Files

Adroit studio /Courtside· Case study

Start a project

Case study · 2026

Courtside

An AI front office for a pro basketball team

A confidential desktop workspace for a pro basketball front office: rosters, trades, the salary cap, scenarios and news in one place, with an assistant that shows its work and asks before it acts.

Product designBrand identityBuilt from the product itself and public research, not from interviews. Read it as a hypothesis to test, not as findings.

Desktop app for 1280 to 1920 windows, dark by default with a light theme

The challenge

Cap rules got stricter and the tools stayed scattered. Staff check the cap in one place, test a trade in another and ask questions somewhere else, then carry the numbers between them by hand. The product had to join that work up and add an assistant people could trust with decisions worth millions.

What we did

  • Brand guideline
  • Design system
  • Product design
  • AI interaction patterns

What changed

  • One workspace links rosters, the trade builder, the cap dashboard, the simulator and news.
  • Every assistant answer carries its steps, sources and data freshness.
  • Nothing the assistant proposes changes data until a person approves it.
  • A brand guideline and a full component library, themed for dark and light.
  • Eight app shells, each laid out for 1280, 1440 and 1920 windows.

Overview

Courtside is a desktop app for the people who build a pro basketball roster. It covers the teams overview, rosters, player profiles, a trade builder, a salary cap dashboard, a scenario simulator and a news feed.

Beside all of it sits an assistant. It reads the team’s own data, explains how it reached an answer, and asks before it changes anything.

We designed the brand guideline, the design system and the app shells. The data is confidential, so the app runs on the team’s own machines and has no public pages at all.

The challenge

Since the 2023 labor agreement, a second spending apron limits what teams above it can do in a trade. Every season the thresholds move, and a mistake can cost draft picks.

Public tools split the work: one site tracks the cap, another checks trades, a third answers stat questions. Inside a front office the same split means copying numbers between tabs at the worst possible moment, the trade deadline.

An assistant could close that gap, but only if staff can see what it read, how current the data is, and that it will not act on its own.

The landscape

We reviewed eight public and team-facing tools for cap work, trade checks, analytics and AI answers. Every claim below links to its source; enterprise tools sit behind logins, so we only score what they show publicly.

In the matrix, "no" means not shown publicly and "unknown" means we could not verify it. The Courtside column reflects the design brief, not a build in use.

Showing 9 of 9 features

The landscape. Each cell says whether a product has a feature: yes, partly, no, not known, or not public (it may exist but nothing public shows it). Our product is the first column.
FeatureOurs
Trade rule checkingYesPartlyYesYesPartlyNoNoNoNot known
Apron and tax modelingYesYesYesPartlyPartlyNoNoNoNot known
Scenario simulationYesPartlyPartlyPartlyPartlyNoNoNot knownPartly
Assistant with visible reasoningYesNoNoNoNoPartlyNoNoPartly
Source citationsYesNoNoNoNoNot knownNoNoYes
Human approval of assistant actionsYesNoNoNoNoNoNoNoNo
On-device or restricted dataYesNoNoNoNoNoNoNoNot known
Light and dark themesYesNot knownNot knownNot knownNot knownNot knownNot knownNot knownNot known
Data freshness signalsYesPartlyPartlyNot knownNot knownNot knownNot knownNot knownNot known
  • Yes
  • Partly
  • No
  • Not known
  • Not public

Where we stand out

  • One workspace instead of four tabs

    Cap tracking, trade checks and analytics are separate products today. Courtside puts cap status on the teams overview and links the roster, trade builder, cap dashboard and simulator.

  • An assistant, not a search box

    The tools we reviewed answer single questions. None publicly documents an assistant that shows its steps for a multi-constraint question like who fits under the first apron at the deadline.

  • People approve every action

    We found no approval step before an assistant action in any reviewed tool. Courtside makes it the default, because a wrong move costs money and picks.

  • Cited answers with freshness

    Only one reviewed tool documents citations, and only for uploaded PDFs. Courtside shows the data scope, season and time behind every answer.

