Learning that changes what people can do.

▼ ▼ ▼

About

I design learning experiences that change what people can do and build the assessments that show they work.

Over the course of a decade I've designed curricula, courses, and assessments from middle school through college: writing and literature programs, dual enrollment courses aligned to postsecondary outcomes, a published ESL writing curriculum, and a teacher-preparation course and field experience for preservice English educators. I served on the University of Montana's writing curriculum review committee, and my graduate research evaluated learner outcomes for a community writing coaching program.

For five years, I leveraged product strategy and design into real customer outcomes as a product content designer in corporate, helping build fintech products used by millions. User research, cross-functional partnership, system-level guidelines, human-centered design - I did it all, all in support of exceptionally clear user experiences.

I work at the intersection of learning science and design principles to develop end-to-end learning experiences, from diagnosing the problem and exploring divergent solutions to crafting learning outcomes and the experiences to reach them.

Portfolio

Performance problem

Authors publish AI-drafted content without recognizing tone failures, missing risk information, or drafts that need compliance review.

My role

Instructional designer: needs analysis, objectives, assessment and evaluation design, practice design, design document, storyboard, and Rise build.

Stakeholders (scenario)

Sponsor: Director of Customer Education. SMEs: Content Design Lead, Compliance Principal. Reviewer: Support Team Lead.

Solution

45-minute scenario-based Rise module, one-page job aid, follow-up practice, and three non-training recommendations.

Timeline and tools

4 weeks. Articulate Rise (SCORM to LMS), forms for scored edits, WCAG 2.1 AA.

Success measure

AI-assisted content meeting standards at publication: 22% baseline, 80% target.

Wayline, its voice guidelines, and its content audit are a fictional scenario that supplies the source material, the way a client brief would.

The brief

Wayline, a fictional retail investing app, gave employees an AI writing assistant six months ago. Its content design team supplied the voice guidelines and an audit of 18 AI drafts: 78% missed at least one standard, and a third of routine drafts drifted into regulated language. The request to L&D was training.

I treated the audit as a symptom, not a diagnosis. It showed what was going wrong in published content, not why authors were letting it through or whether training was the right fix.

Discovery

Sources: the audit; interviews with the content design lead, a compliance principal, and the support team lead; a review of the AI assistant's configuration; and observation of the prompt, review, publish workflow.

  • Training fixes one of four causes. The other three need changes to the AI tool, the escalation policy, and how authors are measured.
  • Topic doesn't predict tier. Authors judged a draft's risk by its subject, but a third of routine drafts contained regulated language. Triage had to be taught as reading for content.
  • One draft exposed a policy gap. An operational article was escalated for a single removable phrase. Whether authors may cut incidental phrases was a policy question, so I routed it to compliance before designing practice around it.

Task analysis

  1. Prompt the assistant and receive a draft.
  2. Triage the draft by its content, not its topic. The audit shows this step is skipped.
  3. For lower and medium tiers, edit against the standards.
  4. For medium tier, check that risks and benefits get balanced treatment.
  5. For high tier or substantive regulated language, route to compliance without editing.
  6. Publish, or send for review.

Steps 2, 4, and 5 are where the failures occur. They are judgment tasks, which calls for scenario-based practice rather than information-only training.

Needs analysis

CauseFixRecommendationOwner
Authors can't reliably triage drafts or recognize regulated driftTrainingScenario-based module and job aidL&D
The AI assistant's instructions contain no guidance or guardrailsToolAdd standards, terminology, and flags for regulated languageContent design, engineering
No defined escalation rule or channelPolicyDefine the rule and a single channelCompliance
Author goals count volume onlyIncentivesAdd a sampled quality measureContent leadership

The training won't fix failures introduced by the tool's configuration, unclear escalation before compliance defines the channel, or quality tradeoffs driven by volume-only goals.

Audience

AudienceCurrent stateConstraintsSolution
Help center and education authors (~25)Strong writers; edit for polish, miss tier and risk balanceDeadline-driven, remote, asynchronousFull module plus job aid
Support specialistsHigh volume; tone slips in problem momentsLittle time away from the queueJob aid and reply templates
Product designersAlready trained in content standardsSmall groupJob aid; rule communicated directly
MarketingAlready routes everything to complianceHigh tier by defaultOut of scope

The full design document is a two-page Learning Solution Design Document. Its core sections:

Performance objectives

Given an AI-drafted help or education article and the voice guidelines, the author will:

ObjectiveAssessed byMastery criterion
1 Triage the draft by tier based on its content, not its topicClassifying 10 unseen drafts9 of 10 match the expert key, including every high-tier draft
2 Edit a lower-tier draft to meet every rubric criterionRubric-scored edit of 2 draftsMeets standard on every criterion
3 Rebalance a medium-tier education draft so benefits and risks get balanced treatmentRubric-scored edit of 1 draftComparable detail on risk and benefit; no regulated language remains
4 Apply the threshold rule: cut incidental phrases, route substantive onesClassifying 6 phrases5 of 6 match the key, with no substantive phrase classed as incidental
5 Escalate a high-tier draft without editing it, with a note naming the triggerEscalation note for 1 draftDraft unedited; note names trigger and tier

Solution approach

The module is built around practice, not presentation: learners make the decisions the job requires, on realistic drafts, with feedback that explains the reasoning. Each skill opens with a worked example, then guided and independent practice. Threshold practice pairs incidental and substantive phrases side by side, and under-escalation feedback shows what would have happened if the draft had published. Short follow-up practice at two and four weeks supports transfer.

