Fantasy’s Four-Step Framework for Deciding Where AI Should Work

By Grace Kim

Fantasy set two measurable targets—meaningfully improve at least half its activities and put AI into at least three internal products—then used a four-step framework and a four-week process to rank projects.

Fantasy designs and builds digital products. When ChatGPT came out, the company said it knew right away that the technology would change its industry. The article’s author says most leaders already know they need AI but do not know where to start or how to scale a successful first experiment across the company.

The framework has since guided executive groups at Harvard Business School. The author says a company team can run it in a few weeks.

What should a company measure before deploying AI?

Before touching a tool, Fantasy set two targets: at least half its activities should improve in a meaningful way, and at least three internal products should use AI. It put small teams and KPIs behind those goals. For teams already experimenting without targets, the author’s instruction is simple: set them now.

1. Identify. Map one workflow from start to finish, such as the path from a client brief to delivered work. Name the owner of each step and mark where the process wastes effort. Then check what AI can handle today.

Fantasy says AI agents now go beyond summaries and first drafts: they can research, analyze data, write and ship code, and carry out multistep tasks across several tools without a person handing off each stage. A capabilities check from six months ago may no longer describe what is possible.

2. Assess. The source recommends scoring each task on five factors:

• Complexity
• Frequency
• Value
• Risk
• Implementation difficulty

The source contrasts a low-risk task performed 50 times a week with one performed twice a quarter that could reach a regulator. Those five scores help teams identify the clearest initial payoff.

3. Decide. Every task receives one of three labels:

• Automate: Low-value tasks that repeat constantly.
• Augment: High-value tasks, with a person still owning quality while AI makes the work faster.
• Preserve: Tasks that depend on judgment, customer relationships or accountability for the outcome.

4. Operate. For each candidate, specify whether the change refines an existing process, adds something new or transforms how the work gets done.

The four steps turn a vague AI mandate into measured targets, scored tasks and an explicit human role.

How did Fantasy rank its first AI projects?

After reviewing the task list, Fantasy plotted every candidate on a map with business value on one axis and ease of implementation on the other. The company started where value was high and the build looked easy. It worked through the entire process in four weeks.

The first quick wins came from inside Fantasy’s operations, such as financial forecasting and a research assistant for its new business team. The largest returns came from Synthetic Humans and Fantasy Canvas, two proprietary tools for synthetic user research and content production. Both now sit at the center of the company’s client work.

Fantasy also recommends resisting the urge to polish before a tool proves itself. Its first AI tools lived inside Google Sheets. The company made them look professional only after they demonstrated value. The operating rule: build the smallest version that tests the idea, put it in front of users and polish it only after it has earned that investment.

That sequence moved AI from a spreadsheet experiment to tools that support client delivery.

How can AI adoption spread beyond a pilot?

The source says companywide adoption comes from employees watching peers build. Fantasy runs regular sessions in which staff create their own tools with Claude and post what they have made in a shared channel. A motion designer and a 3D artist built tools for their own crafts rather than searching for a suitable plugin.

Leadership then has to use the tools. The Fantasy author built Synthetic Humans and Fantasy Canvas, and the whole company now uses them on client work. The source argues that leaders’ daily use removes fear that AI will take people’s jobs while showing teams that it makes them faster and better at work they already love.

The author runs AI agents in the background throughout the day. After one Harvard session, a senior marketing executive said half of what the author had shown felt out of reach for someone without a technical background. The author’s answer was: “Start small, don’t be afraid of the tool, and treat it like a collaborator.”

The author estimates that working this way has made them at least 30% more productive. They also used AI to plan all the logistics for their family’s first ski trip, a trip the family had wanted for years.

The adoption lever is practical use: small builds, peer demonstrations and tools that leaders already rely on.

Why should an AI roadmap use 90 days?

Fantasy rejects an initial three-year roadmap because the technology changes every week. It calls 90 days a better planning horizon. During the first 90 days, teams should pick one workflow, map it with the people who perform the work, score the steps, assign automate, augment or preserve labels, and build the smallest possible version, even if it remains rough.

When the team returns to the map, the source expects the tools to have improved and employees to have built products that no one planned. The leaders it sees extracting the most from AI began with one workflow and kept building.

For Fantasy, the next 90 days will test whether its four-week map can become a repeatable, company-wide operating model built around small releases, improving tools and human ownership of high-value work.