JENI

Jeni — Marketplace Intelligence(Still building at the moment)

Turning fragmented marketplace signals into decisions Bravo can act on.

TL;DR

Bravo is a payments and dining rewards marketplace connecting diners with 500+ restaurant partners across Metro Vancouver. Despite years of customer, merchant, and transaction data, the team lacked a way to understand what those relationships meant—or where to act.

With no PM, predefined requirements, or predetermined solution, I defined, designed, and built Jeni end to end: an AI-powered intelligence system that turns Bravo's fragmented marketplace signals into actionable opportunities.

Role

Product Designer & Builder

Year

2026

Collaborators

CEO · 1 Backend Engineer

Built with

Claude · ChatGPT · AWS · Figma

Video

The starting point

I wasn’t given a product to design. I was given an ambition.

Bravo’s CEO and marketing team had a recurring problem. By the time a new restaurant opened or an existing one was visibly struggling, the team was often already late to the opportunity.

The initial ask was loose: could AI help Bravo spot these signals earlier?

I turned that idea into the first version of Jeni, a market intelligence tool that monitored public signals around restaurant openings, closures, and other merchant activity.

Jeni's monitoring overview: a Needs Attention summary bar over three columns of detected restaurants — New / Opening, Closed / At Risk, and For Sale — with counts for tracked restaurants and signals, and a list of source collectors below. — Jeni V1 — Bringing restaurant openings, closures, and other public signals into one monitoring view.

That first version changed the scope of my role.

After seeing me turn a loosely defined AI idea into a working product, the CEO gave me a much broader mandate: expand Jeni across Bravo.

Monitoring would remain one capability within Jeni, but the larger product was undefined. There was no predefined problem, workflow, or solution. I had to determine where Jeni could actually create value inside the business.

Jeni's VCH report: counts for active closures, reopened, and unresolved, above a table of health closures listing each restaurant, city, closure period, reason, and when it was last checked — with a separate collapsed group for unresolved listings. — One collector in detail — health closure orders, with unresolved merchant matches preserved for review rather than guessed.

Finding the product

The next brief was one sentence: make Jeni useful to Bravo.

After seeing V1, Bravo’s CEO asked me to take Jeni further into the business. There was no feature list, target workflow, or predefined problem to solve.

I started with the systems Bravo already used. The existing merchant portal tracked credit balances and transaction history, while customer records captured balances, rewards, referrals, and dining activity. The data was there. What was missing was a way to interpret it across the marketplace.

The ledger showed credit remaining, but not when a merchant was becoming a problem.

A merchant credit table: rows of previous credit, funding amount, credit acquired, resulting Bravo credit balance, previous and new rate, and funding method. — Bravo’s existing merchant portal exposed balances and transactions, but recognizing urgency still required someone to interpret them manually.

A transaction table: rows of QR code payments showing amount, channel, status, balance used, points used, and coupon used. — Bravo’s existing merchant portal exposed balances and transactions, but recognizing urgency still required someone to interpret them manually.

Understanding both sides still didn’t tell Bravo what to do.

A merchant could need help and a diner could be a strong match, but that still didn’t mean Bravo should intervene. The decision also depended on whether the expected value justified the cost.

Merchant Need

Where does the marketplace need help?

Diner Opportunity

Who could realistically respond?

Bravo Economics

Would acting create enough value?

This became the product thesis for Jeni.

Jeni would connect merchant need, diner opportunity, and Bravo’s economics to identify where intervention was actually worth considering.

Building Jeni

Turning the product thesis into a decision workflow.

Jeni's dashboard: Today's Executive Brief with merchant counts — 514 of 574 need attention, split into high opportunity, needs credit, at risk, and credit outstanding — and user counts flagged for a win-back play, above a Metro Vancouver heatmap of merchant locations and user activity.

The thesis gave Jeni three things to reason about. The product still had to make that reasoning useful to the people running Bravo.

I designed Jeni around a simple progression: surface what needs attention, explain why it matters, connect the relevant merchant and diner signals, and make the next decision easier to evaluate.

01 / See What Needs Attention

Start with the exceptions, not the database.

Bravo already had the data. The problem was knowing where to look first.

I designed Jeni to surface the merchants, diners, and opportunities showing meaningful change, then prioritize them by urgency and potential business value.

