Estimating Token Usage

When planning a Brim project, one of the most common questions is:

“How many tokens will this project use?”

The short answer is: token usage is difficult to predict precisely.

However, you can create a practical estimate to support planning, budgeting, and optimization decisions.

Why Token Usage Is Difficult to Predict

Token usage in Brim depends on several interacting factors across your abstraction workflow, including:

Document length (how much text exists per patient)
Number of documents per patient
Total number of patients
Number of variables defined
Variable structure (including dependent variables and hierarchies)
Iteration cycles during optimization and validation

Because Brim supports iterative, human-in-the-loop abstraction, token usage naturally evolves as projects mature.

A Conservative Estimation Formula

To help teams plan early, Brim commonly uses the following conservative estimate:

Total Token Usage ≈
total_patients × characters_per_patient × number_of_variables_defined

What each component means

Total Patients

The number of patients included in your project.

Characters per Patient

The approximate total text volume in characters across all uploaded documents for a single patient.

Number of Variables Defined

The total number of variables generated during abstraction.

✅ The conservative estimate overestimates usage. For a recent group of 75 real-world projects:

All projects had usage equal to or less than the conservative estimate.
The conservative estimate overestimated usage by an average of 20x relative to real token usage.
Some projects were overestimated by as much as 200x.

Example Conservative Estimate

If your project includes:

500 patients
~25,000 characters per patient
40 defined variables

Your conservative estimate would be:

500 × 25,000 × 40
= 500,000,000 tokens

At $0.40/million tokens, a conservative estimate would be that this project will be under $300 in compute.

This provides a useful upper-bound planning estimate before running generation.

Factors That Increase Token Usage

You may see higher token usage when:

Variables reference large portions of source documents
Projects contain long clinical narratives
Variables are regenerated multiple times during iteration

How to Reduce Token Usage

Brim provides several tools to help manage and optimize token consumption:

Designing clear, focused variables
Using structured variable hierarchies
Iterating intentionally rather than regenerating broadly
Reviewing Variable Scorecard feedback before regeneration

For guidance, see:

Important Note

Token estimation should be treated as a planning tool, not a precise forecast.

The most accurate understanding of usage comes from:

Running small pilot generations
Monitoring token usage during early iterations
Refining variables before generation

Recommended Workflow

We recommend the following approach:

Estimate token usage conservatively using the formula above.
Set a per-project Token Limit using that estimate and what's acceptable for your project.
Run a small pilot cohort
Reviewing token usage results frequently
Optimize variables and hierarchy structure
Scale generation gradually

If you’d like help estimating token usage for a specific project, contact support@brimanalytics.com.