@felix · retention · live
Felix
The loyalty layer that listens, then rewards generously.
Variable reinforcement Whitelabel loyalty Built for LTV Companion to Sid

What it is. Felix is a conversational AI module that appears at the moment a member redeems points inside any loyalty or rewards programme. It opens with a question calibrated to that member's history, holds a short warm dialogue, and closes with a personalised reward — sometimes offered directly, sometimes revealed through a tap-to-unveil mechanic.

How it's integrated. Felix drops into the partner's existing app or web experience as a conversational overlay at the redemption moment. It reads member history through an API (balance, threshold, cadence, category pattern), runs the Syntacy Analyzer Engine on each reply, and writes structured preference data back to the partner's CRM. Brand voice, reward economics, and appearance frequency are all partner-controlled.

Why we built it. Loyalty programmes compete on cashback until they stop being programmes at all. Felix moves the battleground from rate to relationship. Every conversation yields two compounding assets: behavioural signal that increases the accuracy of future curation, and a moment of genuine recognition that materially extends member lifetime value. The chat isn't the cost. The chat is the product.

Live demo · try a scenario
Felix session · dialogic
Felix is available this time. Not every redemption will trigger a chat — availability is randomly allocated.
Felix brain · real-time

Conversation stage

Composite score Hidden

0.50 / 1.00
Reading…
Insight depth
Tone

Signals

— awaiting first reply —

Zero-party data

    Felix reward
    Conversation in progress
    How Felix behaves

    A short, deliberate sequence. Each step changes how the next one feels.

    Felix is not a list of features. It's a single movement, performed in three parts. The chat opens with a question that already knows the member. It listens, scores, and responds in real time. It rewards generously, withdraws gracefully, and never repeats itself the same way twice. Here's what each part is doing, and why.

    First
    It opens a real conversation.

    Most loyalty bots interrogate. Felix talks. Each question is calibrated to what the brand already knows about the member, and each response acknowledges what the member just said before moving on.

    The opener is informed.

    Felix reads the member's history before saying a word. Balance, redemption threshold, cadence, category pattern. The first question lands as if a host had read the file — never generic, always specific to tonight.

    The follow-ups acknowledge.

    Every reply produces a brief reflection before the next question. Felix quotes the member back to themselves. That single move — being heard before being asked again — is what separates dialogue from interrogation.

    Then
    It listens, with consequence.

    While the conversation flows, two parameters score in the background. Insight depth: how much the member is willing to share. Tone: how warm the language reads. The composite is hidden from the member but determines everything that comes next.

    Generosity is rewarded.

    Longer, more considered answers lift insight. Warm, cooperative language lifts tone. Short or neutral replies hold steady. The member who shares more, gracefully, ends up with more — and they never know that's why.

    Rudeness is recognised.

    Hostile language lowers tone and is flagged separately. Past a threshold, Felix politely ends the chat. "Fair enough — enjoy your evening." No reward. No passive-aggressive hold. Boundary-setting is the product, not a workaround.

    Finally
    It rewards, with surprise.

    Felix never trades reward for data. Felix gives reward as a thank-you for the conversation. The form of the reward changes — sometimes a clean menu of options, sometimes a tap-to-reveal of three covered tiles. The member can't predict which, and the small surprise of the reveal carries real behavioural weight on its own.

    Generosity matches the conversation.

    High scores unlock the deepest tier — three reward options, often gamified. Mid scores get a smaller menu. Low scores close warmly with no reward, no friction. The member always leaves on good terms, but only the generous conversations are met with generosity in return.

    Design principles

    Two principles that hold the whole thing up.

    The behaviours above describe what the member experiences. These two principles describe what makes the system durable for the brand. Without them, Felix becomes a survey bot members learn to game. With them, the mechanic compounds value over time.

    i
    Random availability

    Felix doesn't appear at every redemption. The chance is randomly allocated per session, never disclosed in advance. Members can't predict it, which means they can't game it.

    Why it matters. Predictable rewards become baseline expectations. Variable rewards stay valuable. The randomness protects reward economics for the brand and keeps the moment feeling rare for the member.
    ii
    Score hidden · data captured

    The composite score and five behavioural axes are never shown to the member. Only the conversation and the reward are visible. Underneath, a structured preference record flows back to the partner's CRM.