  • Privacy as a feature

    Front offices guard their analytics closely. No reviewed tool documents on-device or restricted-data use; Courtside is built around it.

Where they do better

An honest read: these are the places the others are ahead today.

  • Data depth: Synergy has 20+ years of play-type data and video, and tracking vendors own the raw feeds. The assistant is only as good as the data it can reach.
  • Procurement trust: Teamworks and Synergy are already approved vendors inside front offices. A new tool has to pass security review first.
  • Reach and speed: ESPN and Fanspo are free, familiar and easy to share, and they set expectations for how fast a trade check should feel.

Sources

The landscape

Spotrac

Fans, media, agents and front-office staff who want a quick public cap reference.

Audience
Fans, media, agents and front-office staff who want a quick public cap reference.
Website
spotrac.com/nba/apron (opens in a new tab)

Strengths

  • Team cap tracker with allocations, cap space and dead cap.
  • Apron tracker showing first- and second-apron room for every team.
  • Trade machine that shows the cap impact for each side.

Gaps

  • Cap, apron and trade tools live on separate pages.
  • No scenario chaining and no rationale behind a result (our analysis).
The landscape

Fanspo Trade Machine

Fans and writers.

Audience
Fans and writers.
Website
fanspo.com/nba/trade-machine (opens in a new tab)

Strengths

  • A published rulebook: deals are checked against the 2023 labor agreement with 2026-27 figures.
  • Covers apron hard caps, aggregation limits, frozen picks and no-trade clauses.
  • Shareable trade links.

Gaps

  • A "valid" result can still sit on a second-apron edge case, so people must interpret it themselves (our analysis).
The landscape

ESPN Trade Machine

A mass fan audience.

Audience
A mass fan audience.
Website
espn.com/nba/trademachine (opens in a new tab)

Strengths

  • Free, familiar, and built for sharing.
  • Pass or fail check against league trade rules, including trade exceptions.
  • Projects post-trade records from each rotation.

Gaps

  • The page does not list which rules it checks.
  • The apron rules forced a full rewrite of its rule logic, which shows how hard the rules are to keep current.
  • Record projections rest on a single rating metric (our analysis).
The landscape

Cleaning the Glass

Analysts, writers and agents. The site says pro teams use it.

Audience
Analysts, writers and agents. The site says pro teams use it.
Website
cleaningtheglass.com/stats (opens in a new tab)

Strengths

  • Stats that filter out garbage time by default, with percentiles on every stat.
  • A Cap Calculator that models waive and trade scenarios and includes cap holds.

Gaps

  • Apron and trade-rule checks are not verified.
  • A subscription product run by one operator (our analysis). Current price not verified.
The landscape

StatMuse

Fans and media. In May 2023 it said 20+ pro teams used its API.

Audience
Fans and media. In May 2023 it said 20+ pro teams used its API.
Website
statmuse.com (opens in a new tab)

Strengths

  • Natural-language answers across six sports.
  • A paid tier with unlimited prompts and search history (launched at $20 a month in 2023; current price not verified).

Gaps

  • Answers are stat lookups, not multi-step reasoning over cap rules (our analysis).
  • No visible reasoning, approvals or private team data found (our analysis).
The landscape

Synergy Sports (Sportradar)

Pro, college and high-school coaching staffs. Sportradar reports deploying it across the league; the exact team count is unresolved.

Audience
Pro, college and high-school coaching staffs. Sportradar reports deploying it across the league; the exact team count is unresolved.
Website
sportradar.com/teams-leagues-conferences/synergy-coaching-scouting/coaching-solutions/?lang=en-us (opens in a new tab)

Strengths

  • Game data linked to video, with one-click clips and heavy filtering.
  • Play-type detail and 20+ years of history, now also as an API.
  • Machine-learning tagging on top of player-tracking data.

Gaps

  • No cap, trade or roster-management tools (our analysis).
The landscape

Second Spectrum (Genius Sports)

Pro teams and broadcasters, behind a login.

Audience
Pro teams and broadcasters, behind a login.
Website
secondspectrum.com (opens in a new tab)

Strengths

  • Named the official team analytics provider in March 2023, serving every team in the league.
  • Machine-learning event and play classification.

