In January 2023, administrative life at Taejon Christian International School (TCIS) ran on paper forms, email chains, and disconnected spreadsheets. Many international schools have run this way, and most run reasonably well this way, because talented people quietly compensate for complicated systems every day without complaint. Three years later, the same school runs on more than two hundred custom-built web applications and digital systems. Requests route themselves; records log themselves; parents and staff see the same information in both Korean and English. This case study examines the decisions that took the school from no systems capacity to a governed digital ecosystem, and considers why the most significant product of the most recent year was not an application at all but a coaching program that teaches school leaders systems thinking and how to use AI tools themselves. The sequence of those decisions, more than any individual system, is the key finding.
The first decision was not to start with a purchase. International schools are routinely sold platforms that promise transformation and deliver another login; every purchased platform arrives with someone else's assumptions about how a school works, and international schools, with their blended curricula, multiple languages, and unusual calendars, violate those assumptions immediately and constantly. The gap between what a platform assumes and what a school actually does gets filled by workarounds, and workarounds reproduce the original spreadsheet chaos beneath a subscription fee.
Instead, the school rebuilt one painful process at a time on tools it already paid for: Google Workspace and its scripting layer, free at the margin, already authenticated for every staff member, and already trusted by the community. The first portals were modest: a request form that no longer stalled in email, a tracker that replaced a clipboard, a signup that closed itself when the seats were gone. What mattered was not sophistication but reliability. A teacher who watches a broken process become a working one does not care how simple the fix was; they care that someone saw the problem and removed it. Each solved problem earned permission for the next, and each new portal recruited its own advocates. An ecosystem is not commissioned; it accumulates, one earned win at a time. A quieter advantage compounded underneath: every record stayed inside the same environment the school already governed, which became decisive later, when artificial intelligence (AI) suddenly entered the space and the question of where data lives stopped being theoretical.
By the time there were a dozen systems, the real threat was not failure but inconsistency. Conventions were written early and enforced consistently. Every portal shares a look and feel, so a staff member who has used one has effectively seen them all. Every application writes a log of who did what and when, which is created automatically; when something eventually goes wrong, the log is the difference between a one-minute diagnosis and an afternoon of document archaeology. Every credential lives in one guarded store rather than scattered through code. The rule: a secret that cannot be inventoried cannot be protected. Every application redeploys in place, so a link shared with parents in September still works in June; a published URL is permanently and reliably reachable.
Every user-facing screen presents Korean alongside English, because a bilingual community deserves bilingual tools. A parent community that reads announcements through a translation app is being quietly told it is an afterthought to the main community. Bilingual content changed who actually used what was built.
These conventions are unglamorous, but they are the entire reason a very small team can operate hundreds of systems; they are also what makes an estate teachable, because a successor who learns one portal has effectively learned them all.
Somewhere past 50 systems, memory stops scaling, and anything that exists only in one person's head is one resignation away from not existing at all. Turnover is a guarantee in international schools. The answer is a systems registry. Every system is documented with its name, purpose, owner, links, status, and history. The rule is absolute: all systems are recorded and maintained in the registry. New builds get a row before they get users, and retired tools keep their row, marked as retired, so the estate has a memory of its own decisions.
The registry unlocked something better than record-keeping: documentation. The school's systems manual, the reference that a successor would actually use, is produced automatically from the registry rather than written beside it, which means it cannot drift out of date. Hand-written documentation is a photograph of a moment, accurate on the day it is written and decaying silently from then on; generated documentation is a mirror, showing whatever is true right now.
When leadership asks what the school actually runs, the answer takes seconds. When accreditors want evidence that the school knows its own operations, the registry is the evidence. Schools should never underestimate this step because it produces nothing visible, but it is the step that separates a managed estate from a single-staff failure point.
As the tools grew more capable, and especially as AI entered the workflow, the school adopted one non-negotiable rule: personal identifying information never leaves its own Google environment. Names, email addresses, student numbers, and anything else that could tell a stranger who a record is about stays inside the school's walls. Any external service, however useful, sees only opaque tokens; a token is resolved back to a real name at the moment of need, inside the school's environment, and the answer is never stored outside it.
The power of the rule is the absence of an exceptions clause. Data incidents begin with "just this once;" a convenience cached here, a name stored there, each individually defensible and collectively a breach waiting for its moment. So the rule overrides everyone, including its own author, and the people maintaining the estate are instructed to push back even on the Director if an exception is requested in a hurried moment. In an era when schools are urged to move fast with AI, a bright line the school refuses to cross is not caution for its own sake; it is what makes the speed defensible to a board, to parents, and to the school's conscience. The line is easy to draw when nothing depends on crossing it and nearly impossible to draw afterward; schools should write theirs before the first AI-assisted application, not after.
The arrival of capable AI coding assistants changed the economics of everything above. Work that once took a few weeks now takes a few minutes; a request made at morning break can be live by lunch. The long tail of school life, the worthy-but-never-urgent ideas no vendor will ever serve because no market exists for them, suddenly became reachable. But the sequence mattered enormously. AI amplifies whatever discipline already exists; because the standards, the registry, and the privacy rule were in place first, the acceleration produced a coherent estate rather than two hundred inconsistent experiments. A school that adopts AI tooling before it adopts conventions risks generating chaos faster, and at a scale no human team could have produced in the same time. Discipline first, acceleration second; the sequence determines the outcome.
The current phase of the work concerns people rather than software. At present, ten of the school's leaders, notably pioneered by the Head of School, are working through an internal, self-paced, systems-leader-guided, bilingual course that walks them through installing their own AI toolchain, connecting it safely to their everyday work, and completing a first real task from their own desk, not a demonstration but a genuine piece of their own job done a new way.
The design rests on one conviction: adults do not adopt tools because a trainer was enthusiastic; they adopt tools that visibly return an hour of their own time. Privacy-safe backups of participants' working folders, with all conversation content deliberately excluded, feed a live dashboard showing who is progressing, who is stuck, and who has quietly gone further than the course asked; coaching time flows to where the dashboard points rather than requiring follow-up conversations to gather data on adoption.
The loop has now closed. The position that once built everything has become a position that coaches the people learning to build or at least to ask for themselves; and the measure of the role is no longer how many systems it produces in a year, but how many colleagues no longer need it for the things they needed it for last year. Succession follows the same logic: the registry generated the manual and the manual mapped the position honestly into the several roles it would take to replace the position.
The takeaway? The path is walkable, and it does not begin with a budget line. Solve one real problem with tools already owned, choosing first wins for certainty rather than ambition. Standardize before scaling; the conventions are worth more than any individual system. Register everything, and generate the documentation so it cannot fall out of date. Draw your privacy line before it is needed, and never cross it. Let AI accelerate discipline that already exists rather than substitute for discipline that does not. From the beginning, work toward the day the role is coaching rather than building; a digital ecosystem that depends on one irreplaceable person is not an achievement but an unpriced risk. The goal was never really the two hundred systems; the goal was a school that can do this on its own.
Read more by Dr. Richard Granger in How Generative AI Changed My Career.
Dr. Richard Granger is the Director of Systems Management at Taejon Christian International School in Daejeon, South Korea.
