Skip to content

Open-source cohort ROAS and LTV dashboard for web subscription funnels

Web-to-Web BI Board is a free, open-source dashboard for subscription businesses that sell on the web. It turns ad spend, trials and payments into cohort ROAS from day 0 to day 360, payback day and forecast LTV.

Licence
MIT, commercial use included
Backend
None. One JSON file, any static host
Pages
Cohorts, Predictions, Real-time, User flow
Horizon
ROAS D0 to D360, forecast with 80 % band
The Cohorts page of the Web-to-Web BI Board on demo data: KPI tiles and a cohort table with ROAS from day 0 to day 360

Selling subscriptions on the web means paying today and getting paid for months

The ad bill arrives this week. The money comes in slowly: a trial, a first rebill a few days later, then a renewal every week or month for as long as the person stays. A campaign that looks like a loss on day 7 can be the best thing you bought this quarter by day 180, and the other way round.

Ads Manager only sees the purchases it attributes inside its window. Stripe sees every payment and knows nothing about the ad. Most teams glue the two together in a spreadsheet once a month, by hand, and argue about whose number is right.

So the same few questions come up every week. Which cohorts paid back, and on which day? What is a trial worth before the renewals show up? How much can we pay per trial and still hit our ROAS target? Is today's traffic fine? Where do people fall out between the ad and the second payment?

How to read the cohort ROAS table, mark by mark

This is the first page you land on. Point at a number on the screenshot, or at a line in the list, to see what each part does.

The Cohorts page with twelve numbered marks explained in the list next to it
  1. Date rangePresets from Today to All time, or exact dates. Every number on the page follows this window.
  2. Step and cohort basisGroup cohorts by day, week or month, and anchor them on the signup date or the trial date.
  3. Gross or netNet takes out refunds, disputes and payment fees, so ROAS shows what actually stays with you.
  4. Channel filterFacebook, Google, TikTok, Web2Web partners or organic. The device toggle next to it splits mobile and desktop.
  5. Who counts as a payerAny payment, including a paid intro, or only people who reached a subscription payment. CAC changes with it.
  6. Forecast in young cellsTurn it on and cells that have not matured yet show the forecast instead of a blank.
  7. Headline tilesSpend, trials, cost per trial, LTV per trial, payers and CAC, each against the previous period.
  8. LTV at a horizonPick D30 to D360. The badge says how much of the number is already observed; the rest is forecast, with its range.
  9. One row per cohortSpend, trials, CPT and CAC for the people who signed up that month. Red means expensive against the rest of the table.
  10. ROAS by dayRevenue to date divided by spend. Green from 1.0x: the cohort has paid back by that day.
  11. Dashed cells are forecastsItalic, striped cells have not happened yet. Hover one for the 80 % interval.
  12. ScopeProject, traffic source and ad account. Pinned, so it stays while you move between pages.

A quick example from the demo: the March 2026 cohort cost $5,383 and sits at 0.89x on day 90, then crosses into green at 1.05x on day 180. That cohort paid back somewhere between month three and month six. October 2026 is too young, so from day 7 on its cells are dashed forecasts.

Four pages, four questions

01 · Cohorts

Which months and channels paid back, and on which day?

  • Cohort table with ROAS from D0 to D360 and the payback day.
  • MRR movement: new, expansion, contraction, churned and net.
  • Paid retention by renewal number, per plan or channel.
  • Billing health: declines by dunning attempt, refund rate, and dispute rate against the 0.75 % line.
Open Cohorts in the demo
ROAS curves by cohort, rebills and churn, MRR movement and paid retention charts on demo data

02 · Predictions

What is a trial worth before the renewals arrive, and how much can I pay for one?

  • Forecast ROAS and LTV per trial at D30, D90, D180 and D360, with an 80 % band.
  • Max CPT: the highest price per trial that still meets a target ROAS at your margin.
  • What-if sliders for first rebill rate, renewal churn, refunds and disputes, kept in the URL.
  • A backtest, so you can see how the model did on cohorts it never saw.
Open Predictions in the demo
Predictions page with pROAS from D30 to D360, intervals, payback forecast and the model panel on demo data

03 · Real-time

Is today's paid traffic healthy, right now?

  • Spend pace against the same hour yesterday, CPT, CPI and ROAS D0 by hour.
  • Checkout funnel with errors by payment provider and method.
  • Payment provider auth rate in the last hour against the last 24 hours.
  • Alert rules you can edit: expensive trials, spend with no trials, CPA over a partner contract.
Open Real-time in the demo
Real-time page with today's spend, signups, CPI, trials, CPT and LTV tiles and the pace chart against yesterday

04 · User flow

Where do people drop between the ad and the second payment?

  • The whole funnel as one map: source, landing, onboarding, paywall, checkout, trial, payment.
  • Edge width is people, chips are conversion rates, nodes carry forecast LTV.
  • Losses drawn on the map: drop-off, checkout error, declined card, refund, dispute.
  • Lanes by channel, device or market, and an A/B card with the chance each variant is better.
Open User flow in the demo
User flow page: a funnel map from traffic source and landing to signup, paywall, checkout, trial, payment and losses

How the LTV forecast is made, and where it stops

The model is old and well understood on purpose. You should be able to explain every number to a finance person.