Gaps

  • The evidence dates from 2023; current team features are not verified.
  • Focused on on-court analysis, not cap or trade work (our analysis).
The landscape

Teamworks (Zelus Analytics)

Team operations, performance and analytics departments. Teamworks claims most of the league as partners (vendor claim).

Audience
Team operations, performance and analytics departments. Teamworks claims most of the league as partners (vendor claim).
Website
teamworks.com (opens in a new tab)

Strengths

  • A data warehouse that combines league, team and vendor data.
  • Predictive models positioned for roster construction and trade decisions.
  • Citation-backed Q&A over uploaded PDFs.

Gaps

  • Salary-cap and rule tooling not verified.
  • The AI answers questions about documents; it does not run multi-step analyses or ask for approval (our analysis).

Proto-personas

A proto-persona: drawn from the product itself and public research, not from interviews. It is a hypothesis to test with real people.
  • Assistant general manager who runs trade and cap work day to day

    “Tell me what fits, show me why, and do not touch anything until I say so.”

    Goals
    3
    Frustrations
    3
  • Salary cap and labor-agreement specialist

    “A green check means nothing to me unless I can see which rule it checked.”

    Goals
    3
    Frustrations
    3
  • Head of basketball analytics

    “If it ran a query, I want to see the query.”

    Goals
    3
    Frustrations
    3
Proto-personas
A proto-persona: drawn from the product itself and public research, not from interviews. It is a hypothesis to test with real people.

The front-office manager

Assistant general manager who runs trade and cap work day to day

“Tell me what fits, show me why, and do not touch anything until I say so.”
Illustrative line, not a recorded quote

Context

Works on a large desktop screen through long days that spike around the deadline. Approves moves, briefs the general manager and owns the paper trail.

Goals

  • Bring the general manager two or three real options before the deadline.
  • Know a deal is legal before anyone picks up the phone.
  • Keep the team under the first apron unless crossing it is a deliberate choice.

Frustrations

  • Rebuilding the same cap picture in three tools for every idea.
  • Answers with no trail to show the general manager.
  • Finding out about an apron problem after a call has started.
Proto-personas
A proto-persona: drawn from the product itself and public research, not from interviews. It is a hypothesis to test with real people.

The capologist

Salary cap and labor-agreement specialist

“A green check means nothing to me unless I can see which rule it checked.”
Illustrative line, not a recorded quote

Context

Lives in tables and thresholds. Reads every number twice and wants the rule behind each pass or fail.

Goals

  • Model the cap, the tax and both aprons for this season and the next two.
  • Check salary matching and apron restrictions on every proposed deal.
  • Keep one trusted set of numbers for the whole front office.

Frustrations

  • Rule logic changes each season and public tools lag behind.
  • Spreadsheets drift away from the real contract data.
  • Colleagues quote figures without saying which season or when they were pulled.
Proto-personas
A proto-persona: drawn from the product itself and public research, not from interviews. It is a hypothesis to test with real people.

The analytics lead

Head of basketball analytics

“If it ran a query, I want to see the query.”
Illustrative line, not a recorded quote

Context

Owns the data pipeline and the models. Will judge the assistant on whether its queries and sources hold up.

Goals

  • Feed models and player data into roster decisions.
  • Keep the team’s data private and its methods internal.
  • Spend less time answering one-off questions from colleagues.

Frustrations

  • Requests arrive as screenshots with no context.
  • Hosted tools mean sending private data outside the building.
  • AI answers that hide the query behind them.

Empathy maps: The front-office manager

Built from the product itself and public research, not from interviews. Read it as a hypothesis to test, not as findings.

Out loud, in meetings and messages

  • “What can we add without crossing the first apron?”
  • “Get me something I can take upstairs by tonight.”

What goes unsaid

  • If I miss a rule, it is my name on the mistake.
  • The numbers I have may already be out of date.

Observable behaviour

  • Keeps the cap sheet, a trade tool and the news open side by side.
  • Asks the capologist to double-check every idea.

