Case study / AVANT Communications / May 2021 to Aug 2024

Sixteen tools, one deal.

Pathfinder was sixteen separate tools wearing one logo. The deal they were meant to serve lived in email, chat and a spreadsheet on someone's desktop. I found the one workflow underneath all sixteen and built the system that carries it.

Open the live system Read the story

Consolidation

16 → 5

tools consolidated into five tracked states of one deal

Research

20%

of the active advisor base interviewed before a screen was drawn

Ownership

3+ yr

May 2021 to Aug 2024, owned end to end as the only systems designer

Pathfinder design system overview running in Figma Make: left navigation listing foundations and components beside the Pathfinder 2.0 hero.

The component layer, running from code

Every destination in this story is assembled from one library. That library is live in Figma Make, so you can click through it instead of reading about it.

Open it

The problem

Sixteen tools, no shared idea of what a deal was. The real record of every pursuit lived in inboxes and spreadsheets, so the platform could not report on its own revenue.

What I did

Interviewed a fifth of the advisors, gave each tool a verdict (merge, absorb, retire or rebuild), and rebuilt the product as five states of one project on one component library.

What changed

Sixteen destinations became five. New surfaces shipped consistent on day one, and everything added since has landed inside one of those five states.

Role: Senior Design Systems Lead. The only systems designer, with three engineers and a PM.
Scope: Research, information architecture, foundations, components, theming, governance.
Stack: Figma Variables, React handoff through Dev Mode, coded build in Figma Make.
Timeline: At Avant May 2021 to Aug 2024. The coded 2.1 build shown above shipped June to August 2026.

01Problem

Sixteen front doors and no hallway

AVANT distributes technology services through independent Trusted Advisors. Pathfinder is where those advisors research, quote and close enterprise deals, with AVANT's sales engineers backing them up.

Each of the sixteen tools answered a real need, and each was defensible on its own. Together they were a tax. An advisor working one opportunity re-entered the same customer, locations and requirements, because none of the sixteen agreed on what a deal was or where it stood.

So the deal left the platform. The real record of a six-figure pursuit lived in an inbox. The platform held the data. The advisor held the deal.

The platform could not tell you the status of its own revenue. That is what got the work funded.

The original Pathfinder home screen: a search box above a four by four grid of sixteen tool icons, from Technology Navigator and Provider Portfolio to Uploaded Documents and Back Office View.

What is not on this screen

Sixteen ways to research and present. Nothing that holds the deal.

Register a deal
Contract
My deals

Every tile helps an advisor find, compare, price or hand something over. None of them is where a deal gets registered, signed or tracked. Registration and contracts happened over email.

BeforeEvery tool its own front door

Technology Navigator
Provider Portfolio
Data Center Locator
Fiber Tool
Interactive Quick Assessments
Dynamic Matrices
CableQuote
Quick Quote
Cloud Ninja Knowledge Base
UCaaS TCO Analysis
Provider Battle Cards
Provider One Pagers
Case Studies
Sample Proposals
Uploaded Documents
Back Office View

The deal itself lived out hereEmail threadsChatSpreadsheets

AfterOne tracked project, five states

1
Project InformationWhat needs me today, in five seconds
2
ConfigurationDozens of providers to a defensible shortlist
3
DiscoveryThe longest state, waiting made legible
4
RegistrationThe irreversible step, beside its reasoning
5
ContractOutlives everyone who was in the room

02Research

One complaint in twenty different accents

I interviewed a fifth of the active advisor base before drawing a screen. Each session worked from the advisor's own live deals and not a script, because workarounds only show up in a real pipeline.

The finding wasn't about any single tool. There was no through-line. The assessment lived in one app, the quote in an email, the client's answer in chat, and the actual status in the advisor's head.

If the platform can't hold the whole deal, advisors keep the real one somewhere else. That set the architecture: one tracked project moving through five states (Project Information, Configuration, Discovery, Registration, Contract), with everything attached to a state.

The research set three goals before any design started.

Out of scope: a rebrand, replacing the CRM, and component count as a scoreboard.

Goal 01

Put the deal back in the platform

Measured by the share of active deals tracked end to end in Pathfinder.

Goal 02

Cut the re-entry tax

Measured by how many times an advisor enters the same customer data on one deal.

Goal 03

Ship consistent by default

Measured by the share of new surfaces assembled from the shared library.

Advisors weren't asking for a better interface. They were asking for somewhere to put the deal.

