Ginkgo

Helping BID directors turn fragmented data into outreach plans

I led the design of an AI-assisted reporting workflow for a GIS and CRM platform, guiding users from raw neighborhood data to a decision-ready plan.

AI Product DesignGuided WorkflowsUser ResearchProduct Strategy
Ginkgo AI report dashboard on a laptop
Role
UX/Product designer
Team
Product owner (founder), developer
Company
Ginkgo, GIS + CRM platform
Timeline
March – August 2025

Outcomes

Prototyped
AI-assisted reporting that layers insights over raw data
Validated
AI reporting as a premium-tier feature in user testing
Clarified
Guided report setup that reduced setup confusion for newer users
01 · Context
The product

Ginkgo is an app that integrates geographic information systems (GIS) with CRM, enriching neighborhood data and generating reports to guide decision-making for business improvement district (BID) staff.

What I owned

I led the design of an AI-assisted reporting workflow to help BID directors turn fragmented data into actionable outreach plans, from research and framing through a prototype tested with users.

The brief

From UI improvements to adoption, retention, and renewal

How the brief evolved

I joined to improve the UI. Early discussions with the founder surfaced his larger goals to increase adoption, retention, and renewal, particularly among the enterprise customers he was looking to target more deliberately. Reporting was a critical feature, but users struggled to create reports consistently because the process was fragmented and unclear. That is when I saw an opportunity to increase the value of the platform.

Project goals
  • Client acquisition
  • Create a clear path from the map interface to the report creator
  • Launch a premium tier subscription option
  • Position Ginkgo as a BID tool that offers advanced decision support and intelligence
Current state

BID staff need to create actionable plans from reports

The app’s confusing UI/UX and lack of workflows present roadblocks.

Diagram of the planning tasks BID staff juggle: contacts, email, scheduling, tasks, integrations, and outreach
User challenges

What users had to do

  • Jump between maps, reports, and external files
  • Manually interpret results
  • Translate report data into plans outside the app
Confusing landing page UX

App login, then no report guidance

  1. Fragmented entry points

    BID staff didn’t know what to do after logging in.

  2. No intent framing

    The workflow was limited to on/off settings.

  3. Data without synthesis

    Report creation was confusing and basic.

Pinpointing the opportunity

Ginkgo surfaced data, but didn’t help BID directors plan outreach

BID directors could find maps, records, and reports across Ginkgo, but had to manually connect these pieces to form an outreach strategy. This created friction, slowed down decision-making, and limited the value users got from the platform.

  • Data lived across maps, records, and reports with no unifying workflow
  • Users had to export or manually synthesize information outside the product
  • The app answered “what exists” but not “what should I do next”
Solution framing

BID directors are responsible for turning data into real-world action

BID directors are often under-resourced and time-constrained. They need tools that help them move from raw information to actionable plans, especially when coordinating across stakeholders, timelines, and neighborhoods.

  • Plan and justify outreach efforts to boards, cities, and stakeholders
  • Identify priority areas, trends, and risks quickly
  • Translate complex data into clear recommendations and next steps
02 · Solutions

Proposed workflow

1
Home page
Report creation starts immediately after login.
2
Guided workflow for report setup
  • 3 criteria questions
  • 1 goal-based question
  • 2 refinement questions
3
AI report
  • AI insights and suggestions
  • Raw data infographics
  • AI query field
Key additions that structure intent and enable AI insights and interactivity in the report
Guided workflow

Login, then report setup immediately

  1. Intent framing

    Eliminated report-creation guesswork in favor of guided prompts with a defined report goal. Increased clarity and set up the AI insights.

AI report

Report with AI integration

  1. AI-assisted report

    Raw data output enriched with AI insights above it, and a query field below.

Workflow summary

From guided setup to an AI-enriched report

  1. Report setup

    A Mad Lib style prompt guides users through previously complicated menus (screens 1–4).

  2. Report output

    AI integration lets output and insights be regenerated on the fly (screen 5).

03 · Design decisions

Consolidating the navigation into a clearer entry point

The distributed and unlabeled navigation was the first thing I addressed, because confusion about where to go was happening before users even attempted a task. I consolidated the top and left navigation and added labels, simplifying the UI controls to make the product easier to use.

Before: distributed, unlabeled navigation

Controls were split between the top bar and a left panel, with no labels to say where to begin.

