Meaning emerges from structure

Take the word “objective”. On its own it is ambiguous: a goal you are working toward, or an impartial perspective. You only know which is meant by how it sits in relation to everything around it.

A brand plan works the same way. When someone on your team reads “increase consideration among high-value cardiologists”, they know it is a brand objective, they know it sits at the top of the strategic hierarchy, and they know that strategies, tactics, key results, budgets, audiences, messages, and claims all hang off it. None of that is written down. It is understood from context.

That is the problem the session sets out to solve. The meaning is there. It is just not architected for AI consumption.

How AI actually reads your documents

Most AI tools pointed at company documents use retrieval-augmented generation (RAG): the document library is chopped into chunks, each chunk is turned into a vector that captures its meaning as a list of numbers, and a question is answered by finding the chunks whose vectors sit closest to the question’s vector. Those become the “relevant” results, and the model writes an answer from them.

The session walks through four limitations of that approach, each with a pharma example:

  1. It struggles to connect the dots. Ask how the first-line positioning imperative is performing and the system finds chunks that mention “first-line” and assembles something plausible. It cannot follow the objective, strategy, content, channel chain the brand plan defines.
  2. It has no sense of time. Launch a new indication and the AI happily serves last year’s objectives alongside this year’s, because it has no concept of a plan cycle.
  3. It does not know you are describing the same thing. “First-line positioning” in one section and “preferred initial therapy” in another are the same imperative. The AI treats them as separate topics and gives different answers depending on which phrase you use.
  4. It fills the gaps with guesswork. Asked for the key result on access and pricing, the system retrieves something close, lacks the actual figure, and infers “formulary listing”. The real goal was “formulary submission”.

Standard RAG gives your AI access to words, not to meaning.

The three missing layers

The second half of the session is practical: what has to be built so that a machine can read a brand plan the way a brand team does.

  • Classification is the bridge between your business language and the system’s understanding of your data. Every company has its own vocabulary for therapeutic areas, strategies, segments, channels, and content, and it has to be written down as a taxonomy: brand objectives, strategies, tactics, key results, and KPIs, each defined and sourced to the section of the plan it comes from. This is where most of the work is.
  • Rules encode the logic that currently lives in people’s heads: which imperatives map to which claims, which channels need which approval path, which KPIs belong to which plan cycle. A global rule that every local imperative must map to an active global one, or a local rule that every key result must reference the current plan cycle, catches the misalignments that otherwise surface only at the next review.
  • Relationships make the connections explicit and traversable. When the link from objective to strategy to tactic to key result to KPI is typed and built into the structure, a question like “which KPIs relate to our first-line positioning objective?” is answered by following the chain rather than by guessing from fragments.

Where Forge DC fits

This is the thinking behind DataBridge. The DataBridge series on this site covers the same ground in writing: Your AI vendor is selling you a hallucination on why retrieval alone fails, and Forget more data, you need a memory on the structured layer that replaces it. Building a Data Foundation Fit for the Human-AI Partnership goes into how structured context turns retrieval into reasoning.

Watch the recording

The recording is hosted on Impatient Health’s YouTube channel and embedded with their permission.

Want to see what your brand plan looks like as structured data? Get in touch.