The Staircase Methodology at Work: How One Small Business Picked Its First AI Project (Case Study)
In the first article of this series, I argued that the hardest part of AI for a small or medium business is not the technology or even the cost. It is knowing what to actually do, and in what order. The Staircase Methodology exists to answer that. This article shows how it could be applied end to end on a real business, so the rest of the series has something concrete to point back to.
All numbers below are illustrative. Actual costs, hourly rates, and timelines vary widely by business, vendor, and geography. The point is the method, not the specific figures.
The Business and the Pain
The business that we’ll talk about is a US home-inspection company: busy schedule, strong reputation, a custom way of working rather than off-the-shelf templates. Identifying details and exact numbers are left out for confidentiality.
The owner kept losing evenings because of the need to make inspection reports. The work was repetitive, it happened after every job, and it ate into the time that could have gone toward growing the business. That feeling, a recurring task that quietly drains the week, is a really good example of what the methodology is built to find and fix.
Step 1 and 2: A Goal, Then a Brain Dump
The first step is always a specific, dated goal. The owner picked: “In 3 months, I want to spend 30 fewer minutes per client.” Concrete, time-bound, and tied to a real pain. A vague goal like “grow the business” filters nothing; this one does.
Then came the brain dump: every repetitive task in the business, listed without judgment. Creating reports. Communicating with clients. Conducting inspections. Scheduling. Invoicing. Quarterly tax prep. Equipment maintenance.
Step 3 and 4: Narrow, Then Score
The goal targets per-client time, so the per-client tasks were circled (reports, inspections, communication, scheduling, invoicing) and the one-off tasks were crossed out (tax prep, equipment maintenance). The owner also crossed out client communication on purpose: it saves time in theory, but the personal relationship is something they value and do not want to automate. The methodology explicitly leaves room for that kind of judgment.
What remained was scored with the Phase 1 formula, Bleed = Pain × Frequency, each rated 1 to 5. The scores: report creation 20, inspections 15, scheduling 8, invoicing 6. Reports rose to the top. The score is a ranking aid, not a precise measurement, but it sorts a long list quickly and honestly.
Step 5: The Four Checks
The top candidates then ran through four checks in order.
Check 1, can AI actually do this? Reports: yes, structured text plus images is well within reach. Inspections: only partially, since the on-site work is physical, and weaving AI into that process was judged too complex for a first project, so it was crossed out. Scheduling and invoicing: yes. Three candidates remained.
Check 2, how much effort? The owner had shipped zero AI projects but was bringing in outside help, so small, medium, and large projects were all on the table. Reports came in at medium, about a week, for the descriptive sections only, with safety and code-related sections staying manual. Scheduling and invoicing were both small. Against the effort-versus-value thresholds, all three passed, and Reports cleared by the widest margin. Reports became the lead candidate, with scheduling held as a backup.
Check 3, return on investment, expressed as payback. For Reports: roughly $5,000 to build and integrate, about 20 hours saved per month (30 minutes across about 40 clients), at an assumed blended rate of $75 an hour. That works out to a payback of about 3.3 months. Comfortably under the 6-month “good to go” line.
The Moment That Matters: Risk
Check 4 is reversibility, scored as Risk = Impact × Stickiness. This is where the project almost died, and where the methodology earns its keep.
A bad report could harm a client and create real liability, so Impact was a 4. The tool itself was easy to switch off, so Technical Stickiness was a 1. But a report, once sent, cannot be unsent. That Reputational Stickiness was a 4. Stickiness takes the worse of the two, so it stayed at 4. Risk = 4 × 4 = 16. The methodology’s rule is blunt: anything 10 or above means STOP.
Stopping did not mean abandoning the project. It meant redesigning it. The owner added a simple checkpoint: the inspector reads and approves AI-generated sections before the report goes out. That single change dropped Reputational Stickiness from 4 to 2, because errors now get caught before reaching the client. New Risk = 4 × 2 = 8, in the medium range, proceed with a rollback plan. The project was cleared, but on safer terms than it started.
This is the safeguard that the headline AI failures keep ignoring. The most damaging small-business AI mistakes are customer-facing outputs that reach people before a human has checked them. One cheap checkpoint is often the difference between a useful tool and a public embarrassment.
Step 6: Commit, Then Ship
Before building, the owner committed three things in writing. A success metric: cut the targeted report section from 30 to 60 minutes down to under 10, within 4 weeks, with no drop in accuracy or quality. A check-in date: 4 weeks after launch. And an honest answer to what this project replaces: those lost Sunday-night report hours, which would now go toward growing the business.
The result: time on the targeted report section dropped from 30 to 60 minutes to a few minutes per report. About 10% of the owner’s working time comes back every working day.
If you want to run your own business through the same sequence, the methodology comes with two printable worksheets that walk you through it step by step: one for setting the goal, brain-dumping, and scoring, and one for the four checks and the written commitment. Both are included in the full methodology paper on SSRN, and I might bring them into the later articles in this series so they are easy to use alongside each phase.
The Lesson
The owner’s own takeaway is worth repeating, because it keeps expectations sane. The methodology gets you to the right project. It does not promise the first version will work. Plan for two or three rounds of refinement before measuring against your metric. The 4-week check-in is what gave the owner room to iterate rather than declaring victory or defeat too early.
One clear goal, one well-chosen project, one honest set of checks, and a written commitment with a date on it. A single step that pays for itself and teaches you something for the next one.
What Comes Next
The next article walks through Phase 1, Optimize, as a full how-to, with the worksheets and the scoring and check formulas laid out so you can run your own business through the same sequence. It refers back to these exact numbers, so this case becomes the worked example behind the method.
The full methodology, with complete references, is available on SSRN (“The Staircase Methodology for AI Adoption in US Small and Mid-Sized Businesses”) for anyone who wants the deeper version. These articles are the practical, faster-to-apply companion to it.
Next article “Phase 1 Optimize: A Practical Way to Pick Your First AI Project (“The Staircase Methodology”)”.
Previous article “Why Most Small Businesses Get Stuck on AI, and How the New Staircase Methodology Helps”.
Note: the views expressed in this article are my own and do not represent the official positions of any past, present, or future employers, clients or stakeholders.
