Knowledge Graphs - Your starting point for use of AI in Life Science Marketing

The Problem with How We Work Today

There’s one frustrating truth about life sciences: We’re comfortable being slow — just like the drug discovery process.

Marketing and product management are no different. Nothing much has changed in how we operate compared to 2010, except we now have dashboards and software that let us diagnose more, but not necessarily do better.

Campaign planning is still the same: ten people in a room, brainstorming what the world needs, creating campaign tactics, and calculating impressions and click-through rates.

The AI Hype That Didn’t Deliver

When Large Language Models (LLMs), ChatGPT, and AI entered the scene, we all expected a revolution in marketing. But so far, the results have been… underwhelming. Sure, some people are more “productive” and can crank out content faster. But most marketers use these tools no better than a curious twelve-year-old experimenting after school.

Why? Because those leading AI initiatives in our companies rarely speak to life science marketers in our language.

And so, we marketers keep doing what we’ve always done — making the next product selection guide or battle card. Meanwhile, customers suffer. They want us to focus on improving their buying experience, not spend our time building AI-enabled ad campaigns with no real benefit for them.

I’m not here to vent. I’m here to say we can change — but to do that, we must change our toolbox. Today, I want to start with one of my favorites: Knowledge Graphs. And we’ll explore them through a simple but universal problem: Product Selection.

The State of Product Selection Today

Product selection guides are among the most valuable marketing tools for scientists. Every product manager has their own portfolio — and wants to improve the customer experience without relying on static content.

The result? We end up with dozens of disconnected guides:

• One for filters
• One for solvents
• One for chromatography columns
• One for buffers
• One for antibodies
• One for catalysts

…and the list goes on.

How are these built? Someone spends days (often months) creating static content — PDFs, web pages, or sales aids. Customers receive them in one form or another, but they never adapt to new data, regulations, or customer context.

Imagine if we could merge all these guides into a single tool and deliver it to customers through an AI agent. This would be a powerful opportunity for marketers to truly transform the buying experience.

Three Solutions to the Selection Guide Problem

1. LLM-powered AI agent using PDFs of selection guides as its “database” The easiest upgrade: feed your static guides into an AI interface. Impact: Low — you’re still relying on outdated, static content that must be manually refreshed. The AI can only be as accurate as the documents you feed it.

2. LLM agents leveraging global web data Sounds powerful, but science demands precision, not storytelling. When product selection impacts research outcomes, hallucinations — confident but wrong answers — are unacceptable. The open web can’t replace your validated product data, regulatory rules, and application evidence.

3. Knowledge Graphs as the structured source, combined with LLM agents for the interface A knowledge graph organizes your product, application, and regulatory data into a connected, precise information network. Pair it with an LLM, and you get the best of both worlds — accurate, evidence-backed recommendations, delivered through an easy, conversational interface for customers.

It may sound like the obvious choice, and for many use cases it is. But here’s the reality: most marketers and product managers don’t yet know what a knowledge graph really is, or how to start building one..

What is a Knowledge Graph?

A knowledge graph stores and connects information so that:

• Things (products, people, processes, concepts) are represented as nodes (dots).
• Connections between those things are edges (lines).
• Both the things and the connections have meaning a computer can understand.

It’s not just “data in a table” — it’s data + relationships + meaning.

Example: Cell Therapy Workflow

Nodes (things):

• Products: Cryopreservation Media, Flow Cytometers, Cell Culture Flasks
• Processes: Expansion, Differentiation, Cryostorage
• Entities: Researchers, Institutes, Regulatory Bodies (FDA, EMA)
• Data: White papers, Application notes, GMP guidelines

Edges (connections):

• “Used for” — Cryopreservation Media → Cell Banking
• “Complies with” — Cell Culture Flasks → ISO 10993
• “Referenced in” — Application Note → Specific Clinical Trial

If a researcher looks at the “Expansion” step, the system could automatically show them relevant products, publications, and compliance information in one connected view.

Example: HPLC

We know:

• Product: HPLC Column
• Application: Protein Purity Analysis
• Regulation: USP 621
• Customer: Dr. Smith

Connections:

• HPLC Column → used for → Protein Purity Analysis
• Protein Purity Analysis → regulated by → USP 621
• Dr. Smith → purchased → HPLC Column

If someone asks: “Show me regulations relevant to products Dr. Smith uses”, the system follows the links and finds USP 621 — even though no single table contains that exact connection.

HPLC Example: Then vs Now

Old Way: Scientist scrolls through a PDF or filters a table, hoping they match the right pore size, phase, and dimensions.

Knowledge Graph Way: Scientist says: “I’m doing peptide purity analysis in plasma on a UHPLC under USP <621> guidelines.” The tool instantly finds:

• Validated columns for that scenario
• Regulatory references
• Linked application notes
• Instrument compatibility data …and explains why those columns are recommended.

The system doesn’t just list SKUs — it explains why each option is recommended and links to relevant application notes, regulatory standards, and complementary products.

How This Differs from a Regular Database

How to Get Started — Without a Data Science Degree

Most life science companies have talented AI and database teams — but they don’t know your marketing problems unless you tell them. You need to be the boundary spanner who translates marketing use cases into technical requirements.

Steps to start:

1. Map your world (Nodes): Products, applications, workflows, instruments, regulations, evidence.
2. Define relationships (Edges): e.g., “ColumnX → recommended_for → peptide mapping.”
3. Standardize language: Controlled vocabularies for analytes, matrices, techniques.
4. Harvest evidence: From application notes, publications, webinars, support cases.
5. Collaborate with technical teams: Application scientists and tech support to validate connections.
6. Start small: Pilot with one workflow, then expand.

Suggested First Sprint (4–6 Weeks)

• Scope: e.g., peptide mapping & small-molecule impurity profiling.
• Inventory: 20 top SKUs, 30–50 application notes/publications, 10 frequent instruments.
• Ontology v0.1: Define entities and relationships.
• Data load: Normalize 20 SKUs + extract 100–200 facts from notes.
• Prototype: Simple UI — input analyte, matrix, instrument → ranked columns + “why” + evidence.
• Measure: First recommendation success rate, time-to-decision, click-through to evidence.

The Challenge to Marketers

If we keep relying on static selection guides, we’ll keep delivering 2010-style experiences in a 2025 market.

If we rely solely on LLMs, we risk giving fast but wrong answers in a regulated environment. The future is Knowledge Graph + LLM — accurate, connected, explainable, and customer-first.

The only question is: Will you lead the change, or watch your competitors do it first? Unfortunately there is no third option for you.