03Teardown

Decide what each of the sixteen actually is

With the research in hand, every tool got documented the same way. Five questions, asked in the same order every time, so the sixteen could be compared against each other. An inventory is an argument, and it only works if it is complete.

01What data does it hold, and who else already needs that data?Duplicated data is the tell. If two tools keep their own copy of the customer, one of them is a view and not a product.
02What job does it do in a deal, and at which state?If the answer wasn't a state of a deal, it wasn't a destination. It was a tool used inside one.
03Who touches it, and is that their job or their workaround?There was no usage data, so this came from the interviews. Some of the most-loved tools were people compensating for a gap somewhere else.
04What breaks tomorrow if it's gone?The fastest way to find dead weight.
05Is this a product, a feature, or a report?The question that did most of the work. Most of the sixteen were features that had been promoted, and reports that had been given a login.

Sixteen tiles, six jobs.

Nobody organizes their work by tool. Sorted by the question an advisor is actually asking, the sixteen tiles collapse into six jobs, and every job already belongs to a moment in a deal.

1
What is available at this address?Data Center Locator, Fiber Tool and CableQuote asked one question of three data sets.One map with layers, inside Configuration
2
Who should I recommend?Navigator, Portfolio and Matrices were three doors into the same provider record.The comparison at the heart of Configuration
3
What will it cost?Quick Quote and the TCO calculator priced things with no deal to attach the price to.Quoting inside the project
4
What does the client need?Assessments produced the best evidence in the platform, then emailed it away.Discovery
5
What do I hand the client?Five shelves with one job, differing only by document type. Four fold together. Sample proposals give way to proposals generated from the project.One library, surfaced in context
6
Where does my deal stand?Uploaded Documents and Back Office View were the only traces of an actual deal.Project Information and Contract

Then every tool got a verdict.

Merge

Same job, different door

One job reached through different doors, collapsed into one surface. Where audiences differ, permissions do the separating instead of navigation.

Absorb

Feature, not destination

Real value at the wrong altitude. It became a panel, a filter, or a step inside the state where the advisor already was.

Retire

Nothing breaks without it

Solving a problem the new workflow doesn't have, or serving so few people that upkeep outweighed use. Retired with a documented migration path, never silently.

Rebuild

Right job, wrong century

Load-bearing to the business and beyond patching, so it was rebuilt on the new foundations instead of re-skinned on the old ones.

Sixteen tools, four verdicts, five destinations.

Nobody defends a tool once they can see which state of the deal it belongs to. The argument was about a documented job, never about anyone's product.

Technology Navigator
Provider Portfolio
Dynamic Matrices
Data Center Locator
Fiber Tool
CableQuote
Quick Quote

Provider Battle Cards
Provider One Pagers
Case Studies
Cloud Ninja Knowledge Base
UCaaS TCO Analysis
Uploaded Documents

Interactive Quick Assessments
Back Office View

Sample Proposals

MergeSame job, different door

AbsorbFeature, not destination

RebuildRight job, wrong century

RetireNothing breaks without it

One tracked project, five states

01
Project InformationBack office status and the deal's documents
02
ConfigurationProviders, locations and quoting, compared in one place
03
DiscoveryAssessments and engineer knowledge
04
Registration NewReplaced email
05
Contract NewReplaced email. Orders and commissions attach here

Sunset, with a migration pathProposals are generated from the project

All sixteen tools by name. This is the diagram that turned a redesign request into a platform decision.

04Five states

One tracked project, and every state is a different problem

Mapping the work end to end gave the product a shape. Every screen answers for one state of one project.

01
Project Information
Where the advisor lives between deals. It has to answer "what needs me today?" in five seconds without becoming a wall of numbers.
02
Configuration
The screen the client sees. It turns a market of dozens of providers into a defensible shortlist, one dense comparison row each.
03
Discovery
The longest state and the likeliest to stall, because assessments sit with engineers for days. The waiting had to read as progress and not as anxiety.
04
Registration
The irreversible step. It is its own state in the record, but its control sits inside Configuration, beside the comparison that justifies it. That placement came out of testing.
05
Contract
The state that has to outlive everyone. A signed deal becomes a deployment, a billing relationship, then a renewal handled by someone who was never in the room.

Pathfinder Project Information screen in the dark product theme.

Pathfinder Configuration screen: provider comparison rows with fit scores.

Project Information and Configuration as shipped, in the dark product theme.

05Testing

The model held. Two details didn't

The five states went to advisors as a clickable prototype before build. Advisors described the states back in their own words, unprompted, and stopped asking "where's the assessment I ran?"

