Spaced Learning Practice
Build the practice in 180 days
Services first, recurring revenue second. The first automation priority is not adaptive AI — it is content lineage and QA workflow, because unsafe automation at item creation contaminates every later personalisation decision.
Stage-gated plan
| Period | Build | Commercial milestone | Exit gate |
|---|---|---|---|
| Days 0–30 | Positioning, claims policy, research library, retention map, diagnostic method, security baseline, partner deck | 15 buyer interviews and 3 design partners | One paid diagnostic or signed pilot intent |
| Days 31–60 | Competency schema, item template, fixed scheduler, email/Teams delivery, response capture, baseline dashboard | Convert first curriculum and run internal dogfood | End-to-end traceability from source to response |
| Days 61–90 | SSO, consent, segmentation, delayed test, support playbook, experiment plan | Launch one paid 100–1,000 learner pilot | Data quality and SME acceptance meet agreed threshold |
| Days 91–120 | Reusable connectors, AI drafting workspace, QA queues, multilingual workflow, executive report | Secure rollout decision and second client | At least 50% of delivery assets reusable |
| Days 121–180 | Campaign operations, model monitoring, SLA dashboard, incident process, quarterly review | Convert pilot to annual service and activate one channel partner | Repeatable implementation and positive unit economics |
| Trigger | Automated work | Required human gate |
|---|---|---|
| Source content uploaded | Parse, deduplicate, classify sensitivity, citations, detect expiry | Client SME confirms authoritative sources |
| Objective approved | Suggest prerequisite map, misconceptions, evidence types | Learning scientist and SME approve |
| Item requested | Draft scenarios, distractors, rationales, variants, metadata | ID edits; SME signs correct answer and rationale |
| Item changed | Factual consistency, duplication, reading level, bias, accessibility checks | QA owner resolves failures |
| Low mastery detected | Select approved remediation and shorten next interval | Program owner pre-approves policy boundaries |
| Source or policy changes | Find affected items, pause risky campaigns, draft updates | SME approves replacement content |
| Monthly review | Generate retention, transfer, burden, fairness, and SLA report | Measurement lead signs interpretation |
Worked example: a financial-services client changes an approved customer-verification policy. The lineage graph identifies every scenario whose answer depends on the old clause. Automation pauses those items, drafts replacements from the new approved text, and routes them to the compliance SME. Nothing reaches learners until the SME confirms the answer and rationale.
Every item needs source passage, objective, role, difficulty, correct answer, rationale, distractor rationale, misconception, language, version, approver, validity date, and accessibility status. NIST controls support ground-truth testing, adversarial evaluation, PII detection, subgroup monitoring, and human escalation.