Four parts, one loop
Three of these are working today. The fourth is in development and we'd rather say so than list it as a feature.
Signal radar
Establishes that something moved unusually for this client. Not that a topic is popular in general — that a specific thing changed, for them, relative to their own history.
It runs on statistics, not on a language model. That keeps it cheap enough to run continuously and makes every decision it takes explainable to a client.
Where the signals come from
- Search demand on the client's own property, query by query
- Competitor pages appearing, changing, or disappearing
- Competitor ads launching, and which ones keep running
- Category and forum discussion where the language is shifting
| What it throws away | Why |
|---|---|
| Low-volume noise | Two impressions becoming eight is +300% and means nothing |
| Seasonal repeats | Compared against the same week last year, not just last month |
| Your own campaigns | A rise you caused is not a discovery, and reporting it back costs credibility |
| Flat but large | High volume with no movement is a fact, not a signal |
Each rule and its current failure modes are written out on the method page.
Fit and timing
Anyone can list what's rising. The question your team actually has is whether it applies to this client and whether there's still time. That judgement is the product.
It rejects freely
Most of what gets detected doesn't apply. The engine is built to say so rather than manufacture a connection, because a plausible-sounding bad suggestion costs more time than silence.
It matches the buyer
A smart thermostat story is about heating, but if your client sells to commercial contractors then the buyer is wrong and the signal is noise. Topic adjacency is the most common way these tools fail.
It respects tone rules
If a client never does lifestyle content, a well-fitting trend that needs lifestyle content gets marked down, with that as the stated reason.
It shows its working
Every judgement cites the part of the brand profile it used, so your account manager can disagree with it specifically instead of distrusting it generally.
It puts a clock on it
Act now, this week, monitor, or too late. A trend you can't reach in time is worse than one you never saw, because it costs a planning meeting.
It hands over a draft
An angle and a first draft attached to each item, so the suggestion arrives as work already started.
Client reports
The monthly report is where agency hours quietly disappear. Windward assembles it from what actually happened: what was surfaced, what your team acted on, and what followed.
It goes out under your brand. There's no badge and no footer credit, because a client seeing a third-party logo is a conversation you don't need to have.
What's in it
- What moved in the client's market this period
- What your team chose to act on, and what it chose to skip
- What followed, described as sequence rather than as proof
- What's building for next period, with windows
Recording outcomes is the point, not a feature
Every time someone marks an item used, changed, or ignored, the system learns something about that client and about which kinds of signal are worth surfacing at all.
A trend tool without this gets louder over time. With it, the weekly list gets shorter and better, which is the only direction worth travelling.
Reports are available on the Practice plan and above. The free plan sends the weekly list only.
Risk radar
The same engine run backwards. Instead of finding what a client could act on, it finds what's moving against them — a competitor's positioning attack, sentiment turning in a specific segment, a category headwind, a journalist circling.
For some clients this is worth more than opportunity spotting, because the correct response is to tell someone quickly rather than to publish anything at all.
| Status | Detail |
|---|---|
| Detection | Working — same rules, inverted |
| Escalation routing | In development. Needs to reach a person, not a dashboard |
| Availability | Studio plan, when it ships. Not billed until then |
If this is the part you need most, say so — it moves up the queue based on who asks.
Start with the parts that work
Signal radar and fit judgement are working today. Try them on one client and judge for yourself.
No card. One workspace stays free permanently.