AI Opportunity Map

Instructional Design Workflow
What's visible. What can change.

Learning & Talent Dev
June 2025
Consequence of failure High
Scale of impact ~2,000 employees across all functions
Aggregate time cost Est. 60–80% of project hours in development & review
Strategic relevance Core to onboarding, enablement, and upskilling velocity
Automate
Augment
Keep human
Stop / rethink
↓ Expand each step to see the judgment layer
Workflow Steps
01
Intake & Request Triage
Receiving requests, clarifying scope, deciding whether to proceed
Automate Augment Keep human
+
Hidden judgment layer

What actually happens here

  • Deciding whether training is even the right solution
  • Reading the real need behind the stated request
  • Assessing stakeholder credibility and urgency
  • Identifying missing context before agreeing to scope

Where it breaks down

  • Requests accepted without enough diagnostic questions
  • No consistent intake format — information scattered
  • Scope creep begins here because agreements are verbal
Automate
Intake form routing, request logging, confirmation messages, basic info gathering
⚠ Watch for
Over-automating intake can create a false sense that requests are understood before a human has reviewed them — automated logging isn't the same as scoped agreement.
Augment
AI surfaces diagnostic questions based on request type; flags when training may not be the right solution
Keep human
The actual decision to proceed, push back, or redirect — requires relationship context and judgment
02
Discovery & Needs Analysis
Interviewing stakeholders, reviewing existing content, diagnosing the real gap
Augment Keep human
+
Hidden judgment layer

What actually happens here

  • Knowing which questions to ask — and which to save for later
  • Distinguishing symptoms from root causes
  • Sensing when a stakeholder doesn't know what they need
  • Deciding how much discovery is enough

Where it breaks down

  • Discovery depth varies by designer, not by project need
  • Interview notes not captured in reusable format
  • Findings rarely surface back to team for pattern recognition
Automate
Transcript generation, note synthesis, existing content audit across known repositories
Augment
AI generates discovery question sets per project type; synthesizes interview themes; flags gaps in information before design begins
⚠ Watch for
AI-synthesized themes can surface confident-sounding patterns from thin interview data. If discovery inputs are weak, the synthesis will be wrong in ways that aren't obvious.
Keep human
Reading the room; building stakeholder trust; making the call on root cause vs. symptom
03
Instructional Design
Defining learning objectives, structure, modality, and assessment approach
Augment Keep human Conditional
+
Hidden judgment layer

What actually happens here

  • Choosing modality based on audience, not just preference
  • Balancing what stakeholders want with what learners need
  • Deciding what to leave out — often more important than what to include
  • Applying taste: what does "good" actually look like for this learner?

Where it breaks down

  • Design rationale rarely documented — lives only in the designer's head
  • Inconsistent quality standards across team members
  • Modality decisions often driven by tool familiarity, not learner need
Conditional: Automate objective generation if documented design criteria exist; Keep Human on modality and structure decisions until criteria are defined and validated.
Automate
Learning objective generation from discovery notes; template scaffolding; modality recommendation based on documented criteria
Augment
AI applies team's documented design principles as a live check; flags when design decisions contradict stated learner needs
⚠ Watch for
If design principles aren't yet documented, AI checks will default to generic instructional design conventions — which may conflict with this team's actual standards.
Keep human
The taste layer: final calls on structure, emphasis, and what "good" looks like for this specific context
04
Content Development
Writing, building, producing the actual learning asset
Automate Augment Stop / rethink
+
Hidden judgment layer

What actually happens here

  • Translating design decisions into actual content — where time is heaviest
  • Making real-time quality calls while drafting
  • Knowing when a draft is "good enough" vs. worth another pass
  • Managing the gap between first draft and stakeholder expectations

Where it breaks down

  • Most time spent here — and it's the most automatable stage
  • Content rebuilt from scratch when similar assets already exist
  • Inconsistent voice/tone across team — no shared style baseline
Automate
First-draft generation from design doc; repurposing existing content; formatting, templating, basic visual production tasks
⚠ Watch for
Draft quality is only as good as the design doc input. Automating from a weak brief produces confident-sounding content that requires heavy rework — sometimes more than writing from scratch.
Augment
AI applies documented voice/quality standards; flags content that drifts from learning objectives; suggests existing assets to reuse
Stop / Rethink
Building bespoke content when templated or reusable assets would serve the learner equally well — this is a default habit worth questioning
05
Stakeholder Review & Revision
Gathering feedback, managing revisions, getting sign-off
Automate Augment Stop / rethink Provisional
+
Hidden judgment layer