The emotional undertow

  • Pressed for time near the deadline.
  • Wary of tools that act on their own.

Pains

  • Re-entering the same deal in several places.
  • No record of how an option was reached.

Gains

  • Options that already pass the rules.
  • A proposal to review and approve in one place.

Empathy maps: The capologist

Built from the product itself and public research, not from interviews. Read it as a hypothesis to test, not as findings.

Out loud, in meetings and messages

  • “Which season is this figure from?”
  • “Show me the matching math.”

What goes unsaid

  • Every threshold has a season label or it is wrong.
  • An assistant is fine as long as it cites its source.

Observable behaviour

  • Checks salary matching by hand for every proposal.
  • Keeps a private sheet of thresholds and exceptions.

The emotional undertow

  • Responsible for numbers other people quote.
  • Skeptical of black-box answers.

Pains

  • Figures with no date or source.
  • Rule tools that pass a deal without explaining why.

Gains

  • Rule checks that list each rule with pass or fail and a reason.
  • One cap picture the whole office shares.

Make a trade-deadline move without crossing the first apron

Built from the product itself and public research, not from interviews. Read it as a hypothesis to test, not as findings.

Journey of The front-office manager6 stages

  1. Know where the team stands before looking for a deal.

    Feeling today: Frustrated (−1)After the redesign: Pleased (+1)

  2. Get a short list of moves that fit.

    Feeling today: Frustrated (−1)After the redesign: Pleased (+1)

  3. See why each option works.

    Feeling today: Very frustrated (−2)After the redesign: Pleased (+1)

  4. Be sure the deal is legal.

    Feeling today: Frustrated (−1)After the redesign: Pleased (+1)

  5. Commit only what a person has reviewed.

    Feeling today: Frustrated (−1)After the redesign: Delighted (+2)

  6. Hand over a clear record of the decision.

    Feeling today: Neutral (0)After the redesign: Delighted (+2)

Key design decisions

  1. Blue means AI

    Problem
    In a dense data tool, an accent color used everywhere stops meaning anything.
    Choice
    The light blue #75B9FD marks only the assistant, focus and the brand. Actions use a deeper blue, #0D5594, and status colors mean state only.
  2. The assistant shows its work

    Problem
    An answer without a trail cannot be taken to a general manager.
    Choice
    Every answer carries its steps, queries, sources and freshness. Steps collapse to one line after the answer and stay in plain language, never raw model output.
  3. Ask before acting

    Problem
    A wrong trade or cap move costs millions and draft picks.
    Choice
    An approval card sits between any proposal and any change. It lists the rule checks with icons and text and offers Approve, Review or Dismiss.
  4. Answers in a fixed order

    Problem
    Assistant output can be an answer, a chart, a trade, a warning and sources all at once.
    Choice
    Attachments stack under the answer in one order: reasoning, artifact, approval, sources, feedback, then follow-up questions.
  5. Confidential by design

    Problem
    The data is among the most guarded in the sport.
    Choice
    An on-device badge, a remote-session indicator and audit trails are part of the system. There are no share buttons, no public links, and no player photos or team logos.
  6. Dense but readable

    Problem
    Analysts compare many numbers at once, all day, on large screens.
    Choice
    App density by default, with 40px rows and 36px controls. Numbers use tabular figures and money always shows its unit.

The workspace

The teams overview puts cap status on every team tile, so the league picture and the cap picture start in the same place.

The cap dashboard leads with one headline figure and a cap bar. The trade builder steps through select, evaluate and approve.

Teams overview
Salary cap dashboard
Trade builder

The assistant

The side panel sits beside the page at 1440 and above, and overlays it at 1280. The full-screen view adds conversation history and a larger reading size.

Every chart, table and KPI tile has an Ask button, so a question starts from the data in view.

Assistant, side panel
Assistant, full screen

Near-black canvas, neutral surfaces, one blue. The numbers are the loudest thing on screen and the chrome steps back.

Dark is the default for a desktop analytics tool, and every component also works in light.

