You've configured the building blocks. Now the question becomes how aggressively should the system act on its own? Every major piece of Horizon — triage, assignment, todo creation, watchlist transitions, form lifecycles — has a manual mode and an automatic mode. Picking the right setting per piece, at the right time in the team's maturity, is what separates an account that works with the team from one the team works around. This lesson is the calibration guide.

Why this matters

Problem: Crank everything to fully automatic on day one and you'll auto-misclassify, auto-misassign, auto-escalate to the wrong person, and auto-write the wrong notes back to the helpdesk. The team loses trust and starts disabling automations they don't fully understand. Leave everything manual indefinitely and you've built an expensive ticket queue with no automation payoff. Both failure modes are common.

Horizon's answer: Treat manual vs automatic as a per-piece, per-phase decision. Every major component has a knob. Start most knobs manual. Move them to automatic as the team gains confidence and the underlying configuration produces results you trust.

There's no universally-correct setting. There's the right setting for your team, this quarter.

The maturity arc

Most accounts move through three rough phases:

Don't skip phases. Phase 1 to Phase 3 in a week is how you lose the team's trust. Phase 2 should last weeks at minimum, often a quarter or more. Phase 3 is where you live, but you'll occasionally get knocked back to Phase 2 on a specific piece (e.g. a triage rule starts misclassifying after a helpdesk picklist change — temporarily move triage to require review until you fix the underlying rule).

The per-piece calibration table

For each major component, there's a more manual and more automatic setting. Here's the spectrum across the things you've configured:

For each row, your account is somewhere on the spectrum. Audit where you sit on each one quarterly.

Signals that tell you when to flip a knob

Don't flip knobs on hunch. Flip when you have a signal.

Move a piece toward automatic when:

• The underlying audit log shows clean, predictable behavior over a sustained window (a week or more).
• The team is doing the same manual action repeatedly without judgment calls.
• The cost of a wrong action is low (e.g. wrong assignee = a quick reassign, not a customer escalation).
• The piece is producing measurable benefit even in manual mode that automation would multiply.

Move a piece back toward manual when:

• The audit log shows the system making bad calls more than once a week.
• The team is overriding the system's decisions consistently (a sign of misalignment).
• A change upstream (helpdesk picklist change, new ticket type, team restructure) invalidated a configuration assumption.
• The cost of a wrong action just went up (a bad customer escalation, an SLA breach).

The default direction is toward automation over time. Reversal should be deliberate, scoped to a specific piece, and explained.

Where to look for those signals

Each component has its own audit surface. Make a habit of skimming these:

A quarterly review that opens each of these and looks for outliers will keep the calibration honest.

Quarterly account review — the checklist

Spend half a day per quarter walking the system. The cadence is more important than the depth — small consistent reviews beat heroic annual ones.

1. Onboarding setup tasks — anything still red on the dashboard? Close it out.
2. Triage — open Triage Logs, sort by lowest confidence. Are misclassifications clustered? Add hints, exclusions, or rules.
3. Assignment — open Decision Log. Are assignments distributing as expected? Open Errors. Any patterns?
4. Coverage Report — does on-call coverage match team reality? Any gaps?
5. Watchlists — open Scan Logs. Are stages firing as expected? Any watchlists with zero transitions in the last quarter (i.e. dead)?
6. Lifecycle Workflow Logs — any consistent failures? Pre-filter logs showing rules that should match but don't?
7. Forms — Submissions per form. Any forms with zero use? Any patterns suggesting a form should be promoted (or retired)?
8. Helpdesk integration — Sync run logs healthy? Callback logs healthy? Widget status green?
9. Panel Customization — has the team's workflow drifted from the configured panel? Adjust scope tabs / actions to match current reality.
10. Time Estimate Policies — only after at least one full quarter of NextAvailable data. Are durations roughly tracking actual time-on-ticket?

The goal isn't to fix everything in the review. The goal is to know what the calibration should be, then schedule the changes for the next month.

Common calibration mistakes

A few patterns to actively avoid:

• Cranking everything automatic in Phase 1. Loses team trust. Disables automations get permanently disabled because nobody remembers what they did.
• Never moving past Phase 2. "It's working, why touch it?" Because it's getting worse without you knowing — picklists change, teams restructure, customer mix shifts. Configurations rot.
• Flipping the same knob back and forth. "We tried auto-assign but it was bad, switched back, then tried it again, switched back." If you're doing this, the underlying configuration is wrong, not the manual/auto setting. Fix the config.
• Making changes on hunches instead of audit data. Use the logs. Every component has them. They tell you when something's off; they don't tell you nothing is.
• Treating advanced layers as required. NextAvailable, time estimate policies, complex multi-tier escalation, custom workflows — none of these are required to ship. Some accounts run perfectly well on the basics for years.

When the account is well-calibrated

• The team describes Horizon as "it just works" more often than "I'm fighting it."
• Manual interventions are rare and surgical, not routine.
• Quarterly reviews surface adjustments, not crises.
• Audit logs show high success rates with low override rates across components.
• New team members can onboard in days because the system has absorbed the rules instead of leaving them in someone's head.

You'll never reach a frozen "perfectly configured" state. The team changes, the helpdesk changes, the customers change. Calibration is ongoing — that's the job.

Tip: The single most useful question to ask in a quarterly review is "what manual action did the team do most often this quarter?" That action is almost always a knob you could flip toward automatic, or an underlying config you could tighten. Build the habit of asking it. The automation opportunities will surface themselves.