How long should a Media Forecaster plan take?

The short version

Plan type	What Customer Success should normally expect
Forecast	Usually under 1 minute. Allow up to 3 minutes before treating it as unusually slow.
Typical Optimization	Usually 5–15 minutes for plans with roughly 10–50 campaigns.
Very large Optimization	Roughly 25–60 minutes has been observed for plans with about 500–850 campaigns.

These are operating expectations based on production testing, not guaranteed completion times. A plan's timer includes waiting for available processing capacity as well as loading the models, preparing the plan, finding an allocation, checking the result, and saving it.

Why Forecast is faster than Optimization

Optimization works at the campaign level. More campaigns generally mean more curves and allocations to prepare and return.
Forecast works at the tactic level. Campaigns remain inside the trained model, while the plan allocates across a much smaller number of tactic curves.

In production tests, 18 Forecast plans had a median runtime of about 11 seconds and all completed within 25 seconds. Large-client examples also completed quickly: HexClad used 13 tactic curves across a 92-day period in 24 seconds, while OUAI used 15 curves and completed in 32 seconds.

What affects runtime

Plan mode

Mode is the strongest visible difference. Forecast plans use compact tactic-level planning curves; Optimization plans construct and return campaign-level results.

Number of campaigns

Campaign count matters most for Optimization. In validation:

Plans with 11–50 campaigns typically completed in 5–15 minutes.
Plans with 541–852 campaigns completed in about 25–58 minutes.

Runtime does not increase in a perfectly straight line. The selected company and models, cached data, and available processing capacity also matter.

Number of tactics

Tactic count is the more useful visible measure for Forecast. Tested plans with 2 to 15 tactic curves completed in under a minute. The campaigns behind those tactics still affect model loading, but they are not individually optimized in Forecast mode.

Number of models

Selecting more models can increase the amount of data that must be loaded and combined. In the Optimization test matrix, plans with three or more models had a median runtime of about 8 minutes, compared with roughly 6–7 minutes for one or two models. This is a tendency, not a formula.

Planning period and daily versus weekly results

Longer periods produce more output, but the tested 28-day and 84-day Optimization plans had similar typical runtimes. A Q4 Forecast is therefore not expected to take three times as long as a one-month Forecast simply because it contains three times as many dates.

Queue and warm-up time

The timer begins when the plan is submitted. It may include time waiting for an available worker or for the service to scale up and load data. This is why two otherwise similar plans can finish at different times.

Fixed allocations and risk tolerance

These settings change the plan's allowed solution but were not a major runtime driver in the tested Forecast cases. They are more likely to produce a clear feasibility message than to make a valid plan run indefinitely.

When should Customer Success investigate?

Elapsed time	Recommended action
Forecast: under 3 minutes	Let it continue. This is within the customer-facing allowance, although most tested runs were much faster.
Forecast: 3–10 minutes	Refresh the plan once to check for a completed result. If it still shows Running, note the plan link and monitor briefly.
Forecast: over 10 minutes	Investigate or escalate. This is not expected from tactic count or a Q4 date range alone.
Typical Optimization: under 15 minutes	Let it continue.
Typical Optimization: 15–30 minutes	It may reflect queueing, service warm-up, or greater model complexity. Refresh once and continue monitoring.
Optimization under 50 campaigns: over 30 minutes	Investigate or escalate.
Optimization with 500 or more campaigns: 25–60 minutes	This is within the range observed in large-client validation.
Any job over 60 minutes	Escalate unless it is a known very large plan already making progress.
Any job approaching 90 minutes	Escalate with the plan link and run details; do not assume campaign count alone explains it.

Refreshing the page should check the same job's status; it should not require creating a duplicate plan.

Suggested customer language

Normal Optimization wait

“This plan is still within the normal processing window. Optimization works across the selected campaigns and usually takes several minutes. You can leave the page and return to the same plan.”

Normal Forecast wait

“Forecast usually completes within a minute, although we allow a few minutes for queueing and model loading. If it passes that window, we’ll check the specific run.”

Escalating a slow job

“This plan is taking longer than we would expect for its size. We’ll investigate the run using the existing plan link; you do not need to submit repeated copies.”

What to include when escalating

Organization and plan link
Optimization or Forecast mode
Submission time and current elapsed time
Approximate campaign and tactic counts shown in the plan header
Number of selected models
Planning-period length and daily or weekly cadence
Whether the page was refreshed and remained on the same running job

Notes on the evidence

The primary runtime matrix included 401 completed production tests: 383 Optimization plans and 18 Forecast plans. Its measured time is submission-to-result wall time, so it includes queueing and processing. An older 500-run stress battery intentionally submitted enough work to create a backlog; its long elapsed times describe batch congestion, not how long one ordinary customer plan should take.

The mode definitions and intended planning experience align with PRD: New Forecast & Optimization Tool.