Intelligence: your sites' numbers, and an analyst that has to cite them
Intelligence is a page in the Anticode dashboard that brings a workspace's numbers together: visits since each site was published, leads from its forms, what model builds have cost, and a forecast once there's enough history. Beside it sits an analyst you can ask questions in plain English. Every fact in its answer has to point at one of the measured numbers on the page, and anything that doesn't is removed before you see it.
One page instead of four tabs
A small business with a site or two doesn't need a reporting suite. It needs to know whether people are coming, whether they're getting in touch, and whether last month was better or worse. Those numbers are usually scattered across an analytics tool, an inbox and a billing page. Intelligence is a website performance dashboard AI analyst in one place: it reads them from data Anticode already keeps and shows them for the workspace as a whole, over the last 7, 30 or 90 days. If you work in a Team workspace, you can switch to it and see its sites instead.
What the page shows
| Area | What it shows | Where it comes from |
|---|---|---|
| Traffic | Visits in the period and the one before, the change, visits by source and per day. Search and AI crawlers are shown separately and left out of totals | The cookieless counts behind site analytics, only for days since each site was published |
| Leads | Form submissions in the period and the one before, by form and per day, and leads per 100 visits | Submissions through your published sites' forms |
| Sites | Each published site with its visits and leads | Both of the above |
| Model use | Successful and failed builds, analyst questions, and what they cost | Your build records |
| Revenue | Not measured, and the page says so | Payments go through your own checkout links, which Anticode never sees |
The forecast, and why it waits
A forecast drawn from a few days of data is noise with a confident line through it. So Intelligence shows none until a published site has 14 days of history; before then it says a forecast needs 14 days. After that it projects visits and leads for the next 30 days using the plain average of the last 28 days (fewer if your first site went live more recently), with a range of one standard deviation either side. It's labelled as an estimate, and the assumptions are written out: it assumes nothing changes.
Asking the analyst
Type a question into Ask about your numbers, such as "why did leads drop?" or "which site gets the most from AI assistants?", and press Ask. Behind it, a model receives the page's numbers as a flat list of named metrics (views.total, leads.form.booking, site.<name>.views and so on) and your question, with instructions to answer only from that list. The answer comes back in parts:
- Answer
- One or two sentences that answer the question, or say the data can't.
- Facts
- Measured values, each citing exactly one metric, shown with its number.
- Estimates
- Forecast values or simple calculations from facts, each with its metric.
- Hypotheses
- Possible causes the data can't confirm, with the metrics they rest on.
- Recommendations
- Concrete next steps, with the metrics behind them.
- Missing
- What the question needs that isn't measured, such as revenue or ad spend.
Before the answer reaches you, the server checks every citation against the metrics it sent. A fact citing a metric that doesn't exist is dropped, as is a hypothesis or recommendation resting on one. The model is also told to say when numbers are too small to mean much, roughly under 30 visits or 5 leads in a period, rather than reading a trend into chance.
What it costs and its limits
- Questions use your allowance. Each one is a model call and counts against your daily and weekly allowance like a build; past the allowance it's paid from credit packs. The page itself costs nothing to view. See usage allowance and pricing.
- Only Anticode's own data. No revenue, ad spend, churn or anything else from outside. The analyst lists those under Missing instead of guessing.
- Site-level, not page-level. Visits are counted per site and source, so it can't say which page people read.
- Read-only. It suggests next steps; it doesn't change your sites.
- A few questions an hour. Questions are rate-limited per person.
If you report to clients each month, the workflow for a monthly website report shows how these numbers fit into one.
Questions people ask
Can the analyst make up a number?
It's instructed not to, and any fact it gives must cite a metric from your snapshot. Claims citing a metric that isn't there are removed before the answer is shown.
When does the forecast appear?
After a published site has 14 days of history. Until then the page says a forecast needs 14 days.
Does Intelligence know my sales?
No. Payments go through your own provider's checkout links, so Anticode can't see them, and the page says revenue isn't measured.
Do analyst questions cost anything?
Each question is a model call and is counted against your allowance like a build. Viewing the dashboard is free.