After: one consolidated, labeled navigation

Top and left navigation merged into a single labeled menu, so the next step is clear before a task begins.

Structuring the workflow: drop-downs or a guided sentence

To structure the workflow, I considered multiple drop-downs before choosing a fill-in-the-blanks sentence (Mad Lib).

Multiple drop-downs

An easy, solid sequence to follow, but a less clear way to define user objectives.

Guided Mad Lib sentence (chosen)
  • A clearer way to define user objectives
  • Aligns with the natural language used in AI prompts
  • Maps user intent to the existing system’s capabilities

Open AI prompt or guided sentence

  1. Open AI prompt

    I also considered using AI at the beginning of the workflow, instead of the Mad Lib. But the report setup controls on the map are complex and ambiguous, and an open prompt at the start could reintroduce the same systems problem we were trying to solve. Without structure at the entry point, the rest of the connected workflow could break down. The founder and I disagreed on this. He was interested in letting users interact with the AI from the beginning, make mistakes, and course-correct, the way most AI works. I implemented the Mad Lib and planned to test it with users later.

Technical challenge

Mapping user intent to the data model

To implement the Mad Lib, the system needed to retrieve data from multiple sources (client-provided data, location-specific data, and public databases) so the report output felt cohesive rather than fragmented.

Diagram of a report built from client data, location data, and public data

I mapped how the existing controls in the left, top, and right navigation related to each other and to the underlying data model. I also created a new report goal drop-down to target user intent, which helps the AI process raw data into insights.

The Mad Lib wasn’t just a UX pattern. It created a unified front-end structure over a fragmented back-end, without touching the underlying code.

Defining where to start with AI

  1. Founder request

    The founder asked me to translate the long-term vision into concrete milestones and define where we should start.

  2. My solution

    Created a less expansive guided workflow with a Mad Lib experience, rather than an open prompt, which required a smaller investment.

  1. Open, conversational AI from the start

    A traditional inline chat style, akin to ChatGPT, available from the beginning of the report-building process. But it adds more steps and allows more mistakes, which could slow down report creation.

Next I explored where AI should sit in the report experience: in a side panel next to the report, or within a full-page report.

AI side-by-side
AI within the report (chosen)

Side-by-side was OK, but less direct. I chose AI within the report for a stronger connection to the raw data.

Where AI suggestions live, and what they look like

I also considered how proactive AI suggestions would appear and be most useful to users, first where they would live, then what format they would take.

Dedicated panel

A persistent, context-sensitive panel. Ginkgo has many features, so it could support onboarding and help as well as report creation.

Shared panel with map

Keeps the right panel but shares space with a map card, so new users creating reports don’t feel separated from that experience. The number of suggestions is limited by panel height, with a search bar for previous queries.

Within prompt setup

Keeps suggestion cards in the prompt area, where they are most commonly expected.

I finalized the UI with AI only within the report, then explored suggestion formats near the chat field. I chose one row of pill buttons: the lower profile and proximity to the chat field keep the interaction tightly associated with the report without competing with it.

Testing

Testing Mad Lib combinations shaped the system logic

Not every combination of inputs could produce a perfect AI output, so I designed a validation layer to catch invalid combinations before the user generates a report.

Error handling in the prompt setup

When a user enters a prompt that doesn’t conform to what the AI can successfully parse into a report, we highlight the drop-down that needs to be changed.

Building error handling and validation: Since users could mix different combinations of variables, I had to ensure invalid combinations were caught gracefully.

Mad Lib testing: prompt combinations and what each produced.

I tested Mad Lib combinations to see how the system logic could recognize and handle report goals. In some cases the AI produced results that were incomplete or not applicable. That showed what information we needed to feed it to correct each scenario and keep outputs relevant to user objectives.

Solution hurdles

Engineering constraints shaped how the solution could be delivered

The concept required restructuring report creation and layering AI into the output, but major architectural changes weren’t possible. So the design strategy had to work within existing systems while still creating enough customer value to justify a premium tier.

Constraints
  • Engineering bandwidth was limited for custom development
  • The Omni reporting plugin defined how reporting could be extended
Workarounds
  • Extended functionality within Omni instead of replacing it
  • Leveraged AI APIs to deliver insights without altering core architecture
  • Positioned AI-assisted reporting as a premium feature to support integration costs
04 · Outcome
User interviews

The prototype revealed two distinct user segments

I interviewed users about their work roles, tasks, and app needs, along with a Q&A about my prototype. The feedback clarified where the strongest value exists.

