What are the five stages of process improvement?

A widely used five-stage cycle is DMAIC: Define, Measure, Analyze, Improve and Control, which the Juran Institute describes as a business improvement and problem-solving method run project by project. This guide's six steps map onto it, with the baseline recorded at Measure and the new process held in place at Control.

What are examples of process improvement?

Delivery Hero moved account-recovery approval from IT to the employee's manager, and n8n's case study reports the average time locked out dropped from 35 to 20 minutes. Singapore consolidated 16 separate port-clearance forms into a single application; its 100,000 man-hours saving is described by every government source as an estimate. Omega Healthcare's hours saved appear as two figures from two dates, one in UiPath's release and one from Omega via Business Insider; only the later one comes with a stated method, Omega's before and after tracking of task times.

What is business process improvement?

Business process improvement is the same work applied to a business process such as approvals, onboarding, billing or clearance: find the step that causes the delay or the error, redesign it, and measure the result.

What is continuous process improvement?

Continuous process improvement is a standing cycle rather than a one-off project: change one thing, measure it, adjust, and test again. AHRQ's Plan-Do-Study-Act method is one written form of that cycle, and DMAIC's Control phase is where a single project hands over to it.

How do you identify process improvement opportunities?

Pick one process with a problem you can count, record how it performs today, map it as it actually runs, and find the step where the delay or the error happens. At Delivery Hero that step was IT sitting in the middle of every account recovery.

How do you write a process improvement plan?

Write down the process, its owner, the problem, a baseline figure with its date, the change, what you will measure, the date you will measure it, the result and where that result came from. This guide gives the plan as a table you can copy.

Why is process improvement important?

On this guide's examples, because changing one step can change the outcome a great deal in both directions. Delivery Hero's redesign cut the average time locked out from 35 to 20 minutes by its vendor's figures; the VA's automated rules removed the human step, and its inspector general found at least 8,000 decisions or letters with problems in a review that TIN's case file puts at some 8,100.

Process improvement: steps, examples and measured results, The Internet Ninja

The short answer

Process improvement is finding where a process loses time, money or accuracy, changing how the work is done, and measuring the result against a baseline you recorded first. When Delivery Hero moved account-recovery approval from IT to the employee’s manager, n8n’s own case study reports that “the average time locked out dropped from 35 to 20 mins” source source. Singapore merged 16 port-clearance forms into one application, and its 100,000 man-hours saving is still, on every government source, an estimate source.

What process improvement means, and what it does not

A process is the sequence of steps a team runs to turn a request into a result: an expense approved, a ship cleared into port, a locked account restored. Improving it means changing those steps so the result arrives faster, cheaper or with fewer errors, and then showing that it did. Business process improvement is the same discipline aimed at the processes a business runs internally: approvals, onboarding, offboarding, billing, claims, clearances.

Three things are often confused with it.

Buying a tool. A workflow platform, a robotic process automation licence or a new ticketing system carries out a process. It does not decide what the process should be. If the steps are wrong, the tool runs the wrong steps faster.
Writing the process down. Documentation records a process. It is the last step of an improvement, not the improvement.
A one-off fix. Clearing a backlog by working a weekend changes this month’s numbers and leaves the process as it was.

The Juran Institute frames improvement work as project by project, and defines a project as “a ‘problem scheduled for a solution’”, meaning “management has decided it is important enough to schedule the resources it needs to get the problem solved” source. That is a useful test before you start: if nobody has scheduled time for it, it is a complaint, not a project.

Continuous process improvement

Continuous process improvement is a standing cycle rather than a project with an end date. The team changes one thing, measures it, adjusts and tests again. The US Agency for Healthcare Research and Quality describes its Plan-Do-Study-Act method as “a way to test a change that is implemented”, with steps that guide the work into “evaluating the outcome, improving on it, and testing again” source. A single improvement project ends; the cycle that keeps checking the process afterwards does not.

