toolbox
Home / AI, Data & Technology / Attribution Model Processing Capacity Estimator

Attribution Model Processing Capacity Estimator

The Attribution Model Processing Capacity Estimator converts parallel processing resources, per-worker throughput, operating time, and expected utilization into an effective record capacity. It is designed for planning attribution runs, backfills, or recurring pipelines where nominal throughput overstates what the system can usually sustain. The result separates theoretical capacity from utilization-adjusted capacity. This helps data teams test whether a planned processing window can accommodate the expected event or conversion volume and quantify the capacity lost to idle time, orchestration overhead, retries, or other nonproductive periods represented by the utilization input.

Processing assumptions

h
%
Result
—
Effective records per period
Theoretical capacity—
Effective records per hour—
Utilization applied—
Capacity lost to utilization—

1. Enter parallel workers
Use the number of workers, partitions, or equivalent processing units that can operate at the same time.

2. Set worker throughput
Enter the sustained records processed by one worker per hour under the planned workload.

3. Define the processing window
Enter the number of operating hours available for the run.

4. Apply utilization
Use a utilization percentage below 100% when setup, retries, waiting, or other overhead reduces usable capacity.

5. Compare capacities
Review effective capacity against your workload, and use theoretical capacity to understand the utilization gap.

Theoretical capacity = workers × rate per worker × hoursEffective capacity = theoretical capacity × utilization

Where:

• workers = concurrent processing units
• rate per worker = records processed by one worker each hour
• hours = operating duration
• utilization = productive share of theoretical capacity expressed as a decimal

Assumptions: Worker throughput is treated as constant over the selected period and scales linearly with the number of workers. Bottlenecks such as shared storage, rate limits, and skew should be reflected in the rate or utilization assumption.

What the result means

Effective capacity is the estimated number of attribution records the configured resources can complete during the selected window after utilization losses.

Real pipelines can stop scaling linearly when shared services saturate, so validate the per-worker rate at a concurrency level close to the planned deployment.

Given: 12 workers, 85,000 records per worker per hour, a 6-hour window, and 82% utilization.

Calculation: Theoretical capacity = 12 × 85,000 × 6 = 6,120,000 records. Effective capacity = 6,120,000 × 0.82 = 5,018,400 records.

Result: 5,018,400 effective records per 6-hour period.

Interpretation: The configuration has about 1.10 million records of nominal capacity that is not expected to be usable because of the 82% utilization assumption.

What should I use for a “worker”?

Use the smallest parallel unit for which you have a stable throughput measurement, such as a process, executor, partition worker, or service instance.

Should the rate be peak or sustained throughput?

Use a sustained rate measured under representative load. Peak benchmarks can materially overstate practical capacity for long attribution jobs.

How do retries affect the estimate?

If retries consume processing time, reduce the effective rate or utilization so their cost is reflected. Do not count retried records as extra completed output unless they are distinct required records.

Can I use minutes instead of hours?

The input is in hours. Convert minutes to hours before entry, such as 30 minutes = 0.5 hours.

How does this differ from a sample-size estimator?

Capacity estimates operational throughput. Sample-size calculations determine how many observations are needed for a statistical objective, regardless of how quickly those observations can be processed.