The record, with the approval standard attached
Case Studies
Software development case studies are usually written to flatter the vendor. These are written to be checked: a named situation, what actually changed, and a number the client would sign their name against. Nothing reaches production without that signature.
01Filter by the change you are weighing
Nine records, filtered by the problem.
Browsing by industry tells you who we worked with. Browsing by the kind of change tells you whether we have solved your problem before, which is the question that actually matters. Every entry below is a sample record while written client approvals are collected.
Showing 9 of 9
Every record below is a sample pending written client approval, and none reach production until approved.
- EdTech · Modernisation
A twelve-year monolith, decomposed without a lost order.
1 to 14Deploys per monthAdmissions ran on a system nobody dared deploy on a Friday. Three waves of decomposition took change failure from routine to rare.
Sample - FinTech · Modernisation
An end-of-life platform exited with zero switch-day surprises.
NoneSwitch-day incidentsA settlement engine faced a runtime deadline. Parallel running proved behavioural equivalence before any customer touched the new path.
Sample - Workflow SaaS · Product delivery
From stalled roadmap to monthly releases in one quarter.
5 months to monthlyRelease cadenceA platform team had shipped nothing for five months. Architecture triage and a thinner release path restored a monthly cadence.
Sample - Healthcare · Product delivery
A clinical workflow platform carried from pilot to 40 clinics.
40 in 7 monthsClinics liveA pilot had to become an operable product. Architecture, compliance evidence and release discipline took it across that line.
Sample - Logistics SaaS · AI in production
A support copilot that answered 6 in 10 tickets, safely.
6 in 10Tickets resolvedDeflection without hallucinated refunds. Retrieval over governed content, confidence thresholds and human handoff shipped in one quarter.
Sample - Healthcare · AI in production
Document intake automated with an auditable trail.
Cut by two thirdsManual intake effortClinical intake forms were processed by hand. Extraction with human verification cut the queue while keeping every decision reviewable.
Sample - Workflow SaaS · Data platforms
Fourteen pipelines, one warehouse, zero definition fights.
Days to minutesReport preparationFinance ran on exports and disagreement. A governed warehouse with tested pipelines gave every metric one owner and one number.
Sample - EdTech · Team extension
Three engineers, embedded in a week, still there two years later.
2+ yearsRetentionPlatform muscle was needed without a new delivery model. Three backend engineers joined existing reviews and standups, and nothing else changed.
Sample - Workflow SaaS · Team extension
One engineer took deploys from fortnightly to daily.
Fortnightly to dailyRelease frequencyReleases moved every two weeks behind a manual checklist. A single embedded DevOps engineer rebuilt the pipeline inside the existing process.
Sample
02What earns a place here
Why this library is shorter than most.
Publishing a case study is a governed act here, not a marketing decision. Four conditions must hold before a record appears, and any one of them failing keeps the story off this page no matter how good it is.
Written client approval
The client reads the record and approves it in writing, including the numbers. Silence is never treated as consent.
A signature, not an assumptionMeasured, not estimated
Every figure traces to something observable: release frequency, resolution time, audit findings. Impressions do not qualify.
A number with a sourceWhat went wrong, included
Records state the wrong turns and the trade-offs taken. An engagement with no difficulties did not happen the way the story claims.
The honest versionReviewed, not archived
Claims are re-checked against reality on a schedule. A record that has quietly stopped being true is removed rather than left to age.
A record kept current
03A test you can apply anywhere
Four questions for any vendor's case study.
These questions work on our records and on everyone else's, which is the point of publishing them. A case study that cannot survive all four is marketing wearing evidence as a costume.
Is the client named, or implied?
An unnamed client can mean an approval still pending, which is honest, or a story nobody would sign. Ask which. Vagueness that persists after asking is your answer.
Is the number measurable by the client?
Release frequency and resolution time are observable. Efficiency gained and value unlocked usually are not. Prefer figures the client could verify from their own systems.
Does the story include a difficulty?
Real engagements contain wrong turns. A record with no friction has been edited until it stopped being evidence and became a brochure.
Is the engagement shape stated?
Team size, duration and who directed the work decide whether the outcome is comparable to yours. Without that context, an impressive number tells you nothing.
04Three ways to use this
Take the evidence somewhere useful.
A library is only worth the decision it informs. If a record here resembles your situation, the fastest next step is usually a scoping conversation rather than more reading. The practice pages carry the full method behind each kind of change.
- Product engineering
Read the method behind the outcome
Every record traces back to a practice page with the full scope, delivery system and honest exits stated.
- Project scoping
Compare a record with your situation
A scoping engagement turns the resemblance into a plan, priced against your estate rather than someone else's.
- Engineering blog
Read the working notes
The blog carries the reasoning between engagements: the decisions, the trade-offs and the things that did not work.