The common methods, named

Two named methods, and what each is for:

Method	What it is for
DMAIC (Define, Measure, Analyze, Improve, Control)	A “business improvement and problem-solving method”, run as a project with “five distinct stages”, typically “as a part of a larger Lean Six Sigma program” source
PDSA (Plan, Do, Study, Act)	Testing a change, then improving on it and testing again source

The steps of a process improvement, in order

The six steps below map onto DMAIC. Step 1 is Define; step 2 is Measure, where Juran says the team “begins assessing the current baseline performance of the problem” source; step 3 is Analyze; step 4 is Improve; steps 5 and 6 are Control, where the team “ensures that gains made during the improve phase are held, and the problem does not recur” source.

One example runs through all six: Delivery Hero’s account recovery. Every Delivery Hero figure below comes from n8n’s own case study, which is marketing material source; TIN’s case file records that no independent outlet reports these numbers, and that Delivery Hero’s Director of Global IT Service Delivery, Dennis Zahrt, is named on the record source.

1. Pick one process and one problem

Choose a process you can describe in one sentence and a problem you can count. Delivery Hero’s was account lockouts: “The company saw around 800 such requests per month and on average it took 35 minutes to complete the recovery process” source source.

That sentence holds everything step 1 needs: a named process, a volume and a time per unit. A problem stated as “recovery is slow” gives you nothing to measure against later.

Pick one. A team that sets out to improve “onboarding” usually finds onboarding is a dozen processes owned by four departments. Start with the one whose problem you can already put a number on.

2. Record the baseline before you change anything

The baseline is how the process performs today, measured the same way you will measure it afterwards. Record the figure, the date and how you got it.

Business Insider describes the method of Omega Healthcare, a healthcare revenue cycle management firm: “Omega Healthcare tracked the time it took employees to manually complete different administrative tasks and compared it to the time they spent on them after adding UiPath’s tool to measure the tech’s impact” source source. Time per task before, time per task after, same tasks. That is the whole method.

A missing baseline looks like this. TIN’s Delivery Hero case file shows an average time before and after, but for monthly volume and total hours locked out, its before column reads “not stated” source. The case study states a saving in total hours without stating the total it was saved from. When you write up your own result, do not leave that column empty.

3. Map the process and find the step that causes the delay or the error

Draw the steps as they actually happen, not as the procedure says they should. Talk to the people who do the work; they know the workarounds. The map should show who does each step, what they wait for and where work goes back for correction.

Then look for the step where work waits. At Delivery Hero, the case study describes the team resolving “to eliminate IT as the obstruction in the process” source source.

This step is DMAIC’s Analyze. Juran describes it as the phase where the team “collects and uses data to prove theories of root cause or causes of the problem” source. The point is to test the explanation, not to pick the most plausible one.

4. Redesign, then decide what to automate

Change the step first. Then decide whether software should carry out the new version.

At Delivery Hero the redesign was a change of approver: the manager authorises the recovery instead of IT. The team also set up “automated API calls to Okta, Jira and Google systems to trigger the actual account recovery as soon as the manager authorized it. All of this was implemented through a single n8n workflow” source source. Per the case study, time to deployment was 5 hours source source.

TIN’s reading of this case is that the gain came from moving the approval, and that the workflow tool carried out a decision already made. The case study does not separate the effect of the new approver from the effect of the automation, so this reading is an inference, not a measurement.

5. Run it and measure against the baseline

Run the new process and measure it the way you measured the baseline. At Delivery Hero, per n8n’s case study, “the average time locked out dropped from 35 to 20 mins. This meant that employees were locked out for a total of 200 hours per month less” source source.

Report the result with its source and its date. A result you measured yourself, a result your vendor published and a result someone estimated before the change are three different things, and the section on reading an improvement figure shows how far apart they can sit.

6. Write the new process down and keep checking it

Document the new process so that the next person runs it the same way.

Then keep measuring. This is where a project becomes continuous improvement: the Control phase hands over to a standing cycle of small tests. Delivery Hero’s case study reports that the result led the company “to identify and automate other manual processes using n8n, e.g. account offboarding and software license assignments” source source. Each of those is a new step 1.

Process improvement examples, with the numbers and where they come from

Four process improvement examples. TIN measured none of these figures: its case files re-read each one against the publication it came from, and the last two columns record what that showed. They matter most, because the same kind of figure means very different things depending on who published it and whether anyone measured it after the change.

