← All classes
CLASS 13July 7 · AI Strategic Foresight

"AI does the data journey.You are the judgment."

Angella Tapé
Taught byAngella TapéSVP Group Strategy Director, Havas

The class in brief

Guest strategist Angella Tapé taught the three-canvas framework her Havas team uses on client work: category first, human truth second, positioning last, with a custom GPT doing the research legwork. The class ran the full framework live on a Busch NA brief. The takeaway skill: directing AI like a junior analyst while the judgment stays yours.

Why this matters

Half of all TV advertising produces no emotional response at all. Strategy decides which half your work lands in.

That claim comes from System1's ad-testing data, which Angella opened with. Her framework is the answer to it: three canvases, worked in order, with AI as the analyst and you as the judge. And always the category before the product: start with the product and, in her words, "I will fail, force-feeding people."

52%
of UK TV ads produce a completely neutral response. "I feel nothing, I do nothing." (US: 47%. B2B is worse.)
2.3x
brand attachment for the most adaptable brands compared to the least. From the Age of DULL research.

The framework

The three canvases

01

Category Creationfrom conformists to creators

The longest stage on purpose. What market am I actually in? What is the paradigm, and what shift is coming? Where are the gaps? The product stays out of the room for the whole pass.

02

Human Truthfrom interruption to meaning

Who are the customers, why do they behave that way, and what single insight sits under it all? Brands that win align to me, my community, or the world.

03

Brand Positioningborrow equity, or build it

Borrow from an adjacent category, or build from scratch the way Liquid Death did. Step away from the head-to-head mindset and the space opens. Then be single-minded: pick one.

The exercise

Working the Busch NA brief live

The brief

Fuel new growth for Busch Light through its non-alcoholic offering.

Light beer is a saturated category, and non-alcoholic drinkers are a growing audience. The ask: use AI deep research to evolve Busch NA's positioning, building on the brand's core equity (Mountains of Busch, the great outdoors, and midwestern values), then identify a growth audience and define new rituals for how Busch NA gets used.

The brief slide appears below, in From the board

Open the class Miro board
How the exercise ran

Five steps, from her workflow slide: upload the data pack, let the agent run deep research, do the thinking and strategizing yourself in Miro, upload your canvases back to the agent, and finish by letting AI challenge the positioning to sharpen it.

Category first, product last

The first canvas bans the product from the room: what market is this, what paradigm, what shift is coming, where are the gaps. Start with the product and you end up force-feeding people.

A market for people who don't like beer

One classmate, who doesn't drink and doesn't like the taste of beer, asked the analyst how big the market is for people who want the ritual without the taste. It came back with real segments, and the class watched a personal observation turn into a researchable audience in minutes.

Pushing past the obvious occasions

Another run kept getting camping, tailgating, and boating back from the model, all summer, all weekend. The fix was a direct order: bring it into the days of the week. The lesson inside it: the model defaults to the most photographed version of an audience until you push.

Another run left the brief entirely

Same canvases, aimed at repositioning a church in Madison County. The framework held anyway. That is how you know it is a framework and not a template.

From the board

What the class made in Miro

The three-canvas framework slide: Category Creation, Human Truth, Brand Positioning
The three-canvas framework, from her deck
The closing recap slide: the three-canvas framework with its supporting visuals
The recap: all three canvases with their visuals
The Busch NA AI Brief slide: objective and ask
The Busch NA AI brief

Methods and prompts

Five methods to take with you

METHOD 01 part of: keep the judgment human part of: reject the first draft

Boss of a junior analyst

Treat the model like a bright junior: direct it, redirect it, and ask for conclusions instead of accepting a data dump.

Where it came fromAngella's working doctrine for the whole framework: AI does the data journey and you are the judgment, so she directs the model the way a strategist directs a junior analyst, thanking it, redirecting it, and demanding synthesis.Use it whenYou are working a research-heavy brief and the model keeps handing you raw material instead of a synthesized answer you can judge.

Working prompt

Synthesize everything above into exactly 2 categories of opportunity. Table format. End with one short question for me. Do not ask anything else.

METHOD 02 part of: reject the first draft

Ask for what's missing

The idea usually lives in what the research does not say. Ask for the missing data, not more of the found data. Her words: "Tell me what does NOT exist. From there you get the idea."

Where it came fromHer gap-hunting move from the live Busch NA exercise: ask the model what does not exist and what data is missing, because the idea usually lives in the gap rather than in the summary.Use it whenThe research summary looks complete but nothing in it sparks a direction worth pursuing.

Working prompt

Based on this research: tell me what does NOT exist in this market, and what data I am missing. List the absences, not the findings.

METHOD 03 part of: structure your ask

Give it an executive's job

Renaming the model's role from assistant to executive changes tone, depth, and brevity everywhere at once. One word doing the work of a page of formatting rules.

Where it came fromAngella showed that renaming the GPT's role from assistant to executive changes tone, depth, and brevity globally, which is cheaper than piling on formatting instructions.Use it whenThe model's answers are wordy or shallow and you want a sharper register without writing ten format rules.

Working prompt

From now on you are an executive strategist, not an assistant. Max 5 sentences per answer, tables where useful, no jargon, answer your own research questions instead of asking me. Confirm and continue.

METHOD 04 part of: structure your ask

Say what you want rather than what you do not

Negatives confuse it. Her full list for reining it in: executive style, five sentences max, table format, no jargon, stop asking me questions, answer your own research questions.

Where it came fromHer hard-won prompting rule from years of client work: negatives confuse the model, so every instruction names the behavior she wants instead of the one she is banning.Use it whenYou catch yourself writing do-not instructions and the model keeps doing the banned thing anyway.

Working prompt

Rewrite your instructions to yourself as positive directives only: what you WILL do, in what format, at what length. Then follow them.

METHOD 05 part of: context beats prompts

Build the tool rather than the chat

A workflow you run twice deserves to be a tool. Know the workflow in your head first, then write it as strict, numbered steps with a named role. Custom GPT or Claude Project, either way the magic is the prompt.

Where it came fromShe opened the hood on her Strategic Foresight GPT: a name, a role, and long, strict, sequential step instructions, locked so students could not edit them. The magic is the prompt, and it is portable to Claude Projects.Use it whenA workflow you run more than twice a month still starts from a blank chat every time.

Working prompt

Help me turn this workflow into a reusable assistant. Here are my steps: [list them]. Rewrite them as strict sequential instructions with a named role, an intake step, staged framework steps, and a fixed output format, always addressing the user.

Deeper detail

Inside her custom GPT

She opened the GPT and showed the class what's inside. Know your workflow in your head first, then write it as strict, numbered steps with a named role, always addressing the user. The same thing works as a Claude Project. Her words: the magic is the prompt, and it moves between tools.

1 · NAME + ROLE
Who the tool is
Stated like a job description.
2 · INTAKE
What it collects first
Everything it asks the user for before any work starts.
3 · STEPS + OUTPUT
The framework, numbered
Strict sequential instructions and a locked output format, so every run ends the same shape.
Open the Strategic Foresight GPT

Related classes

Where this class connects

The shelf

Tools and references

Tools that night

Cases she mentioned

  • BTS: the community that never gives up on you
  • Liquid Death: brand equity built from nothing
  • McDonald's Taiwan: pit stops for cyclists
  • Norway's macho non-alcoholic beer, now #1 with new mothers
← Class 12 · Bias in Our AI Data CLASS 13 OF 24 Class 14 → Copy Craft + AI