Confidence Ladder

AI CONFIDENCEENABLE

Purpose Stage-gate AI use so teams rack up safe, fast wins before high-stakes rollout.
Intervention type Behavioural scaffolding & risk-managed adoption
Lead AI programme or team leader
Time 45 min solo, 90 min group

Expected outcomes

  • User sees a clear, low-risk path to try AI

  • Team logs measurable wins on a public ladder

  • Business gets data on speed/quality deltas

  • Org aligns risk appetite with real safeguards

What to bring to the session
  • Confidence-Ladder Board (build via the steps below)

  • Ladder Cheatsheet (below)

Steps

1 | Set up

Step 1: Build the Board
Show the Confidence Ladder (Shadow, Co-pilot, Autopilot) with a blank KPI box visible to all participants.

Step 2: Shortlist Anchor Tasks
Ask participants to identify workflows that:

  • Take ≤ 30 minutes

  • Recur often

  • Are safe to experiment with


Think about:

  • “Would failure here cause real damage?” (Yes → find a new option)

  • “Do we already have workarounds?” (Yes → good friction point)


2 | Task Selection

Step 3: Vote on One Anchor Task
Use dot voting to agree on one priority task.

Step 4: Capture the Baseline
Log baseline data using available metrics or team estimates.

Think across:

  • Time

  • Effort

  • Quality

  • Happiness

3 | Build the Ladder

Step 5: Fill in the Shadow Rung
Define how AI will support without taking over:

  • AI role

  • Human role

  • Guard-rails required (e.g. sandbox only, human approval required)


Tip: Not sure on what this rung looks like? Read the cheat sheet below.

Step 6: Define Co-pilot and Autopilot Rungs
Repeat for each level.

Step 7: Set Risk Gates
Complete the Risk-Gate for each rung:

  • Where can this run (pilot, live)?

  • What would trigger rollback?


Tip: Struggling? Use the Risk-Gate Sheet below to help.

4 | Lock the Loop

Step 8: Schedule the 3Fs Stand-up
Set a recurring 15-minute check-in using the 3Fs:

  • Facts: Did AI save time or match quality?

  • Feelings: Confidence rating (1–5)?

  • Fixes: One improvement for tomorrow

5 | Launch & Progress

Step 9: Assign Owners
Nominate:

  • Ladder Owner → runs stand-ups

  • Scorekeeper → logs metrics

Step 10: Launch the Shadow Rung
Post the ladder in Slack and start Shadow Mode today.

Step 11: Advance Based on Readiness
Only move up if:

  • AI maintains or improves quality

  • Confidence ≥ 80%

  • No rollback triggers for 2+ days

Resources

Risk-Gate Sheet

How to use

  1. Tick the highest threshold you’ll allow at this stage.

  2. Tick / fill triggers that matter for this workflow.

  3. Post the completed sheet next to your Confidence Ladder so everyone knows the fail-safes.

Ladder cheatsheet

Level 1: Shadow

AI suggests. Humans decide.
This is the safest place to start. AI can generate ideas, drafts, or suggestions... but nothing goes live.

What this looks like:

  • AI gives you headline options, summary suggestions, or draft text.

  • You decide what (if anything) to use.

  • You’re still doing the real task, AI is just a silent helper.

Examples:

  • Suggested email subject lines

  • Headline ideas for a report

  • Summary of a Slack thread

Level 2: Co-pilot

AI drafts. Humans edit.
Here, AI takes the first pass, but humans still check and approve everything before it goes out.

What this looks like:

  • AI writes the first version

  • You review, improve, and approve the final result

  • The task gets done faster, but you’re still in control


Examples:

  • First draft of a weekly update email

  • Slide deck based on meeting notes

  • AI writes a KPI report, you check the data and add commentary


Level 3: Autopilot

AI does most of the work. Humans spot-check.
Now AI runs the task by default, and humans step in only when needed. This is only for trusted, proven workflows.

What this looks like:

  • AI generates the final version automatically

  • You check 1 in every 5 outputs

  • You trust the process, but you’re still watching for red flags


Examples:

  • AI sends daily news summaries

  • AI fills in CRM records after calls

  • AI generates standard usage reports every month

Other methods within the enable block