AI can make you faster—and blander. The art is knowing where machines should amplify people, not replace them.
The paradox
A French automaker invited 10,000 engineers to propose new digital features. The prompt was broad; the response was huge: 1,500 ideas and 3,500 comments. Humans couldn’t read it all, so popularity became the proxy. Many original, fringe ideas likely died unseen.
Lesson: At scale, human-only review fails—but AI-only ideation flattens originality. The answer is a deliberate human-in-the-loop model.
What AI is good for (today)
Think in four verbs:
- Augment – spark and enrich human creativity (seed ideas, generate visuals).
- Assist – digest and structure large volumes (cluster, deduplicate, summarize).
- Accelerate – speed the pipeline (faster triage, faster synthesis, quicker tests).
- Automate – remove repetitive toil (translation, routing, tagging, notifications).
The real headaches AI can relieve
- Inspiration on demand
Give teams curated “starter kits”: adjacent examples, patterns, quick AI visuals. Idea quality rises when minds have something concrete to push against.
- Overlap and duplication
Use semantic clustering to merge near-identical ideas expressed in different words, languages, or formats. Stop seven teams solving the same problem.
- Volume triage
Let models score by novelty, feasibility markers, and thematic fit—so humans spend time on the right 10–15%, not the loudest 10–15%.
- Language barriers
Live, high-quality translation allows people to submit and collaborate in their native language without losing nuance.
- Curation of the “innovation lake”
Mine years of ideas, insights, and experiments to:
– Resurface shelved concepts now made viable by new tech.
– Spot emerging themes across business units.
– Auto-connect people with shared interests across time zones.
- Partner scouting
Crawl public signals to find startups/scaleups and set standing searches that alert you when something relevant appears—without manual sleuthing.
Where AI falls short (and why that matters)
- Data you can’t see, you can’t use
Much meaningful knowledge sits behind firewalls or in tools the model can’t access.
- Tacit knowledge lives in heads
The “click” between experience and weak market signals often isn’t written down. Humans still outperform on vague, early cues.
- Assets & competencies aren’t codified
If you haven’t documented what you’re truly good at (beyond current products), AI can’t recommend credible adjacencies.
- Mid-tier idea bias
Out-of-the-box model outputs tend toward “competent average,” not breakthrough.
- Decision context
Final portfolio calls depend on constraints (capacity, regulation, brand risk) that models typically don’t internalize well.
- Multi-company projects
IP, culture, and security realities limit cross-org data sharing; automation helps, but doesn’t erase governance work.
A practical operating model (people + machines)
1) Design the front door
- Write prompts/briefs that are specific enough to control volume and protect IP.
- Require a minimum viable data set for every submission: effort, investment, expected impact, time horizon, confidence.
2) Machine first, human final
- AI clusters, scores, summarizes.
- Humans review clusters, not individual ideas; they make the advancement/kill decisions.
3) Build the “Innovation Shelf”
- Park “good but not now” concepts with lightweight metadata; routinely re-scan with AI as tech/regulations change.
4) Make signals a habit
- Quarterly cadence: trend/tech scans + customer executive dialogs → update focus themes → refresh prompts.
5) Guardrails
- Clear data boundaries, opt-in sources, audit trails, and explainable scoring.
- Recognition for stopping as well as shipping (to avoid sunk-cost bias amplified by automation).
30/60/90-day rollout
Days 1–30 (crawl)
- Enable translation, clustering, and ranking on one active challenge.
- Pilot AI image generation to visualize top ideas.
Days 31–60 (walk)
- Ingest historic idea/insight data; create a searchable map.
- Stand up the Innovation Shelf and a monthly “reactivation scan.”
Days 61–90 (run)
- Launch automated startup scouting for 1–2 themes.
- Add a portfolio view with weighted impact (impact × confidence × time).
What to measure
- Throughput: time from submission → human decision.
- Uniqueness score: % of ideas not clustered into existing themes.
- Reuse rate: shelf → reactivated concepts progressed.
- Partner hit-rate: scouted → qualified → piloted.
- Decision quality: concept survival to next gate; post-mortems on kills.
- Engagement equity: contributions and wins across languages/regions.
Do this, not that
- Do use AI to find, group, and visualize; don’t outsource judgment.
- Do seed with examples and images; don’t let models set the brief.
- Do codify assets/competencies; don’t assume products = strengths.
- Do reward prudent kills; don’t equate persistence with performance.
Bottom line
AI doesn’t kill creativity—uncritical reliance on AI does. Treat models as powerful exoskeletons for human imagination: excellent at volume, structure, and speed; reliant on people for context, leaps, and taste.
.