Human-First AI Automation
Perfect the service with people, then automate the proven experience with AI.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 98%
Human-First AI Automation begins with an intentionally unscalable service delivered by people. The team uses that manual version to discover what excellent looks like, including the questions customers ask, the context staff need, and the judgments that create value. Calls, feedback, and outcomes become the specification for the automated system. Only then does the team encode the motion into prompts, workflows, models, and interfaces, gradually removing human labor while comparing results with the original benchmark. This reverses the usual growth-team instinct to begin with a touchless flow. AI supplies scale, speed, and around-the-clock availability, but humans first establish the quality standard that the system must approximate or exceed.
Origin
Extracted from Marketing Against The Grain while Kieran Flanagan explained how he would validate personalized automation guides through sales and customer-success teams.
Core principles
- 01Start by discovering the best possible experience, not the cheapest scalable one.
- 02Human delivery reveals the context and judgment that automation must reproduce.
- 03AI should replace a validated motion rather than an imagined workflow.
- 04A nearly human-quality experience can still win through speed and continuous availability.
How to run it
- 1
Choose a valuable customer motion
Select one interaction where better guidance, personalization, or responsiveness could materially improve the customer experience.
Pro tip Begin with a narrow job to be done that has an observable outcome.
Watch out Do not automate a vague journey that lacks a clear success condition.
- 2
Build the best manual version
Have a skilled person deliver a highly personalized version without worrying about immediate scalability.
Pro tip Allow the person to use all available customer, role, and product context.
Watch out Cost optimization at this stage can hide what the ideal experience actually requires.
- 3
Test it with real customers
Route the experience through sales or customer success and observe whether customers find it useful.
Pro tip Listen to calls rather than relying only on summary metrics.
Watch out Internal enthusiasm is not proof of customer value.
- 4
Capture the human mechanism
Document the inputs, questions, decisions, explanations, and follow-up actions used by the human operator.
Pro tip Record exceptions as well as the common path.
Watch out Automating only visible outputs without capturing reasoning creates brittle systems.
- 5
Engineer the human out
Reproduce the validated motion using AI, prompts, workflow automation, and product interfaces.
Pro tip Remove human steps incrementally so failures can be traced.
Watch out Do not eliminate escalation paths before the automated experience is reliable.
- 6
Benchmark and refine
Compare speed, quality, availability, and customer outcomes against the manual version, then iterate.
Pro tip Accept a slightly lower answer quality when faster and continuous service creates a better total experience.
Watch out Scalability alone does not make the automated version superior.
In the wild
A team manually prepares automation recommendations based on each prospect's role, domain, technology stack, and jobs to be done. Sales representatives walk prospects through the guide while the team listens for confusion and value signals. Those observations are then used to create an AI-generated version.
→ The team validates demand and quality before attempting touchless delivery.
A software company first assigns specialists to guide new customers through setup and records the questions, dependencies, and recurring decisions. It then builds an AI concierge that handles the proven common path while escalating unusual configurations.
→ Customers reach initial value faster without losing human help on complex cases.
Common mistakes
Automating an unproven experience
Starting with a scalable interface can efficiently reproduce a journey customers never valued.
Ignoring human judgment
A workflow diagram may omit the contextual decisions that make the manual service effective.
Removing escalation too early
Customers need a reliable path to human support while the AI system is still learning its limits.
Is it for you?
Best for
Growth, sales, and customer-success teams building personalized AI-led experiences.
Not ideal for
Processes where experimentation could expose customers to serious safety, legal, or irreversible harm.
From the transcript
“you should start by putting the human in the loop first, and getting the best experience that is unscalable, 'cause it's with a human. And…”
“And AI's incredible 'cause you can engineer the human out of the loop and actually just pass that off to AI, and that's the way…”
“I wanna start at like, what's the best? it's unscalable cause we can't afford to have humans doing these motions, but what's the best version…”
From the episode
The State Of A.I. In 2023: Useful Tools Vs Hype (#120)