Features & roadmap
Thinking Ad is one creative object that takes any shape, writes itself, unfolds into a mini website, answers questions, sells, and knows when to disappear. Here is what it does today and what comes next.
What it does today
- Any shape, any dimensions
- One creative object re-composes into strip, tower, card or badge layouts instead of scaling.
- AI creative generation
- Headline, body, CTA, copy variants, mini-site panels and hero imagery generated from a short brief.
- Five renditions
- Text, static, real-world, animated and video — the advertiser enables any mix and picks the default.
- Magical pop-up mini-website
- The ad morphs from its exact position into a draggable, resizable multi-panel site — a continuous object, not a modal.
- Audio engine
- Web Audio cues for appear, expand, message and reward, muted by default and user-triggered.
- In-ad dialogue
- A streaming AI assistant with a persona, product knowledge and a hard ceiling on the discount it may grant.
- Awareness and self-teaching
- The ad pulls facts from a public source, reads page and session context, and logs the questions it could not answer.
- Offers and checkout
- Colour, size and quantity selection, discount codes granted mid-conversation, and a simulated checkout.
- Disappear and blend in
- Dismiss to a pill, snooze for the session, or render as contextual editorial content marked sponsored.
- Seasonal resolution
- Christmas, Valentine's, Black Friday, Halloween and New Year re-skin palette, tag and decoration.
- Flight awareness and alerts
- The run period is read against the retail calendar and raises alerts — ending soon, ended, or a season worth remixing into.
- Reach and targeting
- Geo markets, age bands, gender, languages, devices and interests, with an estimated reach readout.
- Live feedback and ratings
- Every interaction streams to the advertiser view; a star rating with a reason closes the loop.
- Light and dark modes
- The whole studio themes from one token set and follows the system preference.
Measurement roadmap
Insights currently reports impressions, expands, chat turns, discounts, conversions, dismissals, ratings and segment breakdowns. These are the standard advertiser metrics still to add.
- Reach and frequency
- Impressions are counted today, but unique reach and average frequency per person need a stable visitor identifier and a frequency cap tied to it.
- Clicks and CTR
- Expands are tracked; a first-class click event plus CTR (clicks ÷ impressions) belongs alongside it, split by rendition and placement.
- Engagement rate
- One blended metric across expands, chat turns, media remixes, video completion and dwell — the ad's real engagement signal.
- Conversion rate
- Conversions are counted; the rate against clicks and against impressions is not yet surfaced.
- Cost inputs
- Nothing in the demo knows what a campaign costs. Budget, bid and spend entry is the prerequisite for every money metric below.
- CPC, CPM, CPA
- Cost per click, per thousand impressions and per acquisition, computed from spend against the event stream.
- ROAS
- Revenue per dollar spent, using checkout value net of the discounts the assistant granted — the number an advertiser actually renews on.
- Attribution window
- A defined view/click window so a conversion is credited to the right impression rather than the last touch in-session.
- Viewable impressions
- An impression should only count once the unit is actually in view for a minimum duration, not when it renders.
- Export and reporting
- CSV export, date-range comparison and a scheduled advertiser summary.
Platform roadmap — initial phase
- Cross-site delivery SDK
- A publisher <script> tag, sandboxed iframe, size negotiation with the host page, and ad-blocker realities. The single biggest engineering item.
- Real payments
- A payment provider, PCI scope, refunds, tax and merchant onboarding. In-ad checkout also needs a trust story before shoppers will use it.
- Persistent storage
- Campaigns, alerts and analytics live in memory for this demo; a real platform stores them per advertiser account.
- Video creative
- AI video generation is real but slow and expensive per asset. Practical route: pre-generate a small library per campaign.
- Identity and targeting
- Third-party cookies are gone. Contextual targeting from page content is the workable path and fits this format well.
- Fraud, viewability, brand safety
- Bot filtering, viewable-impression measurement, and rules for which pages a creative may appear on.
- Prompt-injection defence
- An assistant that can grant discounts is an attack surface. Server-side ceilings, code issuance and rate limiting must never be trusted to the model.
Platform roadmap — final phase
- Multi-armed bandit that shifts traffic toward the winning variant automatically
- Auto-rewrite the creative when engagement drops, rather than waiting for the advertiser
- Page-context reading so the ad matches the article's subject and tone in real time
- Frequency capping and a 'never show me this again' preference that actually persists
- Voice mode — the assistant speaks and listens instead of typing
- Per-shopper price sensitivity, with the discount ceiling set by predicted margin
- Handoff to a human seller when the assistant hits the edge of what it knows
- Publisher-side revenue view alongside the advertiser view
- Accessibility as a feature: full keyboard path, reduced-motion variant, screen-reader script
- Calendar intelligence: regional holidays and retail moments per market, auto-scheduled remixes
- Mid-flight season handover — queue a themed variant that swaps itself in on the right date
- Proactive AI helpers across the studio: nudges on weak copy, missing facts, unused renditions
- Voice-first authoring: dictate a brief, then refine the ad by talking to it
- Per-rendition performance — animated versus video weight, completion and cost per placement
- Persistent alerts with email and push delivery and a read/unread history
- Scheduled alert sweeps so an expiring flight is caught even when nobody has the studio open