Organisation	Process	What changed	The figure, as published	Who published it	Measured after the change?
Delivery Hero source source	Employee account recovery	Approval moved from IT to the manager; one workflow triggers the recovery	”the average time locked out dropped from 35 to 20 mins”	n8n, the vendor, in its case study	Reported as a result by the vendor; no independent outlet reports it
Omega Healthcare source source source	Billing, coding and insurance correspondence	Administrative tasks automated with UiPath	”monthly savings of 6,700 worker hours” (UiPath release, Oct 2024); “more than 15,000 hours a month” (Business Insider, June 2025)	UiPath, in its release; Omega, via Business Insider	For the more than 15,000 figure, yes, by Omega’s own before and after tracking; no method is stated for the 6,700 figure. Every figure originates with Omega or UiPath
Singapore, digitalPORT@SG source source	Port clearance for arriving and departing ships	Separate forms to three agencies consolidated into a single application	”is estimated to save the industry 100,000 man-hours per year”	Deputy Prime Minister, October 2019	No: “No government source found gives a method or a measured post-rollout figure”
JPMorgan, COIN source source	Interpreting commercial-loan agreements	A machine-learning system reviews the contracts	The work “consumed 360,000 hours of work each year by lawyers and loan officers”	Bloomberg, 2017, attributing it to JPMorgan’s designers	No: “It has never been independently measured or updated”

Read the table by its last column. Only one figure in it has a published method behind it, and even that method was run by the company reporting the result. The Singapore and JPMorgan figures are both widely repeated source source, and neither was measured after the change.

The digitalPORT@SG row is also the only one here with no automation tool at its centre. MPA’s service page says the system “streamlines vessel, immigration and port health clearances across multiple agencies into a single application by consolidating 16 separate forms” source source. That is a redesign: fewer forms, one submission instead of three. TIN’s case file notes that the 2019 launch text said “up to 16” rather than 16, and shows the difference rather than resolving it source source.

Estimated, claimed or measured: how to read an improvement figure

An improvement figure is one of three kinds. An estimate is a forecast made before or without measuring. A claim is a result reported by someone with a stake in it. A measurement is a result with a stated method, a baseline and a date. The words around the number usually tell you which kind you are reading.

One figure, several wordings

Singapore’s 100,000 man-hours shows how far the wording can travel while the number stays still. TIN’s case file sets the versions side by side source:

Source	Date	Wording
Deputy Prime Minister	October 2019	”is estimated to save the industry 100,000 man-hours per year” source
Ministry of Transport	2021	”we reduced the administrative efforts by an estimated 100,000 man-hours per year” source
MPA service page	Undated, read 3 October 2026	”the industry can save up to 100,000 man-hours per year” source
NCS, the builder, in its case study	Undated	”the shipping community has saved 100,000 hours per annum” source source

The government sources call it an estimate in 2019 and still call it one in 2021. Only the builder turns it into a result, and TIN’s case file calls that vendor page “the weakest source here and the only one that presents the figure as achieved” source. The old-process description, “up to 16 forms for port clearance were submitted to three government agencies separately”, also comes from NCS’s marketing, not from an MPA document source source.

The consolidation itself is not in doubt; the government states it plainly. The saving is a forecast.

One origin, repeated for years

TIN’s record calls JPMorgan’s 360,000 hours “the best-known hours-saved figure in automation” source. TIN’s case file finds it “traces to a single 2017 Bloomberg article attributing it to JPMorgan’s own designers. It has never been independently measured or updated” source. Bloomberg also reported that COIN helped cut loan-servicing mistakes source source; the case file adds: “The magnitude of the error reduction was never quantified publicly.” source.

Two figures from two dates

Omega Healthcare’s hours saved appear as two figures from two sources. UiPath’s release in October 2024 reports “monthly savings of 6,700 worker hours” source source; Omega, via Business Insider in June 2025, reports “more than 15,000 hours a month” source source. The case file treats them as “different figures from different dates” and does not merge or average them source.

The ratios Omega reports repeat across both sources. Business Insider writes that it “reduced the amount of time workers spend on documentation tasks by 40%” and “slashed document processing turnaround time by 50% with a process accuracy of 99.5%” source; the UiPath release gives the same three figures in its own words source source. Because both sources trace to the same company, the case file treats this as “repetition, not independent measurement” source.

