Bottom-up forecasting · ZopDev University

Outcome

By the end of this lesson, you will be able to collect team-level forecasts via a structured template, identify the systematic biases that creep into bottom-up numbers, and aggregate them into a defensible org-level total.


Tier	Engineer
JTBD	”Get team leads to forecast their own cost so the org-level number reflects ground truth, not a leadership wish.”
Personas	FinOps Lead · Engineering Leader · Team Lead
Prerequisites	M4.6.L1: Top-down forecasting
Time	9 minutes
Bloom verb	Collect (Apply), Identify (Analyze), Aggregate (Apply)

1. Concept

Bottom-up runs the other way. Each team says what it expects to need, and those are added up into the company total.

What you gain is detail, and the fact that each number has somebody’s name on it.

What you lose is realism, in a direction that is entirely predictable. Teams are optimistic about the savings they are going to make and forget the costs they do not look at. It is not dishonesty; it is that everybody forecasts the plan rather than the outcome.

BOTTOM-UP PROCESS:
  1. Each team's lead receives a forecast template
  2. They forecast resource by resource (or service by service)
  3. They consider planned events (launches, migrations, growth)
  4. They submit to FinOps
  5. FinOps aggregates; the sum is the bottom-up forecast
  6. The result feeds reconciliation with top-down (L3)

The process is 6-8 weeks for a quarterly forecast. Teams typically need 1-2 weeks to draft; FinOps needs 1-2 weeks to aggregate and review; the reconciliation takes another 1-2 weeks.

When bottom-up wins

USE CASE                                  GOOD FIT
──────────────────────────────────────────────────────────────────
Major projects with clear scope            Yes: teams know their plan
Predictable steady-state workloads         Yes: easy to forecast
Quarterly budget planning                  Yes: standard cadence
Acquired company integration                Yes: different cost shape
                                          than historical
Multi-team migrations                       Yes: each team forecasts
                                          their leg
Team-level accountability building          Yes: forcing function

When bottom-up misses

USE CASE                                  POOR FIT
──────────────────────────────────────────────────────────────────
Sudden unexpected growth                   Better caught by extrapolation
                                          (top-down)
Cross-team impacts                          Each team forecasts independently;
                                          interdependencies hidden
Marketing-driven cost spikes                Marketing isn't on the team list;
                                          falls between cracks
Cloud-rate changes                          External; no team has visibility
Org-wide rationalization                    Top-down captures aggregate

Each team’s contribution

A typical bottom-up template per team:

TEAM A (engineering-platform):
  Current run rate:  $50K/mo
  Planned changes:
    2 new K8s clusters in Q2 (+$8K each ongoing) = +$16K
    Decommission old monitoring stack (-$3K)
    Net: +$13K/mo
  Forecast Q2: $63K/mo

TEAM B (engineering-product):
  Current: $40K/mo
  Planned:
    Feature launch (+5K MAU expected) → +$3K compute
    Migrate API gateway (one-time +$2K Q1, neutral after)
  Forecast Q2: $43K/mo

TEAM C (engineering-data):
  Current: $30K/mo
  Planned:
    Data pipeline migration (+$10K Q2 one-time, baseline unchanged)
    New ML training cluster (+$5K/mo ongoing starting Q2)
  Forecast Q2: $35K/mo + $10K one-time = effective $45K Q2
                ($35K Q3+)

ORG TOTAL (bottom-up): $63K + $43K + $45K = $151K/mo Q2
                       (drops to $141K Q3 after one-time costs)

Reconciling with top-down

The bottom-up total should approximately equal the top-down total. Where they diverge, investigation surfaces real assumptions:

TOP-DOWN sees: $145K/mo (8% growth from $134K baseline)
BOTTOM-UP sees: $151K/mo (sum of teams)

DIFFERENCE: $6K/mo (4%: within tolerance)

RECONCILIATION:
  Talk to teams about specific assumptions
  Cross-check planned events
  Resolve to a single committed number

A 4% difference is acceptable for a quarterly forecast; 15%+ difference signals significant assumption gaps worth investigating.