Newer users
  • Valued guided workflows that helped them get started
  • Found the AI-generated report helpful for surfacing insights and suggested next steps
Experienced users
  • Preferred existing workflows and saved queries
  • Used AI selectively depending on task, role, and time pressure
What I learned

Guided setup and AI each contributed value

Key insights
  • The Mad Lib helps structure intent, especially for newer users or situations with low confidence or limited onboarding
  • The AI-driven report delivers the highest value for speed, synthesis, and decision support under ambiguity or scale
Implication

This validated positioning the guided workflow as an entry point, and the AI-driven report features as a premium capability focused on insight generation and decision making. Premium value wouldn’t be universal. It would concentrate on users in lower-confidence situations or dealing with scale and ambiguity.

What the prototype delivered

The prototype delivered core functionality and validated the business direction

The prototype enabled structured report creation and layering of AI insights in a way that worked within technical constraints. This allowed testing of real user workflows, evaluation of pricing strategy, and a foundation for future product expansion.

  • Introduced AI-assisted reporting to generate insights alongside raw data
  • Validated positioning AI-assisted reporting as a premium-tier capability
Reflection

Meeting the product where it was

I worked with the existing product rather than rebuilding it, and adapted the design as product objectives shifted.

What worked
  • AI integration landed because it was already available as an API within the existing Omni extension
  • A simpler, consolidated UI gave users a path through features they already had access to
How the design adapted
  • Objectives were redefined during stakeholder check-ins, so I rewrote user stories to match
  • Moved from a wizard workflow to a hybrid AI-wizard workflow
Impact

The solution improved user workflows and differentiated the product

The updated reporting experience reduced friction for new users, accelerated decision-making for experienced ones, and positioned Ginkgo as more than a data repository. These improvements strengthened user perception and created strategic room for premium features.

  • Guided workflows clarified how to create reports and reduced setup confusion
  • AI-assisted output enabled faster synthesis and reporting compared to manual methods
  • Strategic differentiation moved the platform toward insight-driven decision support
Ginkgo

Helping BID directors turn fragmented data into outreach plans

I led the design of an AI-assisted reporting workflow for a GIS and CRM platform, guiding users from raw neighborhood data to a decision-ready plan.

AI Product DesignGuided WorkflowsUser ResearchProduct Strategy
Ginkgo AI report dashboard on a laptop
Role
UX/Product designer
Team
Product owner (founder), developer
Company
Ginkgo, GIS + CRM platform
Timeline
March – August 2025

Outcomes

Prototyped
AI-assisted reporting that layers insights over raw data
Validated
AI reporting as a premium-tier feature in user testing
Clarified
Guided report setup that reduced setup confusion for newer users
01 · Context
The product

Ginkgo is an app that integrates geographic information systems (GIS) with CRM, enriching neighborhood data and generating reports to guide decision-making for business improvement district (BID) staff.

What I owned

I led the design of an AI-assisted reporting workflow to help BID directors turn fragmented data into actionable outreach plans, from research and framing through a prototype tested with users.

The brief

From UI improvements to adoption, retention, and renewal

How the brief evolved

I joined to improve the UI. Early discussions with the founder surfaced his larger goals to increase adoption, retention, and renewal, particularly among the enterprise customers he was looking to target more deliberately. Reporting was a critical feature, but users struggled to create reports consistently because the process was fragmented and unclear. That is when I saw an opportunity to increase the value of the platform.

Project goals
  • Client acquisition
  • Create a clear path from the map interface to the report creator
  • Launch a premium tier subscription option
  • Position Ginkgo as a BID tool that offers advanced decision support and intelligence
Current state

BID staff need to create actionable plans from reports

The app’s confusing UI/UX and lack of workflows present roadblocks.

Diagram of the planning tasks BID staff juggle: contacts, email, scheduling, tasks, integrations, and outreach
User challenges

What users had to do

  • Jump between maps, reports, and external files
  • Manually interpret results
  • Translate report data into plans outside the app
Confusing landing page UX

App login, then no report guidance

  1. Fragmented entry points

    BID staff didn’t know what to do after logging in.

  2. No intent framing

    The workflow was limited to on/off settings.

