Skip to main content
SiteSuper runs a continuous loop: it reaches out to your superintendents on a schedule, turns whatever they say into a structured report, watches those reports for trouble, and — when it finds some — runs an automated follow-up chain on your behalf. This page traces that whole flow so you have a clear mental model of what happens, and when.

​
The flow, stage by stage

1

Scheduled outreach

Every minute, SiteSuper checks your active schedules. When a schedule’s day and time match (in its own timezone), it contacts the super on the configured channel — a recorded phone call with the voice agent, an SMS, a WhatsApp message, or an email.If a call hits voicemail, SiteSuper automatically follows up with an SMS and retries after your configured callback window. It won’t contact the same super twice for the same schedule on the same day.
2

Report collection & parsing

However the super responds — talking to the voice agent, texting back, replying on WhatsApp, or emailing — the raw message comes back to SiteSuper. The AI parses it into a structured report: work completed, workforce, issues, delays, weather, and inspections. Phone calls are recorded and transcribed; photos sent over WhatsApp are attached.The report is also checked against your project schedule (the imported Gantt) and flagged if something looks off.
3

Variance detection

Each new report is run against your project’s variance definitions — the rules that say “this situation needs follow-up.” A definition’s detection rules look for signals in the report (a materials issue, an idle trade, a failed inspection, and so on).When a rule matches, SiteSuper figures out which scheduled task it relates to, finds the right contact to chase (e.g. the vendor active for the current milestone), and scores its confidence. Above the confidence threshold, a variance opens — tied to that task, that contact, and a snapshot of the evidence.
Every report is detected against exactly one project — the one its schedule/conversation belongs to. For a super who works several projects, SiteSuper attributes the report to the right job (the schedule that prompted it, or the site they tell the agent) rather than checking it against all of them. If a report genuinely can’t be attributed, it’s held as “Needs project” for you to assign — never run against the wrong project. See Schedules and Reports.
4

The action chain

A new variance immediately starts its definition’s action chain — an ordered sequence of steps you design (see the variance definition builder). Each step is one action: send a WhatsApp/SMS/email, place a call, send a calendar invite, let the AI decide what to do next, escalate, or wait.Every step is either sent automatically or held for approval:

Send automatically

The message/call goes out right away (respecting quiet hours), and SiteSuper waits for a reply or a timeout before moving on.

Require approval

The drafted action waits in the project’s Approvals queue for a manager to review and send. The “Awaiting approval” card on the Outcomes page turns amber, then red if it sits past 48 hours.
5

Resolution, escalation & SLAs

The chain advances based on what happens:
  • A reply comes in — SiteSuper links the vendor’s response to the variance and moves to the next step (an AI decision step can branch on what the reply says).
  • No response / a step times out — the chain moves to its next step, which often steps up the pressure (text → call → calendar invite → escalate).
  • It reaches the end / a “done” step — the variance is resolved.
  • It blows an SLA — if a variance isn’t reviewed within its review SLA, or isn’t closed within its close SLA, it escalates and the Owner is notified.
Managers can Pause a chain at any time (the engine stops advancing it) or Resolve / Dismiss the variance outright from the variance page.

​
Where you control the flow

Schedules

Set when, how often, and on which channel each super is contacted.

Variance definitions

Design the detection rules and the action chain — visually, no JSON required.

Approvals

Review and send anything a step held back. Off by default — turn auto-send on per project in Settings.

Outcomes

See open, escalating, and awaiting-approval variances at a glance, plus the weekly summary.

​
Designing the flow: variance definitions

A project’s Variances → Definitions page is where the whole automation is authored, in three parts that map exactly to the flow above:
  1. Detection rules — when does this variance open? One or more signals (match any of / require all of / exclude). This is stage 3.
  2. Applies to — who does the chain chase? The contact role (vendor, sub, inspector…) and whether they’re scoped to the current milestone or task.
  3. Action chain — what happens, step by step? Each step’s type, tone, whether it needs approval, the message template, and — for AI-decision steps — the branches that route to the next step. This is stages 4 and 5.
Everything is editable as a visual builder, but there’s an Edit as JSON (advanced) toggle for power users who want the raw config.

​
Good to know

  • Approvals are off by default. New projects require a manager to approve AI-driven outbound. Flip a project to “Auto-send everything” in Settings only when you’re ready — that override makes every step send automatically regardless of its individual setting.
  • Quiet hours apply to automatic sends, so the system won’t message a vendor in the middle of the night.
  • Versioning is forward-only. Editing a definition bumps its version; variances already in flight keep running on the version they opened against.
  • Everything is recorded. Calls are recorded and transcribed, and every step, reply, and decision is captured in the variance’s audit log.