    Why it matters. Visible scoring turns conversation into performance. Hidden scoring lets members behave naturally, which produces more accurate signal. The asymmetry is the whole product.
    What the brand actually receives
    A single Felix conversation, unpacked.
    Below is the structured record produced by the Fallow conversation in the demo above. This is what lands in the partner's CRM when the chat closes, alongside three role-by-role views of who in the partner org uses which fields.
    conversation_id #a47f3 Fallow · 12 Apr · member_id #82914
    21 fields captured · 47 seconds elapsed
    adjacent_venue
    "Lita, Marylebone"
    Member named a venue not currently in Yonder's roster.
    taste_profile
    "chef-led Italian"
    Inferred from named venue + reasoning.
    neighbourhood_signal
    "Marylebone"
    Geographic gap, prioritised for next drop.
    occasion
    "reunion dinner"
    Triggered by friend visiting from out of town.
    redemption_trigger
    "scarcity of companion"
    Why this redemption, not a different night.
    balance_above_threshold
    +30.7%
    Member redeemed at 7,840 vs usual 6,000.
    product_wish
    "tenure recognition"
    Member-volunteered product feedback.
    lh_score · long_term
    0.85
    Strong long-term orientation. Retention-positive.
    lh_score · politeness
    0.85
    Warm tone sustained across all three turns.
    lh_score · price_focus
    0.15
    Low price focus. Not redemption-rate driven.
    composite_score
    0.81
    Priority-tier conversation. Highest reward unlocked.
    reward_chosen
    "early_drop_marylebone"
    Member opted for partner-acquisition signal over points.
    Where each field lands inside the org
    Three teams. One conversation.
    Partnerships

    Adds new venues to the roster, negotiates terms with existing ones.

    Reads
    adjacent_venue neighbourhood_signal taste_profile
    Pipeline of named target venues, prioritised by demand frequency. The Lita signal alone, repeated across 50 members, becomes the business case for opening a Marylebone partner conversation next month.
    Curation

    Decides which 15–20 experiences appear in next month's drop.

    Reads
    occasion redemption_trigger taste_profile balance_above_threshold
    Drops weighted by what members are actually using redemptions for. "Reunion dinner" + "scarcity" patterns reveal the urgency-led use case curation can specifically support with same-week availability.
    Product · Retention

    Owns lifetime value, churn, and the loyalty programme structure.

    Reads
    composite_score lh_score · long_term product_wish reward_chosen
    High-LH members flagged for retention-positive treatment. "Tenure recognition" appearing across the wish field, repeated, becomes a roadmap signal for the next programme feature — pre-validated by the members who'd use it.
    What this looks like at scale
    1 conversation
    21 fields · 1 partner gap · 1 product wish
    A single member, vivid and useful.
    100 conversations
    ~2,100 fields · 12–18 named venue signals · 6–8 product wishes
    A pattern emerges. Partnership pipeline names itself.
    1,000 conversations
    ~21,000 fields · ranked venue map · validated product roadmap
    A behavioural intelligence layer no survey could produce. Compounds monthly.
    The conversation is one minute. The asset is permanent.
    Meet the companion bot
    Meet Sid.
    Syntacy · acquisition bot
    The pair, at a glance
    Sid acquisition · first purchase · discount justified by fit
    Felix retention · repeat redemption · reward justified by insight

    If Felix is the retention half of the Syntacy stack, Sid is the acquisition half — applied where blanket voucher codes currently leak margin.

    Brands burn millions on generic 10%-off codes that hit everyone indiscriminately — the bargain-hunter who'd never return, alongside the genuinely aligned buyer whose loyalty could have been cultivated. Sid overhauls that model. At the purchase or signup moment, it runs a short qualification chat. Buyers who share context and whose signals align with the product — cultural literacy for opera tickets, food fluency for a restaurant booking, craft interest for a pottery class — unlock deeper rewards. The rest pay list price, warmly.

    Same five-axis Analyzer Engine underneath. Same dialogic register. Different job: Sid qualifies. Felix cultivates. Together they replace blanket discounting with a loyalty economy that rewards alignment rather than extraction.

    Opera → cultural signal
    Restaurant → food literacy
    Craft class → making interest
    Fitness → routine commitment
    Status
    Felix for retention. Sid for acquisition.
    Two BehaviorBots, one engine. Whitelabel for partners across cultural sector, premium hospitality, membership programmes and considered-purchase D2C. Get in touch to design-partner a deployment, or take a look at Sid to see the acquisition half of the loop.