What failed

The logo outweighed the evidence

The first Configuration table gave the fit score and the provider logo equal weight. Advisors scanned past the evidence and picked on brand recognition, the exact thing the screen existed to prevent.

The fixI rebuilt the row hierarchy until the score and the differentiator outweighed the logo.

What failed

Registration read as a commitment

I had treated registration as a light confirmation. Advisors treated it as the irreversible step it is, and wanted the reasoning in front of them when they took it.

The fixThe register control moved into Configuration, next to the comparison that justifies it.

One explored Pathfinder direction: advisor dashboard with revenue figures and a featured-provider panel in the dark theme.

A second explored direction for the same dashboard, trading the featured panel for a denser task list.

Visual direction was tested the same way. Two of the dashboard directions that went side by side in front of the team, so the choice was between options and not opinions.

06System

Sixteen tools had built the same few things sixteen ways

Each tool had grown its own table, its own card, its own way of saying "pending." If two things differ only in density, tone or content, they are one component with a property. The set got smaller while the coverage got larger.

Families of per-tool parts collapse into single components. The variation moves into properties and tokens.

Three tiers of tokens: primitives, semantic, component. Engineers consume intent, never hex codes.

The system had a second customer. AVANT's marketing team shipped light-theme campaign work at high volume with no shared foundations. Dark product and light marketing resolve from one semantic layer, so a re-theme changes nothing downstream. Every component shipped with loading, empty, error and disabled states, because in-between states are most of what advisors see.

Pathfinder token architecture: primitive, semantic and component tiers.

Pathfinder chart library: themed bar charts with live revenue data.

The three token tiers, and the chart library resolving dark and light from one color logic.

One job, one place

If two surfaces do the same job for different audiences, permissions separate them. Navigation is a promise about where things live.

A view of existing data is a filter, not an application

This is also the rule that stopped the count from creeping back up.

Density is a property, not a new component

The comfortable table and the compact table are the same table. Anything else is two things to fix every time something changes.

Every state, or it isn't done

Loading, empty, error, disabled. Specified once, inherited everywhere.

Prove it on the densest screen

Components were stress-tested against the worst case first. Anything that only worked on the simple screens wasn't finished.

Nothing gets retired silently

Every sunset shipped with where-it-went documentation. Consolidation earns trust or it costs it, and you only get one attempt.

Checked against itselfI ran this same method on my own system and published what it found.Five findings, each traced to a file, and a list of what is still unsourced.

Read the audit ↗

07Outcome

The default path because it was the easier one

Components went to engineering as a React-ready library through Dev Mode: named, specced and versioned. Governance was built for adoption and not enforcement. Docs were written for the person shipping at 4pm on a Friday, weekly office hours caught edge cases before they became forks, and releases came with migration notes.

The library now also runs as a coded build in Figma Make.

Open the live system ↗

Pathfinder design system sheet beside two product screens assembled from it: palette, type, buttons, inputs, badges and data display on the left, a scheduling screen and a project dashboard on the right.

The system sheet beside two product screens assembled from it.

Before and after, layer by layer.

Layer	Before / sixteen tools	After / one platform
Primary navigation	16 tools on the home screen, each its own front door	5 states of one project, plus a global search
Deal record	Email, chat, attachments, personal spreadsheets	One tracked project. Every artifact attaches to a state
Components	Each tool's own tables, cards and status labels	One library, every state specified
Theming	A palette per tool, marketing on its own	One semantic layer for dark product and light marketing
Data visualization	One-off charts per tool	6 themeable chart types on one color logic
Adding a surface	Rebuild the basics, re-litigate the brand	Assemble from the library, consistent on day one

It absorbed growth. Everything new arrived as a component, a property, or a panel inside an existing state. No sixth destination was ever invented.

Adoption came without a mandate. The library became the default path because it was the easier one.

Reviews got better. Brand reviews shifted from policing basics to discussing ideas, and accessibility findings stopped recurring because the fixes lived at the foundation.

The Pathfinder design system today, in its Live Builder: a command panel, quick recipes and floorplans on the left assemble a Partner Performance Overview dashboard on the right, with KPI tiles, a partner tier donut, revenue by region bars, a top partners table and system alerts, all built from the library's components.

The system today. Its Live Builder assembles a partner dashboard from the library's own components.

Sixteen tools was never an interface problem. It was a question nobody had answered: what is the one thing this platform is for?