  3. Data without synthesis

    Report creation was confusing and basic.

Pinpointing the opportunity

Ginkgo surfaced data, but didn’t help BID directors plan outreach

BID directors could find maps, records, and reports across Ginkgo, but had to manually connect these pieces to form an outreach strategy. This created friction, slowed down decision-making, and limited the value users got from the platform.

  • Data lived across maps, records, and reports with no unifying workflow
  • Users had to export or manually synthesize information outside the product
  • The app answered “what exists” but not “what should I do next”
Solution framing

BID directors are responsible for turning data into real-world action

BID directors are often under-resourced and time-constrained. They need tools that help them move from raw information to actionable plans, especially when coordinating across stakeholders, timelines, and neighborhoods.

  • Plan and justify outreach efforts to boards, cities, and stakeholders
  • Identify priority areas, trends, and risks quickly
  • Translate complex data into clear recommendations and next steps
02 · Solutions

Proposed workflow

1
Home page
Report creation starts immediately after login.
2
Guided workflow for report setup
  • 3 criteria questions
  • 1 goal-based question
  • 2 refinement questions
3
AI report
  • AI insights and suggestions
  • Raw data infographics
  • AI query field
Key additions that structure intent and enable AI insights and interactivity in the report
Guided workflow

Login, then report setup immediately

  1. Intent framing

    Eliminated report-creation guesswork in favor of guided prompts with a defined report goal. Increased clarity and set up the AI insights.

AI report

Report with AI integration

  1. AI-assisted report

    Raw data output enriched with AI insights above it, and a query field below.

Workflow summary

From guided setup to an AI-enriched report

  1. Report setup

    A Mad Lib style prompt guides users through previously complicated menus (screens 1–4).

  2. Report output

    AI integration lets output and insights be regenerated on the fly (screen 5).

03 · Design decisions

Consolidating the navigation into a clearer entry point

The distributed and unlabeled navigation was the first thing I addressed, because confusion about where to go was happening before users even attempted a task. I consolidated the top and left navigation and added labels, simplifying the UI controls to make the product easier to use.

Before: distributed, unlabeled navigation

Controls were split between the top bar and a left panel, with no labels to say where to begin.

After: one consolidated, labeled navigation

Top and left navigation merged into a single labeled menu, so the next step is clear before a task begins.

Structuring the workflow: drop-downs or a guided sentence

To structure the workflow, I considered multiple drop-downs before choosing a fill-in-the-blanks sentence (Mad Lib).

Multiple drop-downs

An easy, solid sequence to follow, but a less clear way to define user objectives.

Guided Mad Lib sentence (chosen)
  • A clearer way to define user objectives
  • Aligns with the natural language used in AI prompts
  • Maps user intent to the existing system’s capabilities

Open AI prompt or guided sentence

  1. Open AI prompt

    I also considered using AI at the beginning of the workflow, instead of the Mad Lib. But the report setup controls on the map are complex and ambiguous, and an open prompt at the start could reintroduce the same systems problem we were trying to solve. Without structure at the entry point, the rest of the connected workflow could break down. The founder and I disagreed on this. He was interested in letting users interact with the AI from the beginning, make mistakes, and course-correct, the way most AI works. I implemented the Mad Lib and planned to test it with users later.

Technical challenge

Mapping user intent to the data model

To implement the Mad Lib, the system needed to retrieve data from multiple sources (client-provided data, location-specific data, and public databases) so the report output felt cohesive rather than fragmented.

Diagram of a report built from client data, location data, and public data

I mapped how the existing controls in the left, top, and right navigation related to each other and to the underlying data model. I also created a new report goal drop-down to target user intent, which helps the AI process raw data into insights.

The Mad Lib wasn’t just a UX pattern. It created a unified front-end structure over a fragmented back-end, without touching the underlying code.

Defining where to start with AI

  1. Founder request

    The founder asked me to translate the long-term vision into concrete milestones and define where we should start.

  2. My solution

    Created a less expansive guided workflow with a Mad Lib experience, rather than an open prompt, which required a smaller investment.

  1. Open, conversational AI from the start

    A traditional inline chat style, akin to ChatGPT, available from the beginning of the report-building process. But it adds more steps and allows more mistakes, which could slow down report creation.

Next I explored where AI should sit in the report experience: in a side panel next to the report, or within a full-page report.

