Manual pilot result report tutorial
A week 2 tutorial for collect real feedback through one inspectable output.
When this is the right output
Use this after repeated pain is confirmed and real material is available.
Prepare before you start
- Week-one output and feedback log
- 1-2 people willing to keep talking or testing
- Real material needed for this week's output
Public case breakdown
How Stellantis &You UK used Make for aftersales messaging
The aftersales team was overloaded by phone and in-person communication. In Make's case, intent and sentiment classification routed ordinary messages to automation and subtle dissatisfaction signals to human follow-up.
Source
Make customer care success story47,000+ messages analyzed and 18,000+ handled automatically over 12 months
It did not start as a huge system
Classify first
The workflow judges intent and sentiment before attempting any response.
Keep human boundaries
Negative, ambiguous, and risky messages are escalated to people.
Track outcomes
The team measures handled volume, escalation, and customer feedback instead of showing only a workflow diagram.
What to copy
Sell saved time, not the diagram
Clients buy less repeated work and fewer missed issues, not the automation canvas.
Automate low-risk work first
Classification, sorting, notification, and logging are safer first steps than full auto-replies.
Price from before-after evidence
Record manual time and post-automation time before quoting a pilot.
Build a small version
Collect 20 historical messages
Pick one niche such as clinics, training firms, or local services.
Create 3 labels
Use only no reply needed, needs human, possible churn.
Run 5 reviewed tests
Let AI classify, review manually, then calculate accuracy and time saved.
Do not promise fully automated support first. Message triage and human alerts are easier pilots to sell.
Build the first version
Pick the strongest signal
Choose the person closest to real demand from week-one replies, forms, or DMs.
Collect real material
Ask only for material needed to create this week's output.
Deliver once
Use AI plus manual review to create a result the user can inspect.
Record feedback
Ask what is usable, what is not trusted yet, and whether they want the next step.
Copy-ready templates
Feedback questions
Which part can you use directly? What is not trustworthy yet? If we continue, what should version two change first?
Delivery log
Input material, process, human edits, delivery time, user feedback, next-step intent.
How to decide the next step
Continue when
- The client confirms saved time or metric improvement
- The client wants to continue the pilot
- Human review boundaries are clearer
Change direction when
- Feedback is polite but creates no next material or trial intent.
- The result does not enter a real workflow.
- Every delivery feels like a new custom project.
Return to the result after you finish this output
In week two, run one manual AI-assisted pilot with real data and prove time saved or metric improvement.