Education & EdTech: software that actually moves learning outcomes

The incumbent stack was built for a different decade

Walk into almost any university or large school group and you will find a familiar archaeology: a student information system chosen fifteen years ago, a learning management system bolted on beside it, a separate admissions tool, a finance system that does not speak to any of them, and a thick layer of spreadsheets and manual processes filling every gap. Each product was sold as a platform and each became an island. The result is that the most basic questions — which students are at risk of dropping out, which courses actually correlate with employment, where a learner is stuck right now — are slow and expensive to answer, because the data needed to answer them lives in five systems that were never meant to be joined.

The large integrators that dominate education IT are, in many ways, the reason this persists. Their commercial model rewards long, heavily-staffed implementations and customization that locks the institution in. A 'digital transformation' becomes a multi-year program with a large standing team, and the software that finally lands is often a thin configuration of an aging product rather than something built for how learners and educators work today. Institutions pay enterprise prices and receive enterprise inertia.

AI changed the baseline expectation overnight

The arrival of capable language models reset what learners and educators consider normal. A student now reasonably expects to ask a question in plain language and get a useful, context-aware answer; to receive feedback on a draft in seconds; to have material re-explained at a different level when they are lost. An instructor reasonably expects help drafting assessments, surfacing which students need attention, and automating the administrative load that consumes their evenings. Platforms that cannot meet these expectations do not feel neutral — they feel broken, and they feel that way to exactly the people whose engagement determines whether the institution succeeds.

But bolting a chatbot onto a legacy LMS is not the answer, and the sector is full of exactly that. The hard, valuable work is integrating AI where the pedagogy lives: in how content adapts, how assessment is designed and protected against misuse, how at-risk signals are detected early, and how educators stay firmly in the loop rather than being replaced by an unaccountable model. That requires building from the data and the learning design outward, which is precisely what a configured legacy platform cannot do.

Access, accessibility and trust are not optional features

Education carries obligations that consumer software does not. A learning platform has to work for a student on a low-end phone over a weak connection, for a learner using a screen reader, and for a child whose data is protected by law. Accessibility conformance, safeguarding, data residency and consent are not items to retrofit at the end of a build — they are constraints that shape architecture from day one. Teams that treat them as a compliance afterthought ship products that exclude the very learners education exists to serve, and that expose the institution to real legal and reputational risk.

We start with the learner and the educator, not the feature list

Most education software fails not because the engineering is bad but because it was specified by people far from the classroom. We begin a project by spending real time with the learners, instructors and administrators who will live inside the product. We watch where the friction actually is — the seven clicks to record attendance, the assignment workflow that loses student work, the report that takes a registrar two days to assemble. That fieldwork produces a sharp, honest picture of where software can move an outcome, and it kills the long tail of features that look good in a procurement document but no one will ever use.

From there we work in short, visible cycles. Stakeholders see working software in weeks, on real data, not slideware at quarterly steering committees. That cadence matters in education specifically, because the people who know whether a learning workflow is right are teachers and students, and they can only tell you by using it. The faster they touch it, the faster the product becomes genuinely good.

AI built into the pedagogy, with humans in the loop

When we add intelligence to a platform we treat it as a teaching and learning decision, not just an engineering one. An AI tutor is grounded in the institution's own material so it does not hallucinate a curriculum; an at-risk model is designed with the staff who will act on its signals so it earns trust instead of being ignored; an auto-grading system is built to defer to a human wherever the stakes or the ambiguity are high. We are explicit about what the model is allowed to decide and what it must escalate, because in education an unaccountable automated decision is a serious failure, not a minor bug.

We also design for the reality of academic integrity in an AI world. Rather than waging an unwinnable arms race against students using the same tools everyone else uses, we help institutions redesign assessment so it measures understanding that AI cannot simply produce on the learner's behalf — and build the platform support that makes that redesign practical at scale.

Accessibility, privacy and resilience as defaults

Every product we build for education is designed against accessibility standards from the first wireframe, tested with assistive technology, and built to perform on modest devices and connections. Student data is treated as sensitive by default, with consent, residency and retention designed in rather than patched on. And because education runs on hard calendar deadlines — term starts, exam windows, results days — we engineer for those peaks specifically, so the system is fastest exactly when the whole institution is depending on it.

The next platform is conversational, adaptive and personal

The learning platform of the next few years will not look like a website with a course catalogue. It will behave more like a knowledgeable, patient guide that knows where each learner is, adapts in real time, and frees educators to do the human work that actually changes lives — mentoring, motivating, and intervening when a student is struggling. The institutions that win will be the ones that treat AI not as a feature to announce but as a redesign of the learning experience, with the educator firmly at the centre. The ones that lose will be those who let a vendor bolt a chatbot onto a system from 2010 and call it innovation.

The mistake we see most often is buying the brand rather than the outcome. A long, expensive program with a household-name integrator feels safe to a board, but it frequently produces a configured legacy product that is already behind the moment it launches, and a dependency that is expensive to ever escape. The cost of that safety is paid by learners who get a worse experience and by institutions that cannot move when the ground shifts again.

Outcomes and ownership will define credibility

Education is moving, rightly, toward being judged on outcomes — completion, progression, employment, real learning gain — rather than on enrollment alone. The technology has to follow. That means platforms instrumented to prove impact, credentials a learner genuinely owns and an employer can verify in seconds, and data architectures that can answer hard questions about who is being served and who is being left behind. The organizations that build for that future now, with software they understand and control, will be the ones still standing when accountability tightens further.