Skip to content
BotCircuits Space

Using Space Automate

Foresight: goals and watches

Space reads your customer conversations, notices what people are trying to achieve, and proposes the right moment to help. You approve once.

For teams that use Foresight (it is off by default)

Most automations are rules someone had to think of: “if status is X and days idle over 3…”. Foresight writes the rules for you. It reads what you already have, such as customer conversations, interaction notes or an agent's threads, and works out what each person is trying to achieve or what was promised. Then it proposes watches: little automations that notice the right moment to help.

Your dataScanGoals foundProposed watchesYou approveRuns on its own

The pieces

PieceWhat it is
SourceA dataset Foresight reads: the rows of a table (one subject per customer), or an agent's task threads. You add short guidance saying what matters in your domain, and choose which actions (playbooks) watches may run.
ObjectiveWhat one subject is trying to achieve or was promised. Three kinds: a goal (“buying a house next summer”), a commitment (“the loan officer will call Friday”) and a life event (“changing jobs in March”). Each keeps its evidence, a due date if there is one, and a confidence score. It stays open until the data or a person marks it achieved or dropped.
WatchA proposed trigger: on a schedule, once on a date, or when a table row changes. It either runs one of your allowed playbooks, or simply reports to you.

How scanning works

Foresight scans on a cadence you set (or manually with Scan for goals). It only reads what's new since the last scan, and it never re-reads unchanged rows.

  1. It reads each person's history.
  2. It writes down new or changed objectives and proposes at most three watches for each.
  3. Everything is checked, just like creating an automation: valid schedule, a table you can see, a playbook with the right inputs. Anything that doesn't pass is kept with a problem you can fix when you approve.
  4. Duplicates are skipped, and a watch you declined is never proposed again.
  5. You get one notification per scan. It counts as “needs you” and is emailed.

Proposals with low confidence are dropped, and when a goal is marked achieved or dropped, its watches retire automatically.

Starting over

ActionWhat it does
Force scanRe-reads everything from the start. Goals you dropped aren't recreated unless newer data shows they're back.
ReopenBrings a closed goal back.
Start overDeletes the goals and all non-active watches, then reads again. Active watches keep running.

Signal discovery

Besides personal goals, Foresight can look across the whole dataset for patterns worth watching: unusual activity in transactions, accounts or applications. It runs weekly by default, or on Discover signals, and proposes watches the same way.

You stay in control

ActionEffect
ApproveTurns it into a normal automation. You can edit its title, trigger, which playbook it runs and its inputs first.
DeclineWith a note. The note is shown to later scans: it's how Foresight learns what you want.
RetireStops a running watch and keeps its history.
Re-checkRe-runs the checks after you fix something, such as sharing a table.

Approved watches appear on Scheduled or Triggers, so everything there applies: they act as you, pause for approvals on risky actions, and obey playbook review steps.

AI decides when, playbooks decide how

A source lists the playbooks it's allowed to run, such as Draft customer outreach, Remind the relationship manager of a commitment or Goal progress check. For each watch, the AI picks the right one and fills in its inputs. The approver sees “Runs: Draft customer outreach” with the inputs, and can switch to another action or choose Only report to me.

If no action fits, the watch only reports: it reads what it needs and gives you findings and suggestions, without changing records or contacting anyone.