What's the first thing to fix?

Replace the bibliometric output (claim counts, classification heatmaps) with a decision-anchored output: a numbered list of three to five filings that change what an internal stakeholder should do this quarter. The visualizations stay as appendix.

Why patent landscape work is broken, and what to fix first

What's actually broken

A typical patent landscape report contains: classification heatmaps, filing-velocity charts, top-applicant rankings, citation networks, and a "white space" map. The output is impressive. The output is also nearly useless for the IP team's actual work.

Why? Because the deliverable answers the questions the database can answer ("which CPC class grew?", "who filed?") rather than the questions the IP team asked ("are we clear to ship?", "should we accelerate the X program?", "what's the freedom-to-operate position on Y?").

The decision-anchored model

In our consulting practice, we've found three reframes that move landscape work from theater to leverage:

Define the decision before the search. Not "what's the landscape look like" but "by end of Q3 we need to decide whether to file in claim space X, license from Y, or pivot to Z." The search scope is reverse-engineered from the decision boundary.
Report against the decision, not the database. The headline output is a numbered list of three to five filings (sometimes one) that change what the stakeholder should do this quarter. The classification heatmap is appendix material.
Time-bracket the FTO question. Freedom-to-operate isn't a binary; it's a position with a confidence window. A landscape report should state the window explicitly: "FTO position assessed as clear through end of 2027 against claim set X, with recheck recommended on amendment publication for application Y."

What AI actually changes here

Not what most teams assume. AI doesn't fix bad landscape work, it scales it. A team running the bibliometric default with AI assistance just produces more visualization theater, faster.

The decision-anchored model is what AI accelerates productively: AI can read claim language, summarize amendment history, and surface candidate filings against a stated FTO question. It cannot decide what FTO question to ask. That's where the IP analyst's judgment belongs.

What the standards actually say

This isn't a fringe opinion. WIPO's own Guidelines for Preparing Patent Landscape Reports start from the same place: a landscape "seeks to answer specific policy or practical questions" for a defined audience, not to plot everything a database can plot (WIPO, 2015). The guidelines bake in the "need behind the need" logic (the need drives the question, the question drives the data, the data drives the tool) and treat precision/recall as a deliberate trade-off, not an afterthought.

Freedom-to-operate is its own discipline with a defined process. WIPO's FTO guide teaches a three-stage method (Describe the invention, Search for in-force claims that might read on it, Analyze each claim and its legal status) and explicitly documents the limits and risks of the determination (WIPO FTO guide, CDIP). That is why I insist on stating an FTO position as a confidence window with a recheck trigger, rather than a binary "clear / not clear."

Sources

WIPO (2015). Guidelines for Preparing Patent Landscape Reports. overview · full PDF
WIPO. Guide on conducting a Freedom-to-Operate (FTO) analysis (Describe / Search / Analyze). PDF
Trippe, A. Patinformatics: Patent analytics, types of patent reports & the Law of Linear Patent Analysis. WIPO presentation, PDF