Improve — Designing and Proving Solutions
Improve turns verified root causes into tested, implemented solutions that reduce variation and waste. Green Belts generate options, choose among them objectively, prove the winners work with designed experiments or pilots, and lean the process with flow, pull, and error-proofing. The discipline is that solutions are validated, not assumed — you change the process on evidence that the change actually moves the output.
The purpose of Improve and generating options
The overall purpose of Improve is to develop, test, and implement solutions that eliminate or reduce the verified root causes from Analyze. It begins with generating options — structured brainstorming, benchmarking, and creativity methods — where the most important ground rule is to separate idea generation from evaluation: defer judgment so quantity and range of ideas are maximized before any filtering begins. Criticizing ideas as they surface shuts down the divergent thinking the phase depends on. Only after a broad list exists does the team converge, grouping and screening ideas toward a manageable set of candidate solutions worth testing.
Selecting solutions: Pugh matrix and pilots
Choosing among candidate solutions objectively is what a Pugh matrix (solution selection / criteria matrix) does: it scores each option against a set of weighted criteria relative to a baseline (datum), forcing a transparent, data-informed comparison instead of a loudest-voice decision. Before committing to full rollout, the team runs a pilot — a small-scale, limited implementation of the chosen solution. A pilot is valuable because it confirms the solution actually works and surfaces unforeseen problems under real conditions at low cost and low risk, before the organization spends on a full deployment. Skipping the pilot risks scaling a solution that fails in ways the team could have caught cheaply.
Design of Experiments: full and fractional factorials
Design of Experiments (DOE) deliberately varies several input factors together to learn their individual and combined effects efficiently — its key advantage over one-factor-at-a-time (OFAT) testing is that DOE studies factors simultaneously, so it uses fewer runs AND can detect interactions that OFAT structurally misses. A full factorial tests every combination of factor levels; for k factors at 2 levels each it needs 2ᵏ runs per replicate, so 3 factors at 2 levels = 2³ = 8 runs. An interaction effect exists when the effect of one factor on the output depends on the level of another factor — that dependence is exactly what OFAT cannot see. When factors are many and runs are costly, a fractional factorial runs a carefully chosen subset (e.g., 2ᵏ⁻¹) to screen which factors matter, trading some ability to resolve higher-order interactions for far fewer runs. This screen-then-optimize logic is central to the phase.
DOE integrity: replication, randomization, blocking
Three design safeguards are commonly tested. Replication means running the same factor-level combination more than once (genuinely repeating the whole run, not just re-reading the gage); it estimates pure experimental error and improves the precision of effect estimates. Randomization means running the experiment in random order; it protects against the biasing influence of unknown, time-related lurking variables (tool wear, warm-up drift, ambient change) by spreading their effect randomly across conditions rather than confounding them with a factor. Blocking groups runs into homogeneous blocks — a batch of material, a shift, a day — so a known, controllable nuisance source is accounted for and removed from the error term, sharpening the comparison of the factors of interest. Together they keep the conclusions of a designed experiment trustworthy.
Lean flow: takt, pull, and kanban
Lean tools reduce waste and smooth flow. Takt time paces production to demand: takt = available production time ÷ customer demand. With 480 minutes available and demand of 240 units, takt = 480 ÷ 240 = 2 minutes per unit — the process must complete one unit every 2 minutes to meet demand without over- or under-producing. Balancing each step's cycle time to takt prevents bottlenecks and idle stations. A pull system produces only in response to actual downstream demand rather than a forecast: in a kanban pull system, production or replenishment of a part is triggered by a signal (a kanban card or empty bin) from the consuming step, which caps work-in-process inventory and exposes problems quickly. Reducing batch sizes and moving toward one-piece flow shortens lead time and reveals defects sooner.
Kaizen, 5S, and error-proofing (poka-yoke)
A kaizen event (kaizen blitz) is a short, focused, cross-functional improvement effort — typically a few days — that rapidly analyzes and improves a specific process, delivering change fast rather than over months. 5S organizes the workplace to make problems visible and standard work easy: Sort (remove what is not needed), Set in order / Seiton (arrange remaining items so each has a marked place for quick retrieval), Shine (clean and inspect), Standardize (make the first three the routine), and Sustain (maintain the discipline). Poka-yoke means mistake-proofing: designing the process or a device so errors are prevented or made immediately obvious. Poka-yoke devices split into two functions — a control type physically stops the process or prevents the error from occurring (a fixture that only accepts a part in the correct orientation), while a warning type alerts the operator that an error may have happened (a buzzer or light). Building prevention into the process is far more reliable than relying on downstream inspection to catch defects after they occur.
Keep going: the full Lean Six Sigma Green Belt (IASSC ICGB) guide covers every section of the exam. Lean Six Sigma Green Belt — Complete Study Guide (2026) — PDF + EPUB, $19.99 · 14-day refund →
Practice stays free. The full Lean Six Sigma Green Belt (IASSC ICGB) study guide is the material itself, taught start to finish — a downloadable PDF + EPUB you keep.