Common systematic biases

Teams predictably forecast in specific directions:

BIAS                                        DIRECTION    SIZE TYPICAL
──────────────────────────────────────────────────────────────────
Underestimate cost growth                   under         5-15%
"Our infra is efficient; we won't grow"

Overestimate efficiency gains                under         5-10%
"We're going to right-size everything"

Forget hidden costs                          under         3-8%
"Just compute" misses egress, support,
monitoring, backup

Forget one-time events                       under         varies
"Just our steady state" misses migrations

Optimism about feature ROI                   under         varies
"New feature will be efficient per user"

Sometimes: OVER-forecasting                 over          5-20%
when defending budget headroom

The net bias is usually under-forecasting by 5-15%. Apply a correction factor based on historical accuracy (more in L4).

Improving bottom-up accuracy

IMPROVEMENT                                IMPACT
──────────────────────────────────────────────────────────────────
Include hidden costs in template            +5% accuracy
(egress, monitoring, backup explicit)

Require one-time events as line items       +3% accuracy

Calibrate against last quarter's actuals    +5% accuracy per cycle
(team feedback loop)

Force ranges, not point estimates           +3% accuracy
(team commits to range, not single number)

Cross-team review (peer challenge)          +5% accuracy
(team A reviews team B's forecast)

Each improvement compounds. After 3-4 quarterly cycles, bottom-up accuracy can reach 92-95%.

The forecast template

A minimum-viable template per team:

TEAM:    __________
PERIOD:   __________ (Q__ 2026)
PREPARED BY:    __________
DATE:    __________

CURRENT RUN RATE:    $______ /mo

PLANNED CHANGES:
  Resource/service           Driver               Cost impact
  __________                  __________            $______
  __________                  __________            $______
  __________                  __________            $______

ONE-TIME COSTS this period:
  __________                  $______
  __________                  $______

HIDDEN COSTS check (have you included?):
  □ Data egress
  □ Monitoring + logging
  □ Backup storage
  □ Support overhead
  □ DR / failover capacity

FORECAST (per month for the period):    $______
ONE-TIME ADJUSTMENT:    $______
TOTAL FORECAST for period:    $______

ASSUMPTIONS (key):
  __________________________________________________________

How ZopNight supports bottom-up

ZopNight pre-fills the template per team using current spend data + recent trends. The team lead reviews, adjusts, adds planned events. Submitted templates aggregate automatically. FinOps reviews the aggregate before reconciliation.

ZOPNIGHT BOTTOM-UP TEMPLATE:
  Per team: pre-filled run rate + 30-day trend
  Manual: planned changes line items
  Manual: hidden cost checklist
  Auto: aggregation across teams
  Auto: variance vs top-down forecast

2. Demo

A quarterly bottom-up cycle:

TIMELINE:    Q2 forecast preparation (Feb 1 - Mar 31)

FEB 1: week -8:
  FinOps sends forecast templates to team leads
  Pre-filled with current run rate + 30-day trend
  Template includes hidden-cost checklist

FEB 8: week -7:
  Reminder; templates due Feb 15
  Office hours offered for team leads

FEB 15: week -6:
  Forecasts collected:
    Platform team:  $63K/mo (with new cluster planning)
    Product team:   $43K/mo (feature launch)
    Data team:       $45K/mo (one-time migration)
    Shared services: $15K/mo
    Total: $166K/mo Q2

FEB 22: week -5:
  FinOps review:
    Hidden cost check passed (all teams included egress)
    Data team's one-time migration verified
    Platform's new cluster cost cross-checked against
      RIs available

FEB 29: week -4:
  Top-down comparison: $158K/mo (8% growth from baseline)
  Difference: $8K/mo (5%: within tolerance)

MAR 7: week -3:
  Reconciliation meeting with team leads:
    Platform forecast revised slightly (cluster count refined to 2.5)
    Data forecast confirmed (migration is real)
    Product unchanged
    Resolved to $164K/mo Q2

MAR 14: week -2:
  Forecast committed: $164K/mo Q2, ±10% band
  Communicated to leadership + finance

APR 1: Q2 begins:
  Forecast in dashboard
  Monthly re-forecast scheduled

Variance check at quarter end

JUNE 30: Q2 ends:
  Forecast Q2: $164K/mo
  Actual Q2:   $171K/mo
  Variance:    +4.3%

  Within ±10% band ✓

  Causes of variance:
    Platform: cluster usage higher than planned (+$3K)
    Product: feature launch generated more traffic (+$4K)
    Data: migration as forecast

  Lessons for Q3 forecast:
    Bump platform forecast 3-4%
    Add traffic-driven adjustment to product feature forecasts

The variance feedback improves the next forecast.