Read the return-on-investment figure with care too. Business Insider describes it as “delivering a 30% return on investment since using the AI-powered automation”, and the UiPath release as “a 30% return on investment within the first year”, and both attribute it to Omega’s clients, not to Omega source source source. Whose return a figure measures is part of the figure.

Three questions for any improvement claim

Ask these of a vendor’s case study, a conference slide and your own write-up alike.

Who measured it? The organisation, its vendor, an auditor, or nobody yet.
Against what baseline? A recorded figure from before the change, measured the same way, or a number reconstructed afterwards.
On what date? A result is true as of a date. A later figure is a new figure, not a correction of the old one.

If a claim cannot answer the first question, it is an estimate, however it is worded.

When an improved process has no check

Speed is not the only thing a process change alters. Remove a human step and you also remove whatever that step was catching.

The US Department of Veterans Affairs processes some survivors’ benefit claims with a rule-based automated system. TIN’s case file puts the Office of Inspector General’s review at “some 8,100 automated decisions and notices”; the OIG found that “At least 8,000 decisions or letters omitted favorable findings and had issues such as incomplete evidence summaries and incorrect formatting” source source. The OIG also found that “At least 2 percent of decisions had legal errors, resulting in an estimated $2.7 million in improper payments,” a figure the report records as questioned costs of $2,727,764 source source.

The OIG placed the fault in the design, not in a glitch: “Deficiencies stemmed from flaws in automation rules and processes that, if not addressed, will recur because the system applies predefined rules without human involvement” source source. It also found “the quality review process for automated claims was less rigorous than for traditional claims” source source. This was not artificial intelligence; TIN’s case file records it as a rule-based process source.

TIN draws one inference from this case, which no figure here measures: automation makes a process faster in both directions, so a wrong rule produces wrong decisions at the same speed as a right rule produces right ones. The practical rule for step 6 follows. When a change removes a human check, put a check back on the output, at least as strict as the one you removed, and measure the error rate as well as the time.

For the automation side of this subject, TIN’s process and operations topic collects the case files.

A process improvement plan you can fill in

Copy this table into a document and fill in one row per process. Every cell should be answerable before you start, except the last three.

Field	What to write	Delivery Hero, filled in from its case study
Process	One sentence, start to end	Employee account recovery after a lockout
Owner	One named person	IT service delivery, under Dennis Zahrt
Problem	Countable, with a unit	Time to complete each recovery
Baseline figure and date	The figure, how it was measured, when	35 minutes on average; date not stated
Change	What is different about the steps	The manager approves; one workflow triggers the recovery
Measure	The same measure as the baseline	Average time locked out
Date measured	When the after figure was taken	Not stated
Result	The after figure, exactly	20 minutes on average
Source of the result	Who measured it and where it is published	n8n’s own case study; no independent outlet reports it

The Delivery Hero column is filled from TIN’s case file source source.

What this guide cannot tell you

Every example is one organisation’s published account. None is a controlled comparison, and none tells you what would have happened without the change.
Several figures come from the company or its vendor. Delivery Hero’s figures come from n8n’s case study; Omega’s originate with Omega or UiPath. TIN’s case files report them exactly, with that provenance, and do not re-measure them.
Three of the four examples in the table involve automation or machine learning. They are here as evidence about changing and measuring a process, not as a recommendation to automate.
TIN has no measurement of how often process improvement projects succeed. No success rate appears on this page because TIN has not measured one, and it will not borrow one it has not read at the primary source.
The methods are named, not compared. None of the examples tests one method against another.

Author and method. Written by TIN. Every case-file figure links to the TIN case file it was read from on 2026-10-16, and to the primary source that case file cites. Last updated 2026-10-16.

