From signal to shipped, in one loop.
Tarazu runs the whole arc of a product decision. Each step produces something the next one needs — and the last step feeds the first, so the system gets sharper with every call you make.
01 · Listen — every signal in one place
Requests, customer feedback, support themes, research notes, and internal ideas land in one workspace instead of four tools. You can type them in or import a CSV; the importer maps your columns onto Tarazu's fields and routes rows by type, so an existing spreadsheet is a starting point rather than something to abandon.
The point isn't collection for its own sake. A score you can defend needs evidence attached to it, and evidence scattered across three tools is evidence nobody checks.
02 · Score — the same way, every time
Every candidate is scored on four dimensions — reach, impact, confidence, and effort — each on a 1–100 slider. The score recalculates as you drag.
score = round((reach × impact × confidence) ÷ max(effort, 1))
All four inputs use the same 1–100 scale. Normalizing them this way makes candidates comparable across teams and quarters, and removes the unit-mismatch errors that make classic RICE spreadsheets disagree with each other. The trade-off is that you give up the literal “reach in users per quarter” reading — the number becomes a rank, not a measurement.
Where the AI fits
Tarazu can draft scores for you. Per-candidate suggestions come from Claude Sonnet, grounded in the product context you've written and any prior feedback in the workspace. Every suggestion arrives as an editable draft with its reasoning attached.
The AI never scores on its own. A suggestion has to be accepted by a person before it counts, because the value of a prioritization system is that someone is accountable for the call — and you cannot hold a model accountable.
03 · Decide — see the tradeoff before you make the call
The tradeoff map plots every candidate with effort on the horizontal axis and impact on the vertical, splitting the field into four quadrants. Point size can track reach or score; point colour can track confidence or quadrant.
| Quadrant | Effort | Impact | What it means |
|---|---|---|---|
| QUICK WIN | ≤ 50 | > 50 | Do these first — cheap and consequential. |
| STRATEGIC | > 50 | > 50 | Worth it, but plan for the cost. |
| FILL-IN | ≤ 50 | ≤ 50 | Cheap but minor. Fill gaps, don't build a roadmap on them. |
| THANKLESS | > 50 | ≤ 50 | Expensive and minor. Say no, and record why. |
The advisor
One click runs a whole-backlog analysis through Claude Opus. It returns a recommended top priority, the fastest available win, the risk most likely to bite, a suggested sequence, and a strategic read on the set as a whole — each with the reasoning that produced it, so you can disagree with the argument rather than just the answer.
Without an API key configured, the advisor runs in demo mode against locally generated analysis, so the flow is explorable before you commit anything.
04 · Ship — the call becomes a record
A decision is recorded with what you chose, what you chose against, and why. Score changes are kept in history and can be reverted, so “why is this ranked third now?” has an answer that isn't someone's memory of a meeting.
This is the part spreadsheets lose. The numbers survive; the argument that produced them doesn't — which is why the same debate reopens every quarter.
05 · Learn — feed the outcome back
Once something ships, record what actually happened. That outcome becomes context for the next round of scoring: the estimates your team habitually inflates become visible, and later AI suggestions are grounded in your own track record rather than in generic priors.
Then step 05 feeds step 01, and the loop runs again.
Who it's for
Tarazu is built for the person who has to defend the roadmap, whatever their title:
- Product managers running a backlog bigger than a meeting can hold, who need the ranking to survive contact with stakeholders.
- Founders making sequencing calls without a product org to delegate them to.
- Product ops trying to make scoring consistent across teams that each invented their own.
- Platform and infrastructure teams whose work competes with customer-visible features and loses on visibility rather than value.
- Consultants and fractional PMs who need a defensible artifact to hand a client, not a spreadsheet only they can read.
What you need to start
Nothing. Tarazu runs in guest mode with no account: your workspace is stored in your browser's local storage and never leaves the device. Sign in only when you want it synced across machines.
When you do sign in, workspaces sync to a Postgres database with row-level security, and the Anthropic API key that powers the AI features lives server-side — it's never shipped to the browser.
Curious about the scoring model itself? The overview covers the shape of the product; this page covers the mechanics.
Bring balance to your roadmap.
Start weighing what to build next — and stop relitigating the same decisions every quarter.