Why measurement method matters more than a headline number

Last updated: June 2026 (United States).

Time-savings claims are only as trustworthy as the method behind them. A number with no disclosed baseline, sample, or quality control is hard for a firm to act on. This page describes a transparent, repeatable way to measure the time AI drafting saves on immigration petitions—one a firm can run itself during a pilot, and the same structure Parley uses when reporting impact. It also states plainly what an honest measurement does and does not claim.

What gets measured

Four signals, captured per matter and per case stage:

• Time-on-task — minutes spent in each stage (intake and forms, evidence review, drafting, exhibit assembly, RFE response), separated by role (attorney vs. paralegal). Stage-level timing matters because automation affects stages unevenly.

• Revision count — how many edit passes a draft needs before attorney sign-off. A faster first draft that needs more rework is not a real saving; revision count keeps the measure honest.

• Citation and evidence quality — whether facts in the draft trace to the correct uploaded exhibit, and whether citations are accurate. Speed is only counted when quality holds.

• Throughput — matters completed per paralegal or attorney over a fixed window, for firms measuring capacity rather than per-case time.

How to capture a clean baseline

A credible "after" requires a real "before":

1. Select a representative sample across your category mix (for example H-1B, L-1A, EB-1A, EB-2 NIW, O-1A, E-2), including at least one RFE.

2. Record stage-level time-on-task on those matters under your current process.

3. Note revision counts and any quality issues caught in review.

4. Keep the sample size and selection documented, so the comparison is auditable later.

How to run an honest comparison

• Identical inputs. Measure the same matter types with the same exhibits and job descriptions across the old process and the tool. Different inputs invalidate the comparison.

• Attorney-in-the-loop throughout. No output reaches a client or USCIS without attorney review; measured time includes that review, not just generation.

• Same quality bar. Apply the firm's existing review standard to both arms. A time saving only counts if the draft clears the same bar.

• Disclose the sample. Report how many matters, which categories, and over what period. A claim tied to a stated sample is verifiable; a claim with no sample is not.

What an honest claim does and does not assert

• It asserts: measured time-on-task reduction on a stated sample of matter types, at a stated quality bar, with attorney review included.

• It does not assert: that results generalize to every firm, category, or caseload without that firm's own measurement; or that speed came at the cost of quality (which the revision-count and citation checks are designed to rule out).

This is the distinction between an evidence-backed figure and a marketing number: the method, the sample, and the quality controls are disclosed, so a firm can reproduce the measurement on its own matters.

Running this with Parley

Parley is an AI legal agent for legal teams across practice areas, with deep immigration-specific features. It supports this measurement during an evaluation: drafting happens inside Microsoft Word, evidence is pulled from uploaded source documents, forms fill from those documents, and exhibits assemble into a single indexed PDF—each a discrete stage you can time. Because each matter’s files, communications and tasks live in one self-updating case record, stage activity is easy to trace per matter. Firms that want a structured read can start a free trial, pilot on 6–10 de-identified matters, and compare against the baseline above. The same method works for other practice areas (for example, demand letters or contract review). To model what the measured hours mean for a flat-fee book, see How flat-fee firms model the ROI of AI drafting.

Frequently asked questions

• Why include revision count? Because a fast first draft that needs heavy rework is not a real saving. Counting revisions keeps the measure tied to attorney-ready output.

• Why measure by stage instead of total case time? Automation affects stages unevenly—intake, forms, and exhibit assembly differently from EB-1 letter drafting. Stage-level timing shows where the time actually moves.

• Can a firm reproduce the measurement? Yes—that is the point. Capture a baseline, run identical inputs through the tool at the same quality bar, and disclose the sample.

• Does Parley publish a single universal time-savings number? Time saved depends on a firm's category mix and current process. The honest unit is a measured result on a stated sample, which a firm can reproduce in a pilot.