Sources

TIN, “Delivery Hero and n8n: account-lockout recovery automated in a single workflow” case file, figures from n8n’s own case study. https://theinternetninja.com/stories/delivery-hero-n8n-it-ops-automation/
TIN, “Omega Healthcare and UiPath: document automation across billing, coding, and payer correspondence” case file, figures from a UiPath release (October 2024) and Business Insider (June 2025). https://theinternetninja.com/stories/omega-healthcare-uipath-document-automation/
TIN, “Singapore’s digitalPORT@SG: 16 port-clearance forms became one, and the 100,000 man-hours is still an estimate” case file, figures from the Prime Minister’s Office, the Ministry of Transport, MPA and NCS. https://theinternetninja.com/stories/singapore-mpa-digitalport-sg-port-clearance-consolidated-into-a-single-digital-s/
TIN, “JPMorgan’s COIN: the famous 360,000-hours figure, and why it was never independently measured” case file, figure from Bloomberg, 2017. https://theinternetninja.com/stories/jpmorgan-coin-contract-intelligence/
TIN, “A VA watchdog found errors in nearly all 8,100 automated survivors’ death-benefit decisions it reviewed, with $2.7M in improper payments” case file, figures from VA Office of Inspector General report 25-00153-47, April 2026. https://theinternetninja.com/stories/va-oig-automated-survivors-death-benefit-decisions-8000-errors-2-7m-improper-payments-2026/
TIN, process and operations topic. https://theinternetninja.com/guides/process-and-operations/
Juran Institute, “Lean Six Sigma Process & Methodology: An Essential Guide”, Maureen Goldman, 2024-02-10. https://www.juran.com/blog/impact-of-lean-six-sigma-dmaic-methodology-on-project-quality/
Agency for Healthcare Research and Quality, “Plan-Do-Study-Act Worksheet, Directions, and Examples”, Health Literacy Universal Precautions Toolkit, 3rd edition. https://www.ahrq.gov/health-literacy/improve/precautions/tool2b.html
n8n, “Delivery Hero saved 200 hours/month with a single IT ops workflow”, case study, cited by source 1. https://n8n.io/case-studies/delivery-hero/
Business Insider, “A healthcare giant is using AI to sift through millions of transactions. It’s saved employees 15,000 hours a month.”, 2025-06-04, cited by source 2. https://www.businessinsider.com/omega-healthcare-uipath-ai-document-processing-health-transactions-2025-6
UiPath release on PR Newswire, “Omega Healthcare Processes 60 Million Transactions with Enterprise AI and Automation from UiPath”, 2024-10-24, cited by source 2. https://www.prnewswire.com/in/news-releases/omega-healthcare-processes-60-million-transactions-with-enterprise-ai-and-automation-from-uipath-302285442.html
Prime Minister’s Office Singapore, Deputy Prime Minister’s speech at the Global Maritime Forum Annual Summit, 30 October 2019, cited by source 3. https://www.pmo.gov.sg/newsroom/dpm-heng-swee-keat-at-the-opening-plenary-of-the-global-maritime-forum-annual-summit-2019/
Ministry of Transport Singapore, remarks by the Senior Minister of State for Transport, 13 November 2021, cited by source 3. https://www.mot.gov.sg/news-resources/newsroom/remarks-by-senior-minister-of-state-for-transport-mr-chee-hong-tat-at-the-maritime-singapore-connect-maritime-digital-challenge-event/
Maritime and Port Authority of Singapore, “digitalPORT@SG: Single Window Port Clearance”, undated, cited by source 3. https://www.mpa.gov.sg/finance-e-services/digitalport@sg
MarineLink, “Singapore Launches digitalPORT@SG”, 31 October 2019, cited by source 3. https://www.marinelink.com/news/singapore-launches-digitalportsg-472410
NCS, “Transforming the tech behind marine port clearance”, undated, cited by source 3. https://www.ncs.co/en-sg/knowledge-centre/articles/transforming-the-tech-behind-marine-port-clearance/
Bloomberg, “JPMorgan marshals an army of developers to automate high finance”, 2017-02-28, cited by source 4. https://www.bloomberg.com/news/articles/2017-02-28/jpmorgan-marshals-an-army-of-developers-to-automate-high-finance
VA Office of Inspector General, “Review of Automated Decisions for Veterans’ Service-Connected Death Claims”, report 25-00153-47, 2026-04-30, cited by source 5. https://www.vaoig.gov/reports/review/review-automated-decisions-veterans-service-connected-death-claims

Process improvement: steps, examples and measured results