Automation

What it is

Every Done Deal feature that can act on its own, or get a step ready for a person, runs at one of three levels:

1. Manual — the team does the step by hand, to learn what works.
2. Semi — the system prepares the step and a person approves it.
3. Automatic — the system does the step on its own, within a daily cap.

The level is set per feature, so one feature can run on its own while another is still done by hand. It can also be set for a single account, so a brand that is ready for more automation can have it while everyone else stays where they are.

The level is the answer to one question: for this feature, for this account, does a person have to be in the loop today?

The three levels

Moving a feature up a level is a decision about trust, not a technical setting. Most features that touch a person outside the team start below automatic on purpose, so the team can watch a few real cases first.

Which features are on it

The cap belongs to the feature, not to an account: it is the most automatic runs that feature may make in a day across everybody. The page shows it as "used today / allowed today", so the team can see how close a feature is to its limit.

Today only deal seeding checks its level and cap before it acts. The other rows are on the page so their level is set in one place once they are wired to it, but for now each still runs by its own switch, whatever its row says. The weekly update sends, dry-runs or stays off by its own switch. The first-ten questions after onboarding have their own on/off switch too; the First ten deals row describes suggested next steps for a new account, not that question series.

What each level means in practice

The words on the page change with the feature, because "prepare it" means something different for each one. Outside deal seeding, these describe what each level is meant to do once the feature follows it:

• Research on semi is queued; a person starts each run. On automatic it runs when an account needs it.
• Matching on semi runs the matches, and profile enrichment waits for a person. On automatic, both run. (Enrichment is not connected yet, so today matching never sends a thin profile to research.)
• Inbound IQ on semi prepares a suggestion and a person accepts it. On automatic, the suggestion is filed on its own.
• Offer pricing on semi suggests a price and a person confirms it. On automatic, the suggested price is filled in — it is still only a suggestion, not a decision.
• Weekly update on semi is a dry run: emails are built and shown, nothing is sent. On automatic, emails are sent on schedule. (Today the weekly update's own switch decides this, not the row.)
• Deal seeding on manual needs a click for every step: approve, ask the talent, ask the brand. On semi the team approves each seed; then the asks and the intro run by themselves. On automatic, strong seeds are approved without the team, inside the cap.
• Outreach on semi drafts each message and a person sends it. On automatic, messages go out inside the cap.

Each feature's row on the page says what its current level means in one plain sentence, so nobody has to remember the table.

Setting it for one feature, or for one account

• For a feature: pick 1 Manual, 2 Semi or 3 Auto on that feature's row. That is the level for every account, until an account has its own setting.
• For one account: open the feature's overrides, paste the brand, talent or agency account, pick a level, and optionally a tighter cap. Removing the override puts the account back on the shared level.
• An agency's setting covers its talent. If a talent is managed by an agency, the agency's level applies to that talent. When several agencies manage the same talent, the most careful of their levels wins.
• The order is fixed: the account's own setting, then its agency's, then the shared setting, then the level the feature started at.

High-risk features need a confirmation

Deal seeding and outreach are marked High risk. Moving either of them to Automatic asks for an explicit confirmation first, naming the feature and the daily cap: it will act on its own, at most that many times a day.

Nobody can turn one on automatic by a mis-tap or a stray save. A feature that is already automatic does not ask again when only the cap changes.

Every change is recorded

Each feature keeps its own history of changes. For every one the team can see:

• who made it
• when it was made
• what the level was before and what it is now
• what the cap was before and what it is now
• any note that came with it

The latest changes are listed at the bottom of the page, newest first. Older ones stay with the setting they belong to, and an account override carries the same information as a shared one.

The cap and the off switch still apply

A level is permission, not a promise. Even at automatic, a feature acts only when both of these allow it:

• The daily cap has room. When the cap is reached, the feature stops acting on its own for the rest of the day. The counter starts again the next day (days run midnight to midnight UTC), and an account with its own cap can be held to an even tighter number.
• The feature's own off switch is on. Each feature keeps the switch it already had, and the page shows whether it is on, off or not set. An unset switch counts as off for the features that need an explicit go-ahead.

So raising a level never overrides a cap and never overrides an off switch. It only decides who does the work when the feature is otherwise free to act.

When it cannot run, it waits

If a feature cannot act automatically — it is not on automatic, the cap is full, the off switch says no, a per-account limit is reached, or the count cannot be read — the step does not fail and nothing is lost. It waits for a person, exactly as it would on semi.

Deal seeding shows what that looks like: a seed that does not get an automatic slot stays proposed and waits for the team, with a note saying why it was not approved. Features wired to their level later follow the same shape. The counters also fail closed: if the system cannot read how much of the cap is used, it says no and a person acts.

Automatic also never overrules a check inside the feature. A seed its judge says to drop is never approved automatically, whatever the level. Only a drop blocks it: a seed the judges did not score (judges off, budget used up, or the call failed) can still be approved automatically.

What the Automation page shows

The Automation page is one table, one row per feature:

• What the feature does, its risk, and what its current level means right now.
• The level buttons, with the current level filled in and the shared or starting level noted underneath.
• Used today against the cap. Clicking the number changes the cap for that feature, or resets it to the default.
• The off switch, named and shown as on, off or not set, in red when it is stopping the feature.
• Overrides. How many accounts have their own setting. Opening it lists each account with its level, its own cap, how many automatic runs it has used today, who set it and when, a Remove button, and a short form to add another.

Under the table, Changes lists the most recent adjustments with who, when, the feature, the account or "globally", and the level it moved between.

What it does not do

• It does not make a feature safe on its own. A high level still relies on the cap, the off switch and the checks inside the feature.
• It does not replace the feature's own screens. The team still approves, sends and reviews where they did before; the level only decides whether they have to.
• It does not grade the results. It records what changed and how much ran, not whether the automatic runs were good ones.

Related Documentation

• Weekly update — listed here on semi; its own switch starts it as a dry run
• Deal seeding — the high-risk feature that asks both sides before a deal exists
• Outreach queue — the other high-risk feature, and what its queue looks like
• First 10 — the short questions after onboarding, which run by their own switch