Software judged by whether people finish the course.
Education platforms are rarely abandoned because of a missing feature. They are abandoned because they are slow on the device a student actually owns, or because the content is wrong for the person in front of it.
- 01Typical workLMS, adaptive learning, assessment, SIS
- 02AccessibilityWCAG 2.2 AA as the working standard
- 03DevicesOlder phones and shared machines
- 04SafeguardingAge-appropriate design and data minimisation
- 05StandardsSCORM, xAPI, LTI where relevant
- 06MeasureCompletion and progression, not logins
The feature list is almost never the problem.
Institutional buyers evaluate on features and then find that adoption depends on something else entirely: whether it loads on a five-year-old Android phone over a poor connection, whether a lecturer can upload a course without training, and whether the student can find where they left off in under three taps.
Those are engineering and design decisions, and they are almost impossible to retrofit once a platform is built for a demonstration, not for a Tuesday morning.
Adaptive learning is the genuine advance here, and it is more subtle than recommendation. What helps is adjusting difficulty and sequencing to the individual, spotting where a cohort consistently stumbles, and giving the educator that signal rather than a dashboard.
AI also does something quietly valuable in this sector: generating practice material and formative feedback at a volume no teaching team could produce by hand, which is where much of the measured gain comes from.
Six pieces of education work we do repeatedly.
Learning platforms & LMS
Course delivery, progress and assessment built to work on the devices your learners have.
Web · mobile · SCORM · LTIAdaptive learning
Sequencing and difficulty adjusted per learner, with the signal fed back to the educator, not hidden in an algorithm.
Learner model · content graphAssessment & feedback
Practice generation and formative feedback at volume, with human review where the assessment counts.
Item banks · marking · moderationStudent information systems
Enrolment, attendance, progression and reporting, integrated instead of re-keyed.
SIS · finance · timetablingLearner & staff support
Answering routine questions from your own handbooks and policies, with escalation to a person where it matters.
Handbooks · policies · ticketingProgression analytics
Identifying the learners at risk of not completing early enough to intervene.
Engagement · assessment · attendanceCompletion is the measure. Not every feature moves it.
Education roadmaps fill up with things that demonstrate well and change nothing. This is the grid we use to argue about priorities with an academic board.
Worth planning properly
Adaptive sequencing and genuine personalisation. Real gains, real cost, sequence it deliberately, not starting here.
Do this first
Performance on the devices learners own, accessibility, and knowing where they left off. Unglamorous, and it decides adoption.
Decline politely
Bespoke gamification, VR modules and social feeds. They demonstrate well and rarely appear in completion data.
Fine, but not instead
Cosmetic polish and reporting tweaks. Worth doing when the queue is clear, never worth doing before the quadrant above.
Before you build an education platform.
Three questions we are asked by almost every education client.
Should we build or buy an LMS?
Buy, in most cases, and we will say so before quoting. The mainstream learning platforms are mature, cheap relative to building one, and solve problems you would otherwise rediscover. Building is justified when your pedagogy is the product (an adaptive model, an unusual assessment approach, a subject-specific simulation) or when you are selling the platform itself. A common and sensible middle path is buying the LMS and building the distinctive layer on top of it through LTI, which is what we do most often here.
Is AI marking acceptable?
For formative feedback, generally yes and it is genuinely valuable; practice at a volume no teaching team could mark by hand, with feedback fast enough to be useful. For summative assessment that contributes to a qualification, we would not recommend it without a human in the loop, and most awarding bodies would take the same view. The defensible pattern is AI producing a first-pass mark with the reasoning and evidence attached, and an educator confirming or adjusting, with the record showing who decided.
How do we handle learner data if some are children?
By collecting less and being explicit about what remains. Age-appropriate design obligations in several jurisdictions now go beyond consent into defaults, profiling and nudging, which affects product decisions, not just the privacy policy. Practically that means minimising what is held, avoiding behavioural profiling that serves the institution rather than the learner, setting conservative defaults, and being able to delete on request. It is much cheaper to design this in than to remove it later.