Brand: colour

  • App background, the darkest surface

  • Panels, cards and the sidebar

  • Previews and reasoning steps inside cards

  • Default card and control border

  • The assistant, focus rings and the brand

  • Primary buttons, with a white label

  • Brand surfaces and the assistant mark fill

  • Brand text and small links on dark

  • Primary text

  • Labels and body at 12 to 14px

  • Muted text at 16px and up, placeholders

  • State only: under the cap, a win

  • State only: over the tax, injury doubt

  • State only: over an apron, out

Brand: type

  • Inter

    Display: one headline figure per screen and stat values (40/46 and 32/38, semibold)

    $14.2M

    • 600
  • Inter

    Headings and body: page titles to labels, body at 14/20 in the app

    Three options stay under the first apron.

    • 400
    • 500
    • 600
  • Inter

    Data: tabular figures for stats and money, always with a unit

    $28.4M · −$2.7M

    • 500
    • 600
  • JetBrains Mono

    IDs and code only

    proposal-12

    • 400

Brand: logo notes

  • A type-only wordmark in bold, with two of its letters set in the AI blue. The blue letters tie the name to the assistant.
  • Shown on dark and light, with rules for clear space and minimum size.
  • Misuse rules: do not recolor the blue letters, do not set the whole mark in one color, do not stretch or track it out, and do not add glows or a symbol.
  • Slots are reserved for the approved primary logo and the app icon at 512 and 32 pixels. Three symbol routes were explored alongside, each with primary, symbol and wordmark lockups on dark, light and mono.

Motion and accessibility

Motion explains change and is never decoration. Hovers take 80ms, dropdowns 200ms and the assistant panel 320ms; with reduced motion, durations drop to zero and the thinking ring becomes a static dashed ring.

APCA is the design target and WCAG AA is the floor: every text pair passes both, and the few accepted exceptions are documented. Status is always an icon plus text, never color alone.

Icon-only buttons always get a tooltip and an accessible name. The primary button label is white on #0D5594, a 7.7:1 contrast ratio.

The design system

78 parts in 9 groups

Structure follows Flowbite’s anatomy, so heights, radii and paddings map straight to code. Values come from variables with dark and light modes, a density mode, and three window sizes.

Designers use instances only, and icons come from one outline set of 93.

    • Colors
    • Typography
    • Spacing and layout
    • Radius, borders and effects
    • Motion
    • Icons
    • Accessibility
    • Buttons
    • Button group
    • Chips and filters
    • Checkbox and radio
    • Toggle
    • Forms
    • Autocomplete input
    • Range slider
    • Datepicker
    • Dropdowns
    • KBD
    • App bar
    • Sidebar
    • Nav tabs
    • Breadcrumbs
    • Pagination
    • Stepper
    • Command palette
    • Split view
    • Tables
    • Cards
    • KPI tile
    • Charts
    • Chart parts
    • List
    • Timeline
    • Accordion
    • Avatars
    • Badges
    • Progress
    • Skeleton
    • Empty states
    • Alerts
    • Toasts
    • Modals
    • Drawer
    • Popovers
    • Tooltips
    • Spinners
    • Notification center
    • Entry points
    • Panel
    • Composer
    • Messages and streaming
    • Reasoning and tool calls
    • Sources and citations
    • Artifacts and approvals
    • Suggested prompts and feedback
    • Agent tasks and memory
    • Player card
    • Team tile
    • Trade slot
    • Cap bar
    • Stat line
    • News item
    • Scenario card
    • Rule check
    • Comparison card
    • Assets and availability
    • Trust and data freshness
    • On-device badge
    • Remote-session indicator
    • Audit trail
    • Teams overview
    • Team and roster
    • Player profile
    • Trade builder
    • Salary cap dashboard
    • Scenario simulator
    • News feed
    • Assistant panel and full screen

Reflection

The hardest work was not the screens. It was deciding what the assistant may never do, and then designing every surface around that line.

Next, the hypotheses need real sessions with front-office staff, and the competitor matrix needs a hands-on audit of density and theming.

A confidential mobile companion for remote approvals is scoped for later. The assistant components were drawn to stretch to any width, so they can carry over.

Want something like this?