What actually happens here

  • Deciding which feedback to incorporate vs. push back on
  • Managing stakeholders who want to reopen design decisions
  • Knowing when revision cycles are producing diminishing returns
  • Protecting learner needs when stakeholder preferences conflict

Where it breaks down

  • Revision rounds are often open-ended — no agreed stopping criteria
  • Feedback arrives in multiple formats, untracked
  • Same comments surface repeatedly — not caught earlier in process

This classification is provisional — see gap report for what's missing.

Automate
Feedback consolidation across reviewers; change tracking; version management
Augment
AI flags feedback that contradicts agreed objectives; catches common issues earlier in development to reduce revision volume
Stop / Rethink
Unbounded revision cycles with no agreed criteria — this process pattern should be redesigned before optimizing within it
⚠ Watch for
Automating feedback consolidation inside a broken review process makes the broken process easier to sustain — which is the opposite of the intended outcome.
06
Launch & Measurement
Deploying the asset, tracking completion and impact, iterating
Automate Augment Stop / rethink
+
Hidden judgment layer

What actually happens here

  • Deciding what to measure — and what "good" actually looks like post-launch
  • Connecting learning activity to business outcomes (often not done)
  • Knowing when to iterate vs. retire an asset

Where it breaks down

  • Measurement often defaults to completion rates — not impact
  • Baseline rarely documented, making "improvement" hard to prove
  • Assets launched and never revisited
Automate
Completion tracking, data collection, scheduled performance reports against baseline
Augment
AI surfaces patterns in engagement and feedback; flags assets underperforming against stated objectives
Stop / Rethink
Measuring completion as a proxy for impact — this metric persists by institutional habit, not because it answers whether learning happened
⚠ Watch for
Automating completion-rate reporting makes it even easier to report on the wrong thing with high confidence. Define outcome metrics first.
Summary
Automate
8
Discrete tasks fully automatable today with minimal risk
Augment
10
High-value tasks where AI amplifies human judgment without replacing it
Keep Human
7
Judgment, taste, relationship, and context — not safely delegated
Stop / Rethink
3
Practices that persist by habit — worth questioning before rebuilding
Measurement Baseline

How we'll know it worked

Baseline → target
Time: intake → first draft
Baseline: ~3–5 days (est.)
→
Target: 1–2 days with augmented workflow
Revision rounds per project
Baseline: avg. rounds not yet tracked
→
Target: reduce by catching misalignment earlier
Judgment documented
Baseline: 0 — lives in heads only
→
Target: principles captured per workflow stage, reusable
Skills to Act on This Map
Capabilities needed to execute
Prompt design for content drafting
Step 4 has the highest automation surface on this map. Getting useful first drafts requires practitioners who can write structured briefs that translate design docs into generation-ready inputs — not just "write me a module."
Judgment documentation
Steps 3 and 5 are conditional or provisional precisely because design criteria and review standards don't exist in written form. The skill needed here isn't AI fluency — it's making tacit standards explicit so AI can apply them.
AI-assisted synthesis and analysis
Steps 2 and 6 require practitioners to evaluate AI-generated summaries critically — recognizing when interview synthesis is pattern-matching on thin data, or when performance reports are measuring the wrong thing.
Process redesign before automation
Three steps carry Stop/Rethink classifications. Acting on this map requires willingness to fix the underlying process before layering in tools — otherwise automation makes broken workflows faster to repeat.
Workflow piloting and iteration
Most augment opportunities here depend on what gets documented and validated in a pilot. The skill is running structured experiments — testing one step at a time, capturing what actually changes, and updating the map based on evidence.
This map is a starting point, not a final answer. The baselines are estimates — capturing real data is the first step.
The "stop / rethink" column is intentionally uncomfortable. That's the point.