Jeni's Campaign opportunities panel — the best set that can run at once: if all eight run, $134,159 of spend across 8 campaigns for +$8k net margin; the top three ranked by net margin — spend threshold at +$5k, visit frequency at +$1k, happy hour at +$781 — with five more in the runnable set. — Instead of another dashboard to explore, Jeni opened with what Bravo could act on.

Three health panels from Jeni's dashboard. Merchant health: 426 need attention, 387 with upside, 198 sitting on credit. Customer health: 3,894 need attention, 447 new customers, 81 merchants worth promoting. Business impact: 0 visits Bravo can evidence, $7,951,323 of credit used, $1,136,103 Bravo margin realised as credit burns. — Instead of another dashboard to explore, Jeni opened with what Bravo could act on.

02 / Understand Why

Show the evidence behind the signal.

A flag only mattered if the team could understand what was driving it.

I designed Jeni to connect each state back to the underlying behavior, such as credit burn, transaction activity, visit patterns, and recent change, so the team could judge whether an opportunity was real before acting.

Jeni's Merchant Intelligence page: the credit book — used, outstanding, Bravo margin, average cashback — above a Transaction flow panel with a 30-day window, credit used and average bill versus the previous period, and a stacked bar chart of credit used per day by cuisine. — Jeni first established the network baseline: how credit was being used, where transaction volume was moving, and what normal looked like across the marketplace.

From there, the team could drill into a flagged merchant and see what was actually driving the signal.

One merchant's Credit panel in Jeni: 14 days of runway emptying 5 October, burn speed $152 a day marked above the healthy band, $2,274 outstanding, a $2,952 batch from $4,000 cash; a breakdown of where the batch went via balance and bonus; and a 60-day balance line that saw-tooths through two $4k top-ups, with a dashed projection running to empty and a footer naming the assumption at 75% confidence. — For Happy Day Cafe Kingsway, a static credit balance became a time-sensitive signal once Jeni showed burn velocity, replenishment history, and runway.

But urgency alone wasn’t enough. Jeni also needed to understand the business behind the number.

Credit showed when attention was needed. Operational behavior helped explain what kind of opportunity it was.

One merchant's Operations panel in Jeni: customer mix — only 36% of visits come back, tagged Mostly new — beside service mix split across morning, lunch, and dinner, with morning flagged n=1; a row of 491 visits over 60 days, $21.50 average spend, 200 lifetime visits, and a 4.79 rating; and a Traffic bar chart of visits by day with the busiest day, Saturday 29 August, highlighted. — Visit frequency, customer mix, daypart behavior, and spend patterns gave the financial signal operational context.

03 / Find Who Can Move It

Connect merchant need with diner opportunity.

Understanding the merchant explained the problem, but not who could change the outcome.

I designed the diner side of Jeni around behavioral signals such as inactivity, top-up activity, customer value, visit history, and merchant affinity. This helped the system identify not just an audience, but the diners most relevant to a specific merchant need.

Jeni's User Intelligence page: 9,041 of 43,320 users flagged at score 40 or above, split into 6,949 at risk, 549 top-up ready, and 884 high value; a Top-ups panel for the last 30 days — $288,355 topped up, $21,933 bonus granted, $0.08 cost per dollar loaded, $50 busiest rung — and a bar chart of balance loaded versus bonus granted by day. — Instead of treating diners as a list of accounts, Jeni organized them by behavioral state: who was drifting away, showing top up signal or worth retaining.

But behavioral state alone didn’t make a diner relevant.

Jeni also needed evidence that the diner and merchant actually belonged together.

Jeni's Diners who also visit panel for Happy Day Cafe Richmond: three merchants ranked by lift — Wasan Cafe at 2.5×, Grill King at 2.4×, AJEA Noodle at 1.0× — with shared diners and the overlap the other way, a note that the list is ranked by lift rather than raw overlap, and a prompt to plan a cross-merchant campaign. — Cross-merchant behavior exposed relationships that raw audience overlap would miss. I used lift rather than shared-diner count so large merchants wouldn’t automatically dominate the recommendation.

Relevance could then become a specific play.

Once diner behavior and merchant context were connected, Jeni could move from “who looks interesting?” to “who could change this outcome, and how?”