Module flow, 45 minutes

SectionMinObjectivesActivity
Pre-assessment51, 2Triage 5 drafts; edit 1 short draft
Why this matters2—A published draft causes a compliance finding and a customer complaint
Triage61Expert walkthrough of 2 drafts, then sort 6 by tier
Incidental or substantive54Contrasting pairs, then classify 8 phrases including borderline cases
Editing lower-tier drafts72Spot failures in an annotated draft, then edit and compare with a model edit
Rebalancing education content43Compare versions, then rewrite one risk passage
Escalating35Branching scenario; write an escalation note from a template
Calibration32, 3Score a draft with the rubric, then compare with the expert's scores
Post-assessment10AllItems from the assessment blueprint

Accessibility: drafts are presented as text, not images; sorting activities have a keyboard-accessible alternative; feedback never relies on color alone. Content meets WCAG 2.1 AA.

[Link to design document PDF]

Each major decision, with the options considered and why one was chosen.

DecisionOptions consideredChosen and why
Solution mixTraining alone; training plus fixes owned elsewhereTraining for one cause and three non-training recommendations, because training alone can't move the business measure
FormatWriting instruction; scenario-based practiceSelf-paced scenarios under 45 minutes: the gap is judgment on someone else's draft, and authors are remote and deadline-driven
Escalation ruleConservative: escalate any recommendation or promissory language. Threshold: authors cut incidental phrases in lower-tier contentRouted to compliance as a policy decision, since it determines what practice teaches. Built around the threshold rule compliance adopted
Assessment approachAuto-scored items only; rubric-scored editsTriage auto-scored in Rise; edits scored outside Rise against an anchored rubric, because Rise can't score free text
Support specialistsFull module; job aidJob aid and reply templates: high-volume work with little time away from the queue
Scope cutInclude prompt-writing; exclude itExcluded: better prompts reduce failures without removing them, and the tool fix addresses the same cause more reliably

What each choice costs, and how the design handles it.

ChoiceCostHow it's handled
Threshold escalation ruleRequires a clear definition of "incidental" and more practice on borderline casesWorking definition approved by compliance before the storyboard; dedicated borderline practice
Edits scored outside RiseNeeds human raters and SME timeAnchored rubric and rater calibration; about four SME hours for anchors and calibration
Under-escalation fails an itemPushes learners toward escalating everything, which recreates the review backlogOver-escalation earns partial credit; unnecessary escalations tracked with a target under 10%
Tool fix ships before the trainingDelays the training's launch by the 30-day measurement windowSeparates the training's effect from the tool's; if the fix slips, attribution shifts to a comparison group
45-minute modulePre- and post-assessment take 15 of the 45 minutesAssessments can be separated from the module if one sitting is too long
Job aid instead of a module for supportLess practice for support specialistsReply templates and direct routing rules for high-risk replies

Assessment

AssessmentWhenItemsScoring
Pre-assessmentBefore the moduleTriage 5 drafts; edit 1 draftSame rubric as post; sets each learner's baseline
Practice checksDuringTriage, phrase sorting, edits with model-answer comparisonFormative; feedback only
Post-assessmentEnd of moduleTriage 10 unseen drafts; classify 6 phrases; edit 2; rebalance 1; escalate 1Against the mastery criteria
  • Anchored rubric. Edits are scored on five criteria at three levels, with a real example anchoring each level so more than one rater can use it.
  • Risk-weighted scoring. A correct tier earns full credit, over-escalation partial credit, and under-escalation fails the item.
  • Rater calibration. A second rater scores five anchor edits independently; scoring begins when raters agree on at least 4 of 5 criteria per edit.
  • Separate item banks. About 20 practice and 20 assessment items, balanced across tiers and failure types.

Evaluation plan

LevelMeasureMethodTimingTarget
1 ReactionRelevance and usabilitySurvey plus one open questionEnd of moduleAt least 4 of 5
2 LearningTriage accuracy and edit qualityPre- and post-assessmentBefore and afterMastery criteria
3 BehaviorRubric scores and escalation accuracy on the jobMonthly content sampling; escalation log review30 and 60 days80% meet standard; unnecessary escalations under 10%
4 ResultsContent meeting standards; regulated drafts escalated; compliance findingsBusiness measures90 days80%; 100%; no findings on trained authors' content

Attribution. The AI tool fix will also improve draft quality, so Level 3 and 4 gains can't be credited to training alone. The fix ships first and is measured for 30 days before the training launches.

StageReviewersStatus
Design documentSponsor, Content Design Lead, Compliance PrincipalDrafted
StoryboardContent Design Lead, Compliance PrincipalIn progress
Alpha buildSMEsNext
PilotFive or six authors; Support Team Lead for the job aidNext

The pilot will measure triage accuracy and rubric scores before and after the module. Findings and the revisions they lead to will be added here.

Artifacts

  • Learning Solution Design Document [link]
  • Storyboard [in progress]
  • Rise module [next]
  • Rubric with anchor examples [next]
  • Job aid [next]