Meeting speed.

Each follow-up question started a new cycle.

A planner asks who buys in a category. The request goes to a data team, which pulls the data, builds or queries a model, and sends an answer back later. By then the meeting has moved on. The answer raises new questions, such as what else these buyers have in common or what happens when the definition is loosened, and each one starts another cycle. Smaller questions often go unasked because they can't justify a project.

The limit in that loop is the wait between a question and its answer. When the wait drops to seconds, the next question can be asked while the decision is still open.

What a chatbot can't do on its own.

Working alone, a general-purpose chatbot has three gaps.

It has no measured numbers. Ask it for a market size and it produces one with the same confidence it gives everything else. A plan built on an invented 14 percent looks like a plan built on a measured one until the campaign runs.

It can't see the whole catalog. More than 100,000 attributes is more than a model can hold in view at once, so it tends to miss the less common ones, which is where the less obvious audiences are. It also can't tell whether the audience it is drafting is similar to one the client already has.

Nobody checks its work. When one model writes the answer and also decides it is finished, nothing checks the arithmetic or argues the other side.

The consultant does the reasoning, and the Brain does the measuring.

The AI consultant clarifies the brief, searches the attribute catalog, proposes audience definitions, sends them to the Brain, reads what comes back, and writes the plan. It works from written procedures and rules built up over Intelafy's audience work, from the dictionaries that define each attribute, and from the catalog of audiences already built on the client's data. Because it can see that catalog, it checks whether an audience like the one it is about to build already exists. It is instructed to take counts, scores, and variables from runs of the Brain rather than from its own guesses.

The Brain sizes each definition against the file, ranks every other attribute against the one the planner picks and names the evidence behind each rank, builds the profile, and scores each person on the 0-to-1 scale (Paper 02).

Every request, whether it asks for a full marketing plan or a single number, goes through the same four steps before the planner sees a result.

• Plan. The consultant breaks the request into steps that can each be checked, and shows that plan to the planner before any work starts, so a misread brief is caught at the start instead of in the finished plan.
• Run. The steps go out in parallel: runs of the Brain, analysis of the client's files, and research outside the catalog.
• Review. A separate reviewer checks the draft's logic against the measurements: whether the segments hold up, whether the creative direction follows from the motivation scores, and whether anything is claimed that the measurements don't support. Weak work goes back with instructions.
• Fact-check. Each number in the draft is traced to the run that produced it. A number that can't be traced is recomputed or removed.

For an audience, the plan step lays out an eight-stage procedure that starts by listing the brief's constraints and ends with the finished audience and its profile. That procedure is written down, and the consultant looks it up during the work instead of relying on memory, so each engagement follows the same steps. The review step is there to catch problems before the planner does. In one session, a segment was anchored on a behavior that held up poorly when its link to the target was checked. The consultant switched to a stronger anchor and rebuilt the segment before showing any of it.

Follow-up questions stay in the session.

With answers in seconds, a planner can keep asking:

• What else goes with this audience?
• Which motivations separate these prospects from one another?
• Which channels do they respond to, and which do they avoid?
• What changes if we broaden the definition?
• Is there a bigger or better audience next to this one?

In a project loop each of those is a new request. In V3 each is a turn in the same conversation. The sneaker session in Paper 05 ran this way: a seven-word request at the start, and by the end of one afternoon a scored national audience, four motivation segments, channel and creative plans for each, and a five-tier map of the category.

Five kinds of work from the same measurements.

• A marketing plan: the audience definition, motivation segments, creative and offer direction, and channel choices. Because channel scores share one scale, the plan can take out the people who lean away from a channel as well as add the people who lean toward it (Paper 07).
• Category research: a written analysis of a market, drafted section by section, with charts drawn from the same runs the text cites.
• Executive briefs and charts. V3 drafts 10-page executive briefs and illustrated personas, and charts appear in the workspace as the session runs.
• A super audience (Paper 03): one file, built for a named brand's campaign, with each person's scores for channel, offer, motivation, and the rival brands they lean toward.
• Quick answers: a plain-English question about the catalog, answered in seconds with numbers from the Brain, which the same session can build into any of the above.

The whole session exports as a standalone report. It runs on the client's data, in the client's own deployment (Paper 08).

Put the engine in the meeting.

V3 changes when the answers arrive. The consultant plans and checks the work, the Brain's part of each answer takes seconds, and the plan comes together during the meeting. To watch a session run on a category you choose, book a Data Discovery.

Book a Data Discovery →Continue to 07 · Where Not to Spend →