← All case studies

Prompt to Dashboard

Making Bliss Point’s dashboard builder the obvious choice over Claude

Type what you want to see and get a client-ready dashboard in 30 seconds instead of an hour.

Bliss PointbyTinuiti
A dashboard built from one prompt: podcast reach and site visits, with daily site visits and reach by platform
faster than building a dashboard by hand
120×
faster than building it in Claude
~30×
initiative at the company this year
Largest

The problem

Every client needs a different dashboard, and people were already building their own in Claude, off brand and sometimes with wrong numbers.

What I did

Director of Product Design at Tinuiti, on the Bliss Point platform

I pitched and designed a builder where you type what you want and AI assembles the dashboard from our own components and data, and says so when a request doesn’t make sense.

Bliss Point powers 300+ brands, including

Liquid I.V.OLLYBrooks Brotherse.l.f. CosmeticsBlue ApronpoppiSquareKalshiSPANX

Try it first

Hit New Dashboard, pick the podcast clicks prompt and press Build. Made-up data for a fictional client.

Working prototype

Dashboard Library

Harvest Market · illustrative data

Harvest Market Dashboards3

Performance overview

Monthly spend, revenue, ROAS and conversions across every channel.

Paid SearchPaid SocialStreaming Video+2

Avery Chen · Oct 2

Paid social efficiency

Q4 ROAS and spend by platform for the paid social team.

Paid Social

Jordan Patel · Sep 29

Mobile revenue by region

Where mobile shoppers convert, for the regional promo plan.

Paid SearchPaid Social

Sam Rivera · Sep 24

The real tool lives behind Bliss Point’s login, so I recreated it here. Want to see the Figma files? Just ask.

Background

One dashboard for everyone.

Bliss Point is a massive marketing analytics platform. It works with an incredible amount of data and serves people across many marketing disciplines, and no two clients need the exact same thing.

Its original ethos was the opposite: one curated layout for everyone, so everybody knew what to expect. It showed the data clearly and beautifully, and leadership loved it.

Two colonial-era stick figures present a single chart: 'Let us remove the burden of choice for marketers!' and 'One config for all and for all one config!'
Bliss Point's cross-channel dashboard on a desktop monitor: fixed Spend, Impressions, KPI Volume and ROAS cards above a daily spend vs. impressions chart
Bliss Point’s dashboards before: the same fixed layout for every client. Image source: tinuiti.com.

But one size fits all causes more problems than it solves. To meet user and client needs, Bliss Point has to be dynamic.

The problem

Every client needs a different dashboard.

Tinuiti manages close to a thousand clients. Big ones, too. You’ve seen their Super Bowl commercials (like this one from Ramp).

People check client performance at least weekly: campaigns, targeting, click-through, conversion, spend. Bliss Point is the source of truth for them and for their clients.

Six frustrated users: 'I have to use euros!', 'WTH why can't I flip axes?!', 'I need more tables!!!', 'My client only accepts line charts', 'I need more charts', 'This app SUCKS'

A client optimizing landing page conversion needs a different picture than one dialing in Meta and Google spend, or one measuring a new national TV spot.

One team wants spend and ROI in one table. Another wants them side by side. Another needs none of it.

Every client needs a tailored dashboard, and it has to feel intuitive to every user.

The catch

People were already building their own in Claude.

People at Tinuiti were already using Claude, Gemini and ChatGPT to make slides, sales decks and HTML pages. Plenty of people had started making dashboards too. AI is fast and gives people what they want as fast as they can ask.

But those dashboards were inconsistent at best and inaccurate at worst. Branding suffered. Readability was hit or miss. Data integrity went right out the window.

Happy users and their Claude-made dashboards: 'Claude, break every brand style rule!', 'Share my client's data with a tunneling service.', 'Hey IT, can you host this HTML?', 'I made my own dashboard!'
A designer cheers on top of an avalanche of chart controls, 'Give them what they need!', while three users say 'that's too many clicks! It'll never work!'

The obvious fix, piling on chart settings, wouldn’t work either. Too many clicks. And engineering can only support so many kinds of charts, so whatever we built had to stay inside one flexible framework.

It had to beat what people were doing in Claude

Accurate

Real client data, no made-up numbers.

Consistent

Titles, legends, colors and tooltips that work the same on every chart.

Just as easy

Ask in plain words, like people already do with AI.

Shows its sources

Marketing data comes from conflicting places, and people need to know which.

The pitch

I pitched it twice.

  1. 1

    I pitched configurable dashboards.

    About two years earlier, so people could shape a dashboard to fit each client.

  2. 2

    We went with curated ones instead.

    They were faster to build.

  3. 3

    People didn’t stick with them.

    They went back to spreadsheets and slides, and then to Claude.

  4. 4

    I pitched it again, with AI.

    Describe the dashboard, and it’s built from our own parts. I pitched and designed it with no PM.

  5. 5

    It became the company’s largest initiative of the year.

    Now in alpha.

The solution

Describe it, and AI builds it from our own parts.

I designed exactly what people were already doing, minus the mess: a dashboard you build by describing it. If you’re an AI nerd, think of it as a RAG-based design system with a layer of data connections on top.

If you’re not, imagine prompting for what you want and getting consistency and accuracy without hallucinations.

The designer whistles a Claude hawk down from the sky: 'Don't worry. We got this. Besides, they're all already doing it.'
Asked for the click-through rate of podcast spend, the assistant replies 'Before I build this': podcast ads are audio, so there's nothing to click. It suggests site visits and reach instead.
A dead-simple rule for impossible requests.Ask for the click-through rate of podcast spend (audio ads have nothing to click) and the AI says so before building anything, and suggests the closest real fit. Ignore it, and the chart comes up blank with a note explaining why.

Ask for a dashboard

Try both kinds of request: one that works, and one that can’t exist. (Made-up numbers.)

Interactive
Describe the chart you want
Your dashboard shows up here.

Inside a built dashboard

Tap a number to see the thinking behind that detail.

Tap to explore
Podcast reach and site visits dashboard with five numbered notes

The result

In alpha.

The company’s largest initiative of the year.

faster than building a dashboard by hand: 30 seconds instead of an hour
120×
faster than building it in Claude, which takes about 15 minutes
~30×

Accurate and on brand, every time

Dashboards meet data, client and brand standards.

Shareable outside the company

Unlike with Claude, people can share and export them, which has improved client communication for alpha users.

The dashboard library for a client: saved dashboards with their owners, channel tags and when each was last opened, plus folders for weekly reviews and Q4 planning
Saved dashboards live in a shared library for each client, so the whole team can find them.

Alpha users say it’ll make their work much faster and easier.

Want to talk through this project?

I’d love to walk you through it, including the parts I couldn’t show here. Ask me about:

  • How the AI builds only from our own chart components
  • What happens when a request doesn’t make sense
  • How I pitched it, twice
All case studies