Change one letter and you get a different method. The improvement cycle most owners learned as Plan, Do, Check, Act is the one W. Edwards Deming insisted should read Plan, Do, Study, Act. According to the W. Edwards Deming Institute, Deming found that a focus on Check “is more about the implementation of a change, with success or failure.” His own focus was on predicting the result of a change before making it, studying what actually happened, and comparing the two.
That sounds like a seminar distinction. It is the most practical idea in business process improvement, and it explains why so many small-business process fixes quietly fade. Nobody wrote down what they expected to happen, so when the change lands nobody can say whether it worked, and within a month the old way is back.
This playbook uses that idea to fit a real first pass at process improvement into two working days. Be clear about what two days buys you: not a finished fix, but a mapped process, one chosen change, a written prediction, and a small test that is already running. The learning arrives over the following two weeks. That is the honest version of “48 hours”, and it is still faster than most owners get anywhere with a process they have been meaning to fix for a year.
Before you start: pick the process, not the problem
If you do not yet know which part of the business is holding everything else back, start with our piece on how to scale a business, which finds the one stage setting your output by counting what is waiting in front of each stage. This playbook starts where that one stops. You know which process deserves attention; now you need to change it without breaking something else.
Good first candidates run at least weekly and end in something you can count: job finished to invoice sent, inquiry received to quote sent, order placed to order shipped, a new hire’s first day to their first solo shift. A process that runs twice a year cannot give you enough repetitions to learn anything inside a month, so leave it for later.
Day 1, morning: walk one real job through
Do not draw the process from memory. Take one job that finished this week and follow its actual trail: the email, the text message, the paper form, the spreadsheet row, the entry in your software. Write each step as one line saying who did it, what they did, and when. Wherever the trail goes quiet for a day or more, write that down as its own line, “waited”, because waiting is a step the customer lives through even though nobody performs it.
Then hand the list to the person who does this work every day and ask them to correct it. They will add steps you did not know existed, usually a workaround somebody invented because the official way stopped working. This is not a courtesy. The Institute for Healthcare Improvement, whose Model for Improvement is built around the same Plan, Do, Study, Act cycle, says in its guidance on establishing measures that it is vital to include the people affected by an improvement in deciding which measures matter. The person running the process is the one who knows where it actually hurts.
Day 1, afternoon: turn the list into a map and one number
Paper and a pen are enough for the map. If you want something the team can open and edit, three real options: draw.io is free, open source and needs no sign-up. Lucidchart has a free plan limited to three editable documents, an Individual plan at $9 per month billed annually, and a Team plan at $10 per user per month billed annually. Miro has a free plan with three boards and a Starter plan at $8 per member per month billed yearly. For one process, any of the free tiers will do.
Now choose the one number that says whether this process is doing its job for the customer. IHI calls this the outcome measure and says there is typically one per improvement effort. For job-to-invoice, it might be days from job finished to invoice sent. For inquiry-to-quote, hours from first contact to quote delivered. Pull that figure for your last ten or twenty jobs from the records you already have. You now know where you are starting from, which most process fixes skip.
Day 2, morning: write the prediction card
This is the step that makes the rest work. Take an index card, or a note on your phone, and write four lines:
- The process and its number today. For example: job finished to invoice sent, averaged across your last twenty jobs.
- The one change. Small enough to try on the next few jobs without asking anyone’s permission to rebuild anything.
- The prediction. What you expect the number to become, and by when.
- What could get worse. The thing you will watch so the fix does not quietly move the cost somewhere else.
Here is a hypothetical to make it concrete, not a reported case. A four-person heating and cooling company finds that invoices wait for a paper work order to travel back to the office in a truck, and sit in a Friday pile. The change: the technician photographs the signed work order and texts it to the office before leaving the customer’s driveway. The prediction: most invoices go out the same day, within two weeks. What could get worse: the office keys in details from a phone photo instead of the original, so billing mistakes could rise.
Writing the prediction feels unnecessary right up until the results come in. Without it, any result looks like success, because you will remember the change as the thing you hoped it would be. With it, a miss teaches you something: your picture of how the process works was wrong, and now you know where.
The line most owners leave off
The fourth line on the card has a name. IHI calls it a balancing measure, and frames it as a question: “Are changes designed to improve one part of the system causing new problems in other parts of the system?”
