odo’s analysis with Claude

odo counts. Claude reads. odo’s analysis asks Claude what odo’s numbers mean and shows the answer beside them, in odo’s own words and layout: a verdict you can check against the numbers it came from.

What each post did#

Add a post by its link (Links for posts → Add a post, see Add a post) and odo measures what it did: the visits its platform sent in the 48 hours after it went up against that platform’s usual for those hours, the page it pushed, and the sales those visits led to. Claude reads those numbers with the page you said it’s for, your note, and the post’s title or caption, and writes:

• A verdict: worked, mixed or didn’t work, with one line on what the post did and the number that matters most.
• Where it sent people: the page it pushed, and whether that’s the page you meant.
• Why: two or three sentences citing the numbers.
• What to do next: one concrete step, such as where to point the next post’s link or what the page people landed on is missing.

Claude reads a post once it has a day of visits, and again when its 48 hours are in. Read it now, or Refresh analysis, asks sooner, at most once every ten minutes. On a handful of visits it’s written to say there’s too little to tell, and while the 48 hours are still going, that it’s early. If you change when the post went up, the read is cleared and written again for the new hours.

The weekly review#

odo’s review, at the top of the Overview, is Claude’s read of the whole week: a headline and a short summary, then what’s Working, what’s Leaking, and what to do Next (each step with why it’s worth it and where in odo to look), and a thing or two to Keep an eye on. Each list stays short, three points at most, and is left empty rather than padded. Under it, Claude says how confident it is, high, medium or low, and why. Earlier reviews stay a pick away.

When it’s written#

Every Monday from 7:00 in your time zone, for each project with a week of real activity, about Monday to Sunday just gone. Write this week’s on the card writes one for the last seven whole days, for a project or for Global; that works once every six hours, or 15 minutes after one that failed, and takes a minute or two. If none of the numbers have moved since the last review, that one still stands. A review needs a week of visits, or earnings on a few days; with less, odo says there’s too little for one yet.

Where it reaches you#

Monday’s 9:00 weekly summary carries the review’s headline and its first thing to do, in every notification channel with the weekly summary on. When Monday’s review is finished after a channel’s summary went out, it follows on its own, with what’s working, what’s leaking and the next steps. Signed webhooks carry the headline, summary and next steps as data.

What it reads#

The week against the one before, with the last 28 days for what’s normal lately, from the same numbers your dashboards show:

• Earnings: what the week earned by source, its payments, refunds and disputes, and ad earnings where you have them.
• Visits: visitors, visits, pageviews, bounce rate, time on site, returning visitors, and the top channels, sources, campaigns, landing pages, pages, countries and devices.
• Sales traced to the visits that brought them, and how many visitors bought.
• Your goals and funnels, and the A/B tests still running.
• Posts: links for posts, waves odo found, and the posts you added, with what you said each was for and Claude’s read of it.
• Changes made to the site in the last 30 days, and what the numbers did after.
• What odo noticed, such as a source that stopped syncing, and its own Next moves.

What it can’t tell#

The facts end with what odo can’t see in that project: no website measured, sales not traced to visits, no goals or funnels. The review makes no claim that needs what’s missing; when a gap stops it answering something that matters, it says so, and filling the gap can be one of its next steps. It says when a change rests on numbers too small to mean much, and calls a cause “likely” when the numbers can’t settle it.

Experiments#

When odo calls an experiment, Claude writes its read-out: what happened and by how much, why it likely went that way, the segments it matched best and worst, the test to run next, and its caveats. Read it out asks for one sooner, at most every ten minutes. In the New experiment form, Suggest versions has Claude propose versions for a page and a goal, from the page’s numbers, what people click there and your past experiments; odo checks every change before it can run.

Who has it#

All of it comes with the Growth plan, and with the free trial, which has everything Growth has. On Starter, the numbers show on their own: a post’s sheet says Claude’s read comes with Growth, the Overview’s review card is a single line saying the same, an experiment’s sheet says its numbers are the whole call, and everything else works as before.

The owner, admins and managers of a project see its reviews and can write one. Shared views and public pages never show them.

What Claude sees#

Aggregates odo has already counted: totals, rates and how many people each rests on, a post’s caption and the pages it led to, the names of your sources, goals, funnels, tests and site updates, an experiment’s versions and what they change, and notes you wrote. Never a visitor’s or a buyer’s personal data: no IP addresses, visitor ids, names, emails or addresses.

Each answer has a fixed shape that odo checks before anyone sees it, and Claude is told to state only numbers odo sent it, or simple arithmetic on them. When Claude declines, or an answer doesn’t fit, odo says so rather than showing a made-up one.

Asking your own questions#

The analysis answers the questions odo knows to ask. For your own, connect Claude, ChatGPT or another assistant to odo over MCP: it reads the same numbers, live, when you ask.