Eight rows of Jeni's Ranked users table: each diner with their average spend or inactivity, a High value or Inactive signal, a suggested action of VIP perk or win-back bonus, net to Bravo from +$874 down to +$684, and network value with diners referred. A hover tooltip on one action reads: Card boost — spend $210 at Bar Bravo, $15 in balance. — Each diner carried the signal, suggested action, and expected value forward, preserving the reasoning behind the recommendation.

04 / Make the Economics Visible

A useful intervention still had to be worth running.

Connecting the right diner to the right merchant still wasn’t enough. Every intervention had a cost to Bravo, and more activity did not necessarily mean more value.

I made the economics part of the recommendation itself, so the team could evaluate an opportunity against credit cost, expected margin, and network value before deciding to act.

The same action could look very different once Bravo’s economics entered the decision.

A crop of Jeni's ranked users table: seven rows with their signal — churn risk, referral potential, inactive, high value — a suggested action of re-engagement offer, referral invite, win-back bonus, or VIP perk, net to Bravo sorted from +$667 down to +$620, and network value with diners referred — including one referral invite at +$655 net whose network value is −$21. — I surfaced expected value beside the recommended action, making the tradeoff visible before the team committed budget.

I didn’t reduce value to a single number. Direct value to Bravo and broader network value could point in different directions, so I kept them visible separately rather than hiding the tradeoff inside one score.

Economics also changed prioritization.

Once Jeni could estimate the value of individual plays, the same logic could compare opportunities across the business.

Jeni's Campaign Intelligence page: 8 of 14 campaign levers can run at once, with +$8k net if all run; below, seven diner cohorts ranked by movable credit — Lapsing regulars at $458k, Big-basket diners at $198k, Lapsed regulars at $183k, Card-first diners at $134k — each with its definition, size, a one-line rationale, and a suggested play. — Opportunities were ranked by estimated net margin rather than activity alone, showing both the spend required and the expected return.

I deliberately showed expected upside together with the spend required to create it. The goal wasn’t to make Jeni sound confident; it was to give Bravo enough context to judge whether the recommendation was worth taking.

The decision model was now complete.

Merchant Need × Diner Opportunity × Bravo Economics → Intervention

05 / Turn the Decision Into Action

A recommendation only mattered if the team could do something with it.

By this point, Jeni could identify an opportunity, explain the evidence behind it, connect the relevant diners, and make the economics visible.

The final step was turning that reasoning into a specific next move the team could evaluate and act on.

The same framework could support different kinds of decisions, from creating demand to replenishing credit to maintaining a healthy state.

A Recommended Action card in Jeni: 'Promote to nearby cafe fans to accelerate burn' with 66% confidence, split into Why now ($2,959 sitting unused, burning $0 a day), The move (promote to nearby fans, run a win-back push), and If it lands (+34 visits in 30 days), with Create campaign and Show the working buttons. — The same framework supported different decisions: create demand, replenish credit, or simply keep monitoring. Each recommendation explained why it mattered, what to do next, and what could change.

A Recommended Action card in Jeni: 'Sales follow-up — credit is nearly gone; buy the next tranche now' with 66% confidence, split into Why now (3 days of runway left, burning $203 a day against $545 outstanding), The move (buy more credit, offer a return-visit coupon), and If it lands ($545 left, 77 visits a month), with Open the credit book and Show the working buttons. — The same framework supported different decisions: create demand, replenish credit, or simply keep monitoring. Each recommendation explained why it mattered, what to do next, and what could change.

I structured each recommendation around three questions: why does this matter now, what should Bravo consider doing, and what could change if it works.

The recommendation stayed close to the evidence, with the underlying reasoning available for the team to inspect before acting.

Actionable did not mean automatic. The team still controlled whether and how to intervene.

The Outcome

From a side project to the roadmap.

Jeni started as an open-ended experiment around restaurant intelligence. It has since become a live internal product for understanding Bravo’s marketplace and identifying where the team can act.

The product is now part of the CEO’s roadmap and has been included in the investor pitch deck as part of Bravo’s evolving intelligence capabilities.

The project also changed my role.

I entered Jeni as a product designer with a loosely defined AI opportunity. I ended up defining the product strategy, designing the decision model, and building the working system myself.

More importantly, it changed how I think about product design: when the path is unclear, designing the interface is only one part of the job. Sometimes the larger responsibility is defining what should exist in the first place.