Start Small, Start Fast
Win within 24 hours, then expand the automation iteratively
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 99%
Start Small, Start Fast converts AI enthusiasm into a sequence of narrow, measurable wins. The team begins by naming one business problem, then repeatedly shrinks the scope until it can produce a useful result within roughly 24 hours. The initial workflow does not need to automate an entire department or handle every exception; it only needs to work and deliver real value. That first win develops practical knowledge of prompts, integrations, data, and internal systems while giving stakeholders evidence that the technology is useful. The team then extends the workflow in small increments, validating each addition before moving on. Momentum, expertise, and organizational confidence accumulate together rather than depending on a risky all-at-once transformation.
Origin
Flo Crivello gave this advice on Marketing Against the Grain after describing customers whose excitement led them toward overly broad AI-agent projects.
Core principles
- 01Reduce ambitious automation ideas to one problem
- 02Prioritize a fast operational win over a comprehensive launch
- 03Use early value to build expertise and trust
- 04Expand only after the current workflow works
- 05Treat adoption as an iterative capability-building process
How to run it
- 1
Identify one valuable problem
Choose a specific source of missed revenue, wasted time, or poor customer experience. State the outcome rather than proposing a broad AI role.
Pro tip Look for frequent, repetitive work with an obvious payoff.
Watch out Do not begin with a goal such as automating an entire department.
- 2
Shrink the scope
Remove secondary use cases, edge cases, and optional integrations until the first version has one primary job.
Pro tip Ask repeatedly whether the workflow can be made smaller while retaining value.
Watch out Excitement often disguises unnecessary launch scope.
- 3
Target a 24-hour win
Define a minimal workflow that can operate and demonstrate value by the following day. Establish a simple success signal such as one booked appointment or one resolved inquiry.
Pro tip Select a use case whose result is directly observable.
Watch out A prototype that merely looks impressive but completes no useful work is not a win.
- 4
Deploy and learn
Run the workflow against realistic cases and observe where prompts, integrations, or internal systems break down.
Pro tip Capture failures as specific improvement tasks.
Watch out Do not confuse the agent platform's capability with the readiness of internal APIs.
- 5
Expand incrementally
Add one new behavior, integration, or use case after the existing workflow is stable. Verify the expanded version before adding more.
Pro tip Use each successful increment to build stakeholder confidence.
Watch out Rapidly adding multiple features makes failures harder to diagnose.
In the wild
A restaurant could begin with an agent that only answers opening-hours questions during peak service. After that simple workflow proves useful, the business could add reservation lookup, booking, modification, and escalation one capability at a time.
→ The restaurant captures immediate value without attempting a complete receptionist replacement.
A software company starts with a voice agent that schedules only 15-minute demo calls for inbound prospects. Once calendar lookup, booking, and follow-up work reliably, the company can add qualification questions or support additional meeting types.
→ The company gains a fast scheduling win and a stable foundation for broader automation.
Common mistakes
Starting with every use case
Trying to support sales, service, qualification, and complex exceptions at launch multiplies failure points before the team has learned the technology.
Demanding perfection before value
Waiting for human-level performance across every scenario delays useful automation that already works for bounded cases.
Expanding before stabilization
Adding capabilities to an unreliable first workflow compounds defects and weakens stakeholder confidence.
Is it for you?
Best for
It is best for organizations beginning AI-agent adoption or trying to prove value to skeptical stakeholders.
Not ideal for
It is not ideal when legal, safety, or infrastructure requirements make a 24-hour production experiment inappropriate.
From the transcript
“the biggest advice I give to people is start small and start fast.”
“let's get a win on the board like tomorrow, right?”
“and then iteratively add to it, little by little by little, by little by little.”
From the episode
Everything You Need To Know About AI Voice Agents in 2025