Skip to main content
Most AI products hallucinate before admitting limits. Forepost takes the opposite stance:
“I won’t guess from zeros. That’s not what a chief of staff does.”
This document explains the rule, where it applies, and what you’ll see when it triggers.

The rule

Every AI surface in Forepost (Daily Brief, Horizon, Board Pack, AI Business Case) checks workspace completeness before generating. If you don’t meet the bar, the surface returns a templated message in the Forepost voice naming exactly what’s missing. No Anthropic call. No fabricated story. The threshold: at least 4 of the 5 core metrics filled, plus at least one named agent. Specifically:
  • Volume (core)
  • CSAT (core)
  • First response time (core)
  • Open tickets (core)
  • Agent utilisation (core)
  • AI deflection (optional)
  • Backlog age (optional)
  • One-touch resolution (optional)
A field is considered filled if it’s a non-zero, finite number. Empty strings, zeros, and null all count as not filled, because in support data, real zeros are extraordinarily rare and almost always indicate a missing input.

Where you’ll see it

The CompletenessBanner

An amber strip appears at the top of the Daily Brief, Horizon, and Watchlist pages when the workspace isn’t ready. It names the missing fields with a one-click link to Settings:
I need CSAT score, first response time, and open tickets before I can give a real read. I won’t guess from zeros.

Inline brief text

If you click into a brief, you see the templated message instead of streamed prose. Same for Board Pack and AI Business Case.

Email

The hourly cron checks completeness before sending. If your workspace isn’t ready, no email is sent. Better silence than a bad brief.

Manual sends

Send brief now and Send digest now return a 400 with a structured response listing what’s missing.

What about Ask Forepost?

The chat surface isn’t pre-gated, you can ask anything, even with no data. Instead, the system prompt has an explicit honesty clause:
A metric showing “0” almost certainly means the user has not entered it yet. Never extrapolate from a zero. If a relevant metric is missing, name that gap explicitly rather than inventing a story.
So if you ask “what’s driving my CSAT trend?” without CSAT entered, the model will say so directly rather than guess.

Why this matters

Trust is the substrate. A CX leader who forwards a Forepost brief to their CEO needs to know it’s grounded in real numbers, not an inference from absent fields. The first time a leader catches Forepost making something up, the product is dead in their mind. Honesty over hallucination. Every time.