  1. Segments. Every user sits in a segment: plan, geo tier and channel. A segment with too little history borrows from the level above it, and the page tells you which level each row uses.
  2. Trial to paid. The share of ended trials whose first rebill went through, weighted toward recent trials (45-day half-life), plus the chance a failed first payment is recovered on a retry.
  3. Renewals. A shifted-beta-geometric curve (Fader and Hardie): people differ in how likely they are to cancel, so the survivors get stickier. Two parameters, fitted on the renewals you already have.
  4. Revenue to day N. Money already received plus the expected payments of everyone still in play, with an 80 % band.
  5. Check it. The backtest predicts day 30 from what was known on day 3 or 7, and a hold-out test refits on older data. Look there before you trust a number.

What the forecast cannot see

Anything that is not in your data yet: a new price, a new creative, a seasonal dip. With a few hundred renewals per segment it is good for comparing channels and setting bid caps. With fewer, the page says "Not enough data" and falls back to broader segments. The repository does not claim any accuracy on real data.

Documentation: run it on your own data

1. Run it locally

Plain HTML, CSS and JavaScript. Nothing to install or build. You need a static server only because browsers block reading files from file://.

Clone

git clone https://github.com/ua5er/web2web-bi-board.git

Serve

cd web2web-bi-board && npx serve .

2. Publish it on GitHub Pages

  1. Fork the repository to your GitHub account.
  2. Settings, Pages: source "Deploy from a branch", branch main, folder root.
  3. Wait a minute and open your-user.github.io/web2web-bi-board.

Private numbers do not belong on a public Pages site. For real data, host it somewhere only your team can reach.

3. Export your data to one JSON file

The board reads one file with these tables. Timestamps can be ISO strings, milliseconds or seconds. Big tables can use a compact column form. The full field list is in the data contract.

TableStatusWhat goes in
usersRequiredOne row per signup: time, country, channel, campaign and ad, device, landing, paywall and plan.
trialsExpectedTrial start, plan, paywall, paid intro amount, cancellation time.
chargesExpectedEvery payment attempt: type (intro, rebill, renewal), renewal number, amount, status, decline code.
refunds, disputesExpectedAmount and time. They come off net revenue and end the subscription.
spendExpectedSpend by date and hour, campaign and ad, with impressions and clicks if you have them.
campaigns, ads, plans, paywalls, events, incidentsOptionalNames and lookups, the plan catalogue, product events, and outage windows the forecast should ignore.

4. Point the board at your file

Any of these three works:

  • Add ?data=./data/my-export.json to any page address.
  • Change data-src on the core.js script tag in the four HTML pages.
  • Set window.BB_DATA_URL before core.js loads.

The loader checks the file. Errors stop the load and show up on an error card; warnings become notes on the page. You can also run BB.adapters.validate(json) in the browser console.

5. Make it yours

Title, contacts and links live in board.config.js. The logo comes from your GitHub avatar. Dark and light themes, English and Russian, phones down to 390 px.

Every view, with its dates, filters, model and what-if scenario, lives in the URL. Send a link and your colleague sees exactly what you see.

New demo dataset

node tools/generate-demo-data.mjs --seed 42

When it fits, and when something else is better

This boardA spreadsheetA paid BI tool
CostFree, MITFreeLicence plus setup
Cohort ROAS to D360 with forecastBuilt inYou build and keep itYou build it
Live data connectorsNot yet, you export a fileManualUsually built in
Where the data livesYour file, your hostYour fileTheir cloud
Time to first viewMinutes on demo dataDaysWeeks

If you need dashboards that refresh themselves from ad accounts every hour, a paid BI tool or a warehouse does that better today. This board is for teams that want the subscription maths done right first, and can export a file.

Questions people ask about it

What is the Web-to-Web BI Board?

A free, open-source BI dashboard for subscription web funnels. It shows cohort ROAS from day 0 to day 360, payback day, forecast LTV, real-time spend and a funnel map. It is a static site that reads one JSON file.

What is a web-to-web funnel?

A subscription funnel that stays on the web: ad, landing page or quiz, web paywall, web checkout, and renewals on the web. There is no app store purchase, so the business sees every payment and has to measure payback itself.

How is cohort ROAS different from the ROAS in Ads Manager?

Ads Manager counts the purchases it attributes inside its window. Cohort ROAS follows everyone who signed up in a period and adds every payment they make over time, renewals included, then divides by what that period cost.

How does the LTV forecast work?

It estimates trial-to-paid per segment, fits a shifted-beta-geometric (sBG) curve to observed renewals, and adds expected future payments to money already received. The band is 80 %. Thin segments borrow from broader ones.

Do I need a backend, a database or a warehouse?

No. The board is plain HTML, CSS and JavaScript. Any static host serves it, including GitHub Pages, and every calculation runs in the browser from your JSON file.

Can it pull data from Meta, Google Ads, TikTok or Stripe on its own?

Not in version 1.0. You export your data into one JSON file that follows the data contract. Connectors are on the roadmap.

Is it free for commercial use?

Yes. The code is under the MIT licence, so you can run it, change it and use it at work without paying.

Is the demo data real?

No. The demo shows a fictional app with about 16,000 synthetic users over 14 months, generated by a script in the repository. Every campaign, user and payment in it is invented.

Web-to-Web BI Board on AlternativeToThe social preview of the ua5er/web2web-bi-board repository on GitHub

Want this on your numbers?

I can wire predictive analytics into your funnel

I am Vlad Yasinski, and I built this board. If you sell subscriptions on the web and want cohort payback, LTV forecasts and max CPT running on your own Stripe and ad data, I can set it up: the export, the model, the bid caps, and a weekly read of what changed.