Case-study standard

See the shape of the work without invented proof.

These examples demonstrate how Mainspring scopes and evaluates an engagement. They are explicitly illustrative and do not represent customer results.

What will qualify as a case study

Named evidence, verified outcomes, and stated limitations.

A future customer case study must identify the measured baseline, evaluation window, relevant constraints, and customer-approved result. Anonymous or representative material remains labeled as such.

01Customer permission before publication

02Baseline and measurement window disclosed

03Observed result separated from causal claim

04Limitations and continuing ownership stated

Scenario 01Illustrative—not a customer case study

Representative merchant scenario

Situation

A merchant has a large catalog, inconsistent product discovery, and limited evidence about which placements influence orders.

Approach

Activate Recommend in one placement, establish an attribution baseline, review relevance and exclusions weekly, and expand only after stable behavior.

Useful result

A measured path to determine whether recommendation quality and attributed outcomes justify broader rollout.

Limitations

No revenue-lift claim is made without real merchant evidence.

Scenario 02Illustrative—not a customer case study

Representative document scenario

Situation

An operations team retypes fields from recurring documents and manually routes uncertain records.

Approach

Define the required schema, test representative documents, add deterministic validation, and route low-confidence fields to a reviewer.

Useful result

A bounded pilot that measures extraction accuracy, handling time, exception rate, and reviewer effort.

Limitations

Human approval remains where the workflow has financial, contractual, or customer consequences.

Scenario 03Illustrative—not a customer case study

Representative knowledge scenario

Situation

Employees search across approved policies and technical documents but need direct answers with evidence.

Approach

Use a bounded corpus, golden questions, permission tests, citation thresholds, and explicit refusal behavior.

Useful result

A go, revise, or stop decision based on answer quality, source fidelity, access control, freshness, and support burden.

Limitations

The scenario does not imply production readiness or replace accountable policy owners.

Your situation

Start with a real workflow and a measurable boundary.

Find your starting point