M5.6 (forecasting general)
Time	9 minutes

1. Concept

Forecasting unit cost means saying where the cost-per-unit line is heading.

It feeds three things: the budget, how much capacity you buy, and what leadership is told. Three ways to do it, each right in different circumstances:

A. EXTRAPOLATION:
   Linear or exponential fit to recent data
   "Cost-per-MAU is dropping 2% per month; project forward."

B. DRIVER-BASED:
   Forecast cost and denominator separately; combine
   "Cost will grow $13K next quarter due to new feature;
    MAU will grow 15%; cost-per-MAU = (cost + $13K) / (MAU × 1.15)"

C. SCENARIOS:
   Best/likely/worst case projections
   "Best: $9; Likely: $11; Worst: $13"

Method A is fastest; Method B is most accurate; Method C handles uncertainty.

Method A: Extrapolation

Last 12 months cost-per-paying-user:
  Month -12:  $13.50
  Month -6:   $12.20
  Month -1:   $11.83

TREND: -1.2% per month average ((11.83 / 13.50)^(1/11) - 1)

FORECAST month +3: $11.83 × (1 - 0.012)^3 = $11.41
FORECAST month +12: $11.83 × (1 - 0.012)^12 = $10.23

Simple, fast. Useful for "first sketch" forecasts.
WEAKNESS: doesn't account for upcoming changes (launches, optimizations).

Method B: Driver-based

SEPARATE forecasts for numerator and denominator:

Numerator (cost) forecast:
  Current: $142K/mo
  Planned launches: +$13K/mo (new product features)
  Optimization in flight: -$5K/mo (Q2 sprint)
  Forecast: $150K/mo by end of Q2

Denominator (paying users) forecast:
  Current: 12,000
  Sales pipeline: +1,500
  Churn rate: -2% per month → -240/mo (-720 over the quarter)
  Forecast: 12,000 + 1,500 - 720 = 12,780 by end of Q2

COMBINED forecast:
  Cost-per-paying-user = $150K / 12,780 = $11.74

COMPARED to extrapolation:
  Extrapolation: $11.41
  Driver-based: $11.74 (worse than the trend line, because it
    counts the new launches AND the churn the trend ignores)

Method C: Scenarios

BEST CASE:
  Cost optimization exceeds plan
  User growth meets plan
  cost-per-paying-user: $9.50

LIKELY CASE:
  Cost optimization meets plan
  User growth meets plan
  cost-per-paying-user: $11.74

WORST CASE:
  New feature consumes more compute than expected
  User growth misses by 30%
  cost-per-paying-user: $13.20

USED FOR:
  Budget planning (commit to LIKELY; reserve for WORST)
  Variance analysis (was the actual closer to BEST or WORST?)
  Leadership conversations (the range, not a single number)

When to worry

PATTERN                                    INVESTIGATE?
──────────────────────────────────────────────────────────────────
Trend reverses (decline → increase)         Yes: what changed?
Sudden spike (one-month jump)               Yes: anomaly or
                                              expected event?
Stagnant for 6+ months                      Maybe: efficiency
                                              plateau? Worth a
                                              re-examination
Continued decline                          Note: investigate
                                              what's working;
                                              document the pattern
Sudden drop                                 Yes: definition
                                              change? Denominator
                                              spike? Anomaly?

A trend reversal is the strongest signal. The narrative changes from “improving” to “worsening”; investigate immediately rather than waiting for the next monthly review.

Per-team forecast variation

Different teams have very different trajectories:

TEAM A (mature platform team):
  Cost-per-MAU stable or declining slowly
  Forecast: continued slow decline → $10 in 6 months
  Story: efficiency plateau; minor optimizations ongoing

TEAM B (new feature ramping):
  Cost-per-MAU rising as feature scales
  Forecast: $14 in 6 months
  Story: new product cost ramp-up; acceptable
  Expected to drop once user base grows to match infrastructure

TEAM C (legacy / sunset):
  Cost-per-MAU rising as fixed infra is amortized over shrinking user base
  Forecast: per-MAU rises further (fewer users sharing fixed cost)
  Story: end-of-life economics; consider deprecation timeline

TEAM D (M&A integration):
  Cost-per-MAU spiking during integration
  Forecast: returns to baseline 6-12 months post-integration
  Story: one-time migration cost; should normalize

Each team’s trajectory tells its own story. Aggregate forecasts mask this; per-team forecasts surface it.

Forecasting accuracy

Be honest about uncertainty:

HORIZON                              TYPICAL ACCURACY
──────────────────────────────────────────────────────────────────
Month-ahead                          ±5-10%
Quarter-ahead                        ±10-20%
Year-ahead                           ±20-30%
Multi-year                            ±40%+

Accuracy decreases with horizon. Communicating “Q4 will be $X ± 20%” is more honest than “Q4 will be $X”: and audience trust improves over time when uncertainty is communicated and outcomes fall within the band.

Forecast vs budget

FORECAST: what we think will happen based on current data
BUDGET:   what we have committed to spend

A forecast that exceeds budget is a signal to investigate;
it does not automatically mean the budget will be exceeded
(the team has time to adjust).

