The Next Input — Issue #098

In partnership with

⚡ The Briefing — 60 sec

• Riyadh Air and IBM partner to launch the world’s first “AI-native” airline
  AI airlines? If it knows my favourite drink at 33,000 ft, we’re cooking. If it forgets… we riot.

• China and the US ramp up tech and engineering plans as Australia rethinks its AI role
  We’re neutral here at The Next Input—but it is good to see someone other than the US holding the AI crown. Keeps the field competitive.

• OpenAI reportedly set to drop GPT-5.2 within 48 hours
  Oh yeah—5.2 is basically here. Might be a good week to clear your calendar.

🛠️ The Playbook — The Multimodal Customer Memory Engine

Mission Create a system that remembers customer preferences—across text, voice, images, and behaviour—and serves them personalised experiences automatically.
Difficulty Advanced
Build time 3–4 hours
ROI Unlocks “they remembered that?” moments that turn users into loyalists.

0) Why This Matters

If airlines are going “AI-native,” customer memory becomes the new battleground.
Whether it’s a drink order, a product preference, or an onboarding style—people respond to being known.
This engine gives your product that same superpower without creepy overreach.

1) Architecture

2) Workflow

1. Collect signals: text chats, voice transcripts, uploads, app behaviour.

2. Claude 4.5 Sonnet processes content into structured traits—tone, likes/dislikes, recurring choices.

3. GPT-5-mini infers behavioural preferences:

   • likely purchase style

   • communication tone

   • content length preferences

4. Store all attributes in Supabase with timestamps + confidence levels.

5. When user interacts again, system retrieves their top attributes and applies them to:

   • responses

   • recommendations

   • notifications

   • UI layout

   • product options

6. Continuously refine memory based on new signals.

3) Example Prompts

Preference Extraction (Claude 4.5 Sonnet)

From the user's message history, extract:
- explicit preferences
- implicit preferences
- habits or repeated behaviours
Return structured attributes with confidence scores.

Behaviour Inference Prompt (GPT-5-mini)

Looking at these attributes, infer:
- likely communication tone the user prefers
- how they make decisions (fast, methodical, visual-first, etc.)
- ideal content length
Keep it simple and actionable.

4) Guardrails

• Never store sensitive categories (race, health, religion, etc.).

• All preferences must be user-controlled and editable.

• Use confidence scoring—don’t assume consistency from one-off signals.

• Clear data-retention policy (rolling 12 months is common).

5) Pilot Rollout — 3 hours

1. Build a Supabase customer_profile table.

2. Run 20 real conversations through the extraction pipeline.

3. Validate attributes manually to tune accuracy.

4. Create a simple API endpoint to fetch + apply preferences.

5. Test personalised replies in a staging environment.

6. Roll out progressively to 10% of users.

6) Metrics

• Attribute accuracy (manual review)

• Increase in response satisfaction

• Repeat engagement rate

• Reduction in support friction

• Personalisation-driven conversions

Pro Tip: Use a “memory decay” curve—recent signals weigh more, older ones fade. Keeps the system from clinging to outdated personas.

🎯 The Arsenal — Tools & Platforms

• Supabase Edge Functions · Real-time storage + retrieval of user profiles · https://supabase.com

• Segment · Capture behavioural signals from every touchpoint · https://segment.com

• Mixpanel · Analyse preference impact on engagement · https://mixpanel.com

• Resend · Send personalised emails using customer memory variables · https://resend.com

Copy-paste prompt block:

Extract user preferences from:
- explicit statements
- repeated behaviours
- tone patterns

Output structured attributes with confidence levels.

💡 Free Office Hours

Want help implementing anything? Book a free 15-minute Office Hours slot—no sales pitch, just workflows solved.

→ https://calendly.com/aaron-cylentis/the-next-input-office-hours

Create how-to video guides fast and easy with AI

Tired of explaining the same thing over and over again to your colleagues?

It’s time to delegate that work to AI. Guidde is a GPT-powered tool that helps you explain the most complex tasks in seconds with AI-generated documentation.

1️⃣Share or embed your guide anywhere
2️⃣Turn boring documentation into stunning visual guides
3️⃣Save valuable time by creating video documentation 11x faster

Simply click capture on the browser extension and the app will automatically generate step-by-step video guides complete with visuals, voiceover and call to action.

The best part? The extension is 100% free

Check out Guidde

🕹️ Game Over

Personalisation isn’t magic—it's memory used well.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: https://cylentisai.beehiiv.com/subscribe