AI side-by-side
AI within the report (chosen)

Side-by-side was OK, but less direct. I chose AI within the report for a stronger connection to the raw data.

Where AI suggestions live, and what they look like

I also considered how proactive AI suggestions would appear and be most useful to users, first where they would live, then what format they would take.

Dedicated panel

A persistent, context-sensitive panel. Ginkgo has many features, so it could support onboarding and help as well as report creation.

Shared panel with map

Keeps the right panel but shares space with a map card, so new users creating reports don’t feel separated from that experience. The number of suggestions is limited by panel height, with a search bar for previous queries.

Within prompt setup

Keeps suggestion cards in the prompt area, where they are most commonly expected.

I finalized the UI with AI only within the report, then explored suggestion formats near the chat field. I chose one row of pill buttons: the lower profile and proximity to the chat field keep the interaction tightly associated with the report without competing with it.

Testing

Testing Mad Lib combinations shaped the system logic

Not every combination of inputs could produce a perfect AI output, so I designed a validation layer to catch invalid combinations before the user generates a report.

Error handling in the prompt setup

When a user enters a prompt that doesn’t conform to what the AI can successfully parse into a report, we highlight the drop-down that needs to be changed.

Building error handling and validation: Since users could mix different combinations of variables, I had to ensure invalid combinations were caught gracefully.

Mad Lib testing: prompt combinations and what each produced.

I tested Mad Lib combinations to see how the system logic could recognize and handle report goals. In some cases the AI produced results that were incomplete or not applicable. That showed what information we needed to feed it to correct each scenario and keep outputs relevant to user objectives.

Solution hurdles

Engineering constraints shaped how the solution could be delivered

The concept required restructuring report creation and layering AI into the output, but major architectural changes weren’t possible. So the design strategy had to work within existing systems while still creating enough customer value to justify a premium tier.

Constraints
  • Engineering bandwidth was limited for custom development
  • The Omni reporting plugin defined how reporting could be extended
Workarounds
  • Extended functionality within Omni instead of replacing it
  • Leveraged AI APIs to deliver insights without altering core architecture
  • Positioned AI-assisted reporting as a premium feature to support integration costs
04 · Outcome
User interviews

The prototype revealed two distinct user segments

I interviewed users about their work roles, tasks, and app needs, along with a Q&A about my prototype. The feedback clarified where the strongest value exists.

Newer users
  • Valued guided workflows that helped them get started
  • Found the AI-generated report helpful for surfacing insights and suggested next steps
Experienced users
  • Preferred existing workflows and saved queries
  • Used AI selectively depending on task, role, and time pressure
What I learned

Guided setup and AI each contributed value

Key insights
  • The Mad Lib helps structure intent, especially for newer users or situations with low confidence or limited onboarding
  • The AI-driven report delivers the highest value for speed, synthesis, and decision support under ambiguity or scale
Implication

This validated positioning the guided workflow as an entry point, and the AI-driven report features as a premium capability focused on insight generation and decision making. Premium value wouldn’t be universal. It would concentrate on users in lower-confidence situations or dealing with scale and ambiguity.

What the prototype delivered

The prototype delivered core functionality and validated the business direction

The prototype enabled structured report creation and layering of AI insights in a way that worked within technical constraints. This allowed testing of real user workflows, evaluation of pricing strategy, and a foundation for future product expansion.

  • Introduced AI-assisted reporting to generate insights alongside raw data
  • Validated positioning AI-assisted reporting as a premium-tier capability
Reflection

Meeting the product where it was

I worked with the existing product rather than rebuilding it, and adapted the design as product objectives shifted.

What worked
  • AI integration landed because it was already available as an API within the existing Omni extension
  • A simpler, consolidated UI gave users a path through features they already had access to
How the design adapted
  • Objectives were redefined during stakeholder check-ins, so I rewrote user stories to match
  • Moved from a wizard workflow to a hybrid AI-wizard workflow
Impact

The solution improved user workflows and differentiated the product

The updated reporting experience reduced friction for new users, accelerated decision-making for experienced ones, and positioned Ginkgo as more than a data repository. These improvements strengthened user perception and created strategic room for premium features.

  • Guided workflows clarified how to create reports and reduced setup confusion
  • AI-assisted output enabled faster synthesis and reporting compared to manual methods
  • Strategic differentiation moved the platform toward insight-driven decision support