Our view, and it is the part of this playbook worth arguing for hardest: in a small business, the balancing measure matters more than the outcome measure. A large company has departments whose whole job is to notice when one team’s speed-up lands on another team’s desk. A small business has one owner and a few people, so a fix that makes invoices faster and billing errors more common will look like a clean win for weeks. The errors surface later as customer calls, credit memos and a bookkeeper’s overtime, and nobody connects them back to the change. One extra number, checked on the same schedule as the first, is what catches it.
Good balancing measures are usually the thing the old, slower way was protecting. Paper work orders were slow, but they were legible. A second approval step was slow, but it caught pricing errors. Ask what the old step was quietly doing, and measure that.
Day 2, afternoon: start small, and plot it
Run the change on a slice, not the whole business: one technician, one crew, one product line, or just the next ten jobs. Deming’s cycle, as the Institute describes it, broadens from a small-scale experiment to a larger implementation in its final Act step, once the learning is in. A small test is cheap to reverse, and it keeps the people doing the work in charge of how it goes.
Then make a simple running chart in a spreadsheet: one row per job, the date, the outcome number and the balancing number. IHI’s guidance is blunt about why: “Tracking a few key measures over time is the single most powerful tool a team can use.” A chart with a dot per job shows you a change that is working and a change that only had a good first week, which a monthly average hides.
Two weeks later: study, then act
Put the card next to the chart and compare. There are only three outcomes, and each one tells you what to do next.
- It did what you predicted, and the balancing number held. Adopt it. Write the new way as a short checklist the next person can follow without asking, and extend it to the rest of the team.
- The outcome improved but the balancing number got worse. Adapt it. In the hypothetical above, that might mean the office confirms the photo is readable before the technician leaves. Write a new card and run a second test.
- Nothing moved. Abandon the change, not the process. Your theory of where the delay lived was wrong, which is useful, and the walk-through notes from Day 1 usually point at the next candidate.
If you want someone to hold you to the two-week review, a short weekly report to one outside person works better than a plan kept in a drawer; we covered the evidence for that in our piece on why weekly reports beat plans.
Where AI helps, and where it cannot
An AI assistant is genuinely useful for the clerical parts of this. Record yourself talking through the Day 1 walk-through and ask the assistant to turn the transcript into a numbered step list with every “waited” line flagged. Give it your change and ask for five things that could get worse, then pick the one you can actually count. After the test, give it your notes and ask it to draft the one-page checklist for the new way.
What it cannot do is find the workaround only one person knows about, or tell you whether a photo is readable in the office at 4 p.m. That knowledge lives with the people doing the work, and the whole point of the method is to put it on paper where the business can use it. Done well, process improvement gives those people back the hours they currently spend chasing, re-keying and fixing, and makes what they know the way the business runs.
If you want a structured starting prompt, BusinessPrompter’s Process Improvement Analyzer is built to identify and remove bottlenecks in key business processes. It is a Pro-tier prompt on BusinessPrompter.com. One caution: at the time of writing, the “About This Prompt” panels on that page describe a financial-health report rather than this prompt, so judge it by its title and summary. Whatever it produces, feed its suggestions into the prediction card rather than acting on them directly.
Frequently asked questions
What is business process improvement?
It is the practice of finding, testing and standardizing changes to a piece of work your business repeats, so it produces a better result for the customer. In a small business it works best one process, one change and one measured test at a time, using a cycle such as Plan, Do, Study, Act (Source: https://deming.org/explore/pdsa/).
What is the difference between PDSA and PDCA?
The third step. The W. Edwards Deming Institute explains that Deming emphasized Study rather than Check: Check focuses on whether a plan succeeded or failed, while Study means predicting the result of a change, studying the actual result and comparing the two to revise your theory (Source: https://deming.org/explore/pdsa/).
What is a balancing measure?
A number you watch to check that improving one part of a process is not creating new problems in another part. The Institute for Healthcare Improvement describes it as looking at a system from different directions, alongside outcome and process measures (Source: https://www.ihi.org/library/model-for-improvement/establishing-measures).
One question to take into this week
Every business has at least one process that works only because somebody quietly patches it every time it runs. Which one is yours, and does anyone besides that person know how the patch works?
