Most companies don’t know which of their patents are the best. And how could they? To find out, you’d first have to know where to start, and then you need outside counsel to build claim charts costing 5 or 6 figures.

If you have <10 patents, you can try to chart these yourself using AI & Google Search. Any more, it starts to get painful. If you have a list of 1,000 or 10,000 patents, you need triage.

At ArcPrime we give counsel three tools that scale from a hammer to a surgical laser:

1. A quality score.
2. Product charts.
3. Patent charts.

Product and patent charts give you conviction about which patents matter. They’re high-signal work product generated by AI—we process millions of tokens per patent to search for real-world evidence that maps to the patent. They answer the question: “Does this claim have a decent likelihood of reading on a product?”

But before you invest at that depth, you want a broadstroke pass that separates the bottom percentile of patents. Our friends at Richardson Oliver Group published a practical framework. We’ve adopted it because it does one valuable thing quickly: it discards the obvious bottom slice so you can focus human attention and AI horsepower where it pays.

ROL’s analysis looked across millions of patents—litigated assets, brokered assets that actually sold, and the “hero” examples brokers highlight. Out of that, five signals show up again and again in the patents that end up mattering. Think of them as a fast filter, not a verdict. The goal isn’t to crown winners; it’s to throw out the near-certain losers.

Here are the 5 factors along with their weight of importance:

• 45% - Number of Forward Citations
• 19% - Age from earliest priority date
• 14% - Independent claim count (adjusted for means-plus-function)
• 12% - Claim 1 word count (penalize extremes)
• 10% - Family size & international filings

Factor 1 — Forward citations (45%)

What it measures. Later patents that cite the subject patent.

Why it matters. It’s a practical proxy for industry R&D momentum in the claimed space. More investment → more products → higher infringement exposure → higher value.

How we score it. Full formula here.

1. Normalize by age: compare a patent’s forward citations to two baselines as a function of years since publication:
   • Issued-patent baseline (broad population), and
   • Litigated-patent baseline (value proxy).
2. Ignore first 3 years after publication (differences are minimal early on).
3. Assign region-based credit once the patent is ≥3 years from publication:
   • Region A: “Massively exceeds” litigated baseline → top credit.
   • Region B: Above litigated baseline (same width as C) → strong credit.
   • Region C: Above issued baseline but below litigated baseline → moderate credit.
   • Region D: Below issued baseline → low credit.

Factor 2 — Age from earliest priority (19%)

What it measures. How far the asset is into its lifetime.

Why it matters. Buyers want adoption (maturity) but also remaining life (enforcement runway). Empirically, patents roughly 8–12 years from priority were most licensing-productive, with at least ~5 years remaining desirable for dispute optionality.

How we score it.

We map age to a curve peaking in the 8–12-year band, tapering down for very young (not yet adopted) and very old (short runway) assets.

Factor 3 — Independent claim count, adjusted (14%)

What it measures. Breadth and investment in prosecution.

Why it matters. Paying for >3 independent claims correlates with litigated and sold patents, but form matters.

How we score it.

• Start with independent claim count.
• Means-plus-function (MPF) adjustment: each MPF independent claim counts as 0.1× a non-MPF claim (under current U.S. case law, MPF is often narrower/less reliable). Exception: if there are ≥5 non-MPF independent claims, no MPF penalty is applied.

We then map the adjusted count to a sub-score (diminishing returns at high counts).

Factor 4 — Claim 1 word count (12%)

What it measures. Over- or under-specificity of the lead claim.

Why it matters. Across value proxies, the distribution of Claim 1 length looks similar to the general population—so length alone isn’t predictive. But extremes are.

How we score it.

• Heavy down-rank if Claim 1 has <25 words or >~250 words.
• Sweet spot (no penalty) roughly 63–163 words, aligned with litigated distributions.

Factor 5 — Family size & international filings (10%)

What it measures. Follow-on investment and geographic breadth.

Why it matters. Bigger, internationally diversified families often reflect strategic importance—but the incremental predictive power is modest compared to citations.

How we score it.

1. Base family score (0–5 pts): scale linearly by INPADOC publication count from 0 to 12 (cap at 12).
2. Multiply that base score by the strongest applicable multiplier:
   • ×25 if there’s a PCT publication <2.75 years from priority,
   • ×25 if <1.75 years from priority (data-sparsity risk adjustment),
   • ×5 if there’s a published EP/JP/CN,
   • ×2 if there’s an issued EP/JP/CN,
   • ×1 otherwise.

Cap the final contribution so the whole factor remains 10% of the total score.

Putting it together — the total score

We compute a weighted sum of normalized sub-scores (each 0–1):

TotalScore =
0.45 * ForwardCitationsScore
+ 0.19 * AgeScore
+ 0.14 * IndClaimAdjustedScore
+ 0.12 * Claim1LengthScore
+ 0.10 * FamilyIntlScore

Important implementation notes

• Transparency: every sub-score is fully explainable and adjustable per client goals (e.g., risk tolerance, target geographies, or runway requirements).
• Field-aware: the forward-citation component is relative to age-matched baselines for issued vs. litigated patents, reducing area-bias.
• Guardrails: weights are tunable within 10–60% per factor; we default to the baseline above.

Why forward citations “win”

In repeated tests across millions of assets, sold patents and representative brokered assets showed dramatically higher forward-citation counts than the broad issued population, and even outpaced litigated sets in some views. We interpret this as the clearest, scalable signal of market-validated relevance.

How do I find the best patents in my portfolio using this?

Now, let’s apply the quality score to every patent to find your best patents no matter how big your portfolio. You'll be able to know your best patents in <1 week.

• Start wide. Use the quality score to eliminate the bottom 50-80%.
• Then go narrow. For the list that survives, generate product charts. This is where you confirm which patents are relevant to which products.
• Finally, be surgical. Generate patent charts for the patents that have the highest number of relevant touch points to products. The point is that by the time you’re deciding, you’re looking at a few sharp needles, not a haystack.
• Optionally, send your favorite patent charts to counsel for a deeper dive.

Credit

This would not have been possible without ROL Group’s work. Many thanks to Erik Oliver, Kent Richardson, and Michael Costa for their work.

License: Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)