The difference between AI projects that succeed and those that fail almost always comes down to planning. Projects that rush into tool selection and configuration without clear scope, success criteria, and realistic timelines tend to lose momentum, exceed budgets, and deliver disappointing results. Projects that invest time upfront in rigorous planning execute faster, stay on budget, and deliver measurable value.

This lecture walks you through the planning phase of your AI pilot—the foundational work that makes everything that follows possible. You'll learn how to identify good pilot candidates, define scope with precision, set success metrics that actually matter, and build timelines and budgets that are realistic rather than optimistic.

By the end of this lecture, you'll have a framework for evaluating which AI ideas are worth pursuing and the discipline to plan them properly.

Why Pilots Matter More Than You Think

Before you invest in a company-wide AI transformation, you run a pilot. Not because pilots are mandatory—though they should be—but because pilots work. They let you test your approach on a small scale, learn what actually works versus what seemed like a good idea in theory, build internal capabilities and confidence, gather real data to justify larger investments, and course-correct quickly if needed.

A well-executed pilot also creates organizational momentum. When people see AI working in practice—solving a real problem, delivering measurable results—skepticism converts to support. Pilots turn abstract concepts like "AI could help us" into concrete examples like "AI is already helping us here."

For a small business, the pilot phase is even more critical. You don't have the budget for massive mistakes, the team capacity to manage complexity, or the tolerance for years-long transformation projects. Your pilots need to be surgical: solve one specific problem well, prove it works, and then expand from there.

Choosing the Right Pilot Project

The first step in planning is choosing which problem to solve. Many businesses approach this backwards—they start with a tool ("Let's use your AI tool") and hunt for problems it can solve. This almost always fails because you're forcing solutions onto problems rather than solving problems you actually have.

The Three Criteria for a Good Pilot

A strong pilot candidate meets three criteria, each of which you should evaluate explicitly.

1. It solves a documented, recurring problem. The problem should be something your team experiences every day or every week, not a theoretical "it would be nice if." Ask your team: what's the single most frustrating, time-consuming task that you wish you didn't have to do? That's your problem. Write it down. Quantify it. How many hours per week does this problem consume? How much money does it cost (in wasted time, lost efficiency, missed opportunities)? A good pilot solves something that hurts right now.

2. It's technically solvable with tools you can reasonably access. Not every business problem is a good AI problem. Some problems need process changes, people changes, or different technology entirely. Before committing to a pilot, assess: Does this problem have enough data to train on, or does it mostly involve tacit knowledge that lives only in people's heads? Is the solution reasonably self-contained, or does it require integrating with five different systems that don't talk to each other? Can you solve 80% of the problem with existing off-the-shelf AI tools, or would you need custom development that's beyond your current capacity? Ruthlessly honest answers here save you months of wasted effort.

3. Success is measurable and achievable in 8-12 weeks. You need to be able to answer "did this work?" with data, not opinions. The metric doesn't have to be perfectly precise—"we saved about 5 hours per week of admin work" is fine—but it should be specific and grounded in reality. And you should be able to achieve measurable success within your pilot timeframe, which should be 8-12 weeks. Projects that stretch longer lose momentum, run out of budget, and accumulate scope creep.

Identifying Pilot Candidates

Start by soliciting ideas from your team. The people doing the work every day have the best ideas about what could be automated or improved. Distribute a simple survey: "What task do you spend the most time on that feels repetitive and frustrating? What would you automate if you could?" Collect 10-15 candidate ideas.

Then score each candidate against a simple rubric:

Criterion	Scoring (1-5)	What to Look For
Impact Potential	Hours/money saved weekly	5 = saves 10+ hours/week | 1 = saves <1 hour/week
Technical Feasibility	Data availability & complexity	5 = clear data, straightforward solution | 1 = unclear data, needs custom build
Ease of Execution	Resource requirements	5 = existing tools, 1 person, <4 weeks | 1 = custom solution, team effort, >16 weeks
Cross-functional Buy-in	Stakeholder support	5 = executive sponsor, team champion, no politics | 1 = contested, low support, political

Multiply the scores (not add—multiplication means all criteria must be reasonably strong). The pilots with the highest composite scores get serious consideration. The ones scoring in the middle get refinement—are there ways to simplify them? The ones scoring lowest—below 100 when multiplied—are probably not ready for pilot status.

Defining Pilot Scope and Success Criteria

Once you've chosen your pilot, the next step is defining exactly what you're going to do—and equally important, what you're not going to do. This is where a scope document becomes essential.