3. Hands-on (5 min)

Draft a bottom-up forecast for your team:

TEAM:    __________
PERIOD:   __________ (next quarter)

CURRENT RUN RATE:    $______ /mo

PLANNED CHANGES (with dollar impact):
  __________  +/- $______
  __________  +/- $______
  __________  +/- $______

ONE-TIME COSTS this quarter:
  __________  $______
  __________  $______

HIDDEN COSTS check:
  □ Data egress
  □ Monitoring + logging
  □ Backup storage
  □ Support overhead

FORECAST per month:    $______
PLUS one-time:    $______

KEY ASSUMPTIONS:
  __________________________________________________________

CONFIDENCE in this forecast:   high / medium / low
WHY:    __________

Compare with the FinOps function’s top-down number for your team. Identify the largest difference and investigate.

Ask for it instead. The same task you just did in the console, asked in one sentence.

BEFORE  A ZopNight account with one cloud connected. Six months of billing history, or the forecast is a guess wearing a chart.
ASK     "Give me six months of cost trend, by team and by tag, so I can build a bottom-up forecast."
CHECK   the series length before you model anything.

Tools behind it: get_cost_trends (read, Cost), get_team_cost_trends (read, Cost), get_tag_cost_trends (read, Cost). The full catalogue is at zop.dev/learn/mcp-tools.

4. Knowledge check

Q1

Bottom-up vs top-down forecast:

A. Same approach
B. Bottom-up is detailed (team-level, line items); top-down is aggregate (org-level, growth rate)
C. Bottom-up is simply a rollup of the top-down figure across all of the teams involved
D. Only use one

Show answer

Correct: B. Use both; reconciliation between them surfaces real assumption differences and improves overall accuracy. Hybrid (L3) is the typical mature practice. Hybrid is best. Each method catches things the other misses.

Q2

Each team’s bottom-up forecast typically:

A. Underestimates cost growth, overestimates own efficiency gains, forgets hidden costs (egress, monitoring, backup)
B. Overestimates growth
C. Accurately reflects truth
D. Matches the top-down figure closely, since both are derived from exactly the same underlying cost data

Show answer

Correct: A. Net bias is usually 5-15% under-forecasting. Apply a correction factor based on historical accuracy. Bias toward optimism in self-forecasting. Correct via calibration.

Q3

Bottom-up forecast diverges from top-down by 15%:

A. Pick whichever is lower
B. Average them
C. Use the top-down (it’s leadership-driven)
D. Reconcile

Show answer

Correct: D. Investigate which is closer to reality Big divergence signals real assumption differences worth understanding; could be missing events (top-down higher), forgotten growth (bottom-up lower), or genuine new info. Use the discrepancy as an investigation prompt. Reconcile to find truth. The reconciliation is the value of running both methods.

5. Apply

Run quarterly bottom-up + top-down reconciliation. Use the template in this lesson; pre-fill via ZopNight’s forecast tool. Track team-level accuracy quarterly; calibrate biases.

Glossary terms touched

Bottom-up forecast · Forecast template · Systematic bias · Hidden costs · Calibration

Outcome

1. Concept

When bottom-up wins

When bottom-up misses

Each team’s contribution

Reconciling with top-down

Common systematic biases

Improving bottom-up accuracy

The forecast template

How ZopNight supports bottom-up

2. Demo

Variance check at quarter end

3. Hands-on (5 min)

4. Knowledge check

Q1

Q2

Q3

5. Apply

Related lessons

Glossary terms touched

Start with the bill.