Where to build the forecast

Build the driver-based forecast in your planning sheet: one tab for the driver forecast (from the business plan), one for the cost model, one for the scenarios, and a band column for the range. ZopNight’s unit economics report adds a simple average-of-last-N projection when there is enough history; treat it as a sanity check against your driver-based number, not as the forecast.

For variance analysis, keep a log tab: each quarter, record the forecast you published and, a quarter later, the actual. The accuracy trend comes from that log.

2. Demo

A SaaS company’s Q4 unit cost forecast:

CURRENT (end of Q3):
  Cost: $142K/mo
  Paying users: 12,000
  Cost-per-paying-user: $11.83

PLANNED CHANGES Q4:
  Product launch in November:    +$20K/mo for new features
  EU expansion:                    +$5K/mo for regional infra
  Q4 optimization sprint:           -$10K/mo expected savings
  Sales growth target:              +2,400 paying users (20% growth)
  Churn rate:                       -2%/mo (assume continued; ~-6% over Q4)

DRIVER-BASED FORECAST Q4:
  Cost: $142K + $20K + $5K - $10K = $157K/mo by end of Q4
  Paying users: 12,000 × 1.14 (net of churn) = 13,680
  Cost-per-paying-user: $157K / 13,680 = $11.48

SCENARIOS:
  BEST: optimization saves $20K (double plan), user growth meets: $10.75
  LIKELY: optimization meets, user growth meets: $11.48
  WORST: optimization saves nothing, new users 30% below target
         ($167K / 12,960 users): $12.89

DECISION (leadership review):
  "Cost-per-paying-user continues to improve despite new launches.
  Forecast Q4: $11.50 (likely), range $10.75-$12.89.
  Acceptable trajectory; on track for $10/user target by end of Q1 2027."

VARIANCE CHECK against prior forecast:
  Q3 forecast made in Q2: $12.10
  Q3 actual: $11.83 (better than forecast)
  Variance: -2.2% (within ±10% band)
  Track record: forecast accuracy improving over time

3. Hands-on (5 min)

Forecast your unit cost for the next quarter:

CURRENT cost-per-unit:    $__________

METHOD A: Extrapolation:
  Recent trend: ____% per month
  Forecast next quarter: $__________

METHOD B: Driver-based:
  Numerator forecast: $__________
    + planned increases: __________
    - planned decreases: __________
  Denominator forecast: __________
    + planned growth: __________
    - churn: __________
  Cost-per-unit forecast: $__________

METHOD C: Scenarios:
  BEST: $__________ (assumes: __________)
  LIKELY: $__________
  WORST: $__________ (assumes: __________)

WHICH METHOD do you trust most for your forecast?    __________
WHY: __________________________________________________________

VARIANCE check against previous quarter's forecast:
  Previous forecast: $__________
  Actual: $__________
  Variance: ____%
  Accuracy improving? Yes / No

If you don’t have a previous forecast to check against, start producing forecasts now. Variance analysis becomes possible after 2-3 cycles.

4. Knowledge check

Q1

Driver-based forecasting (vs extrapolation):

A. Same accuracy
B. Worse
C. Only better for long horizons
D. More accurate

Show answer

Correct: D. Forecasts cost and the denominator separately, combining them. Accounts for planned events (launches, optimizations, user growth) that extrapolation cannot see. Extrapolation is fine for first sketches; driver-based is for committed forecasts. Driver-based is more accurate because it incorporates known future events.

Q2

A team’s cost-per-MAU rising as a new feature scales:

A. Always bad
B. Often acceptable
C. Proof the feature should be cut
D. Always good

Show answer

Correct: B. New features add cost before users grow to match. Monitor; if the pattern continues 6+ months without MAU growth catching up, investigate. Short-term cost-per-MAU rise during feature ramp-up is normal economics. Temporary increase during ramp-up is acceptable; sustained increase warrants investigation.

Q3

Forecast accuracy at year-ahead horizon:

A. ±20-30%
B. ±2%: same as month-ahead
C. ±10%, the same as quarterly
D. ±5%

Show answer

Correct: A. Communicate this band honestly. Long-horizon forecasts have high uncertainty (many unknowns: launches, growth, optimization, market). Audience trust improves when uncertainty is acknowledged and actual outcomes fall within the band; trust erodes when forecasts are presented as certainties.

5. Apply

Build your quarterly forecast in your planning sheet with driver-based inputs and scenarios for budget planning. Compare it with the cost-per-unit trend on ZopNight’s Unit Economics page and log forecast against actual each quarter.

Track variance: actual vs forecast for prior periods. Forecast accuracy is itself a metric: improving accuracy is a sign of mature unit economics practice.

Glossary terms touched

Forecast · Driver-based forecasting · Forecast accuracy · Confidence band

Outcome

1. Concept

Method A: Extrapolation

Method B: Driver-based

Method C: Scenarios

When to worry

Per-team forecast variation

Forecasting accuracy

Forecast vs budget

Where to build the forecast

2. Demo

3. Hands-on (5 min)

4. Knowledge check

Q1

Q2

Q3

5. Apply

Related lessons

Glossary terms touched

Start with the bill.