What Goes Into a Scope Document

Your scope document doesn't need to be 50 pages. A clear, single-page scope is better than a bloated document nobody reads. The goal is to force clarity on your team—shared understanding of what you're trying to do and why.

Setting Success Metrics That Actually Work

The most common mistake in pilot planning is setting metrics that are too vague or too hard to measure. "Improve efficiency" is too vague. "Increase productivity by 47.3%" is probably too precise (you won't be able to measure that accurately for a pilot). The sweet spot is metrics that are specific, grounded in reality, and measurable.

Good pilot metrics:

• Before: "Customer service emails take 4-5 hours per batch to respond to." After: "Response time is <1 hour, with AI drafting initial responses."
• Before: "We manually categorize 200 invoices per week." After: "AI categorizes 80% of invoices automatically; manual review is 15 minutes per batch instead of 3 hours."
• Before: "Lead qualification calls take 30 minutes each." After: "AI screening reduces time to 10 minutes per qualified lead, with 90% of the pre-call research done automatically."

Notice these metrics compare a before state to an after state, are grounded in actual time or volume, and are achievable within the pilot window.

Resource Planning and Timeline

Now you know what you're building and how you'll measure success. Time to plan resources and timeline realistically.

The Four Phases of a Pilot

Most AI pilots follow a predictable structure. Understanding these phases helps you build accurate timelines.

Phase	Duration	Key Activities	Common Blockers
Setup & Planning	1-2 weeks	Tool selection, account setup, data access, team training	Slow vendor responses, data governance issues, access delays
Configuration & Testing	2-4 weeks	Tool configuration, workflow design, sample data testing, edge case handling	Tool limitations not apparent until testing, need for custom integration, quality issues
Limited Pilot	2-4 weeks	Run with small user group (1-3 people), gather feedback, iterate	Low adoption due to skepticism, workflow not intuitive, real data uglier than test data
Measurement & Decision	1-2 weeks	Collect metrics, compare to baseline, determine next steps (scale, refine, kill)	Metrics weren't tracked during pilot, too much variation to draw conclusions, unclear causation

Building Your Budget

For a typical small business AI pilot, expect to invest:

• Labor: 200-400 hours (the biggest cost). Usually 1-2 people at 50% of their time for 8-12 weeks.
• Tools: $500-$3,000 for most LLM-based pilots; $3,000-$15,000 if you need custom integration.
• External support: $0-$5,000 if you need consulting help with setup or integration.
• Contingency: Add 20% for the unexpected (always happens).

A modest pilot might cost $5,000-$10,000 total ($4,000 in labor + $1,000-$3,000 in tools + 20% contingency). More complex pilots might hit $20,000-$30,000. The important thing is to estimate conservatively and build in buffer—pilots almost always take longer than anticipated.

The Pilot Charter: Your Planning Document

Pull all of this together into a simple Pilot Charter—a 1-2 page document that becomes your north star for the project. Here's the structure:

Avoiding Common Planning Mistakes

Here are the planning mistakes that derail pilots, and how to avoid them.

Mistake 1: Scope Creep. You start with "improve customer service email response" and by week 4 you've added "also integrate with Slack" and "also analyze sentiment" and "also implement in sales." You now have a 6-month project masquerading as a 2-month pilot. Solution: Have a shared definition of "out of scope" and enforce it ruthlessly. If new ideas come up during the pilot, write them down for the next project—don't add them to this one.

Mistake 2: Optimizing Too Much Too Fast. During the pilot, you'll feel pressure to make the solution perfect before expanding it. Resist this. The pilot's job is to validate the approach works, not to build the production system. You optimize in the second phase, after you've proven the concept.

Mistake 3: Measuring the Wrong Things. You measure adoption (yes, people used it) instead of impact (did it save time? did it improve quality?). Adoption is a leading indicator; impact is the real metric. Measure both, but the impact metrics are what matter for deciding whether to scale.

Mistake 4: Insufficient Data or Access Prep. You don't realize until week 3 that the data you thought you had isn't available, or it's in a system you can't easily access. You then spend 2 weeks working around this instead of using the time to build and test. Solution: Audit data and access requirements in your planning phase, well before the pilot starts. Make sure everything is accessible before day 1.

What You'll Learn Next

With your pilot planned and charter in hand, you're ready to move into execution. The next lecture covers —the work of actually setting up your AI tools, configuring them for your specific use case, testing rigorously, and ensuring the quality is production-ready before you expand beyond your limited pilot group.