Looking for new invention ideas for the future usually starts with speculation. It does not have to.
Filing activity is public and countable. Every application is classified, and counting classifications over time shows where inventive effort is actually going.
But rising activity is not straightforwardly good news. It signals commercial interest and it signals crowding, and crowding is what makes claims narrow.
The useful signal is usually adjacent — a specific unsolved problem inside a growing area rather than the area itself.
Read the data yourself
| Tool | Shows | Cost |
|---|---|---|
| USPTO Patent Public Search | US filings by CPC and date | Free |
| PatentsView | Bulk US data, counts and trends | Free |
| Espacenet | Worldwide families | Free |
| Google Patents | Full text, translations | Free |
| CPC scheme browser | What each class covers | Free |
Classification groups inventions by what they do, not what they are called, which is exactly what a trend question needs.
Counting is the method. Pick a CPC group, count applications per year across five or ten years, and read the direction.
What the counts tell you
| Pattern | Reading |
|---|---|
| Rising steadily | Sustained commercial interest, increasing crowding |
| Sharp recent rise | New entrants; prior art still thin but closing |
| Flat and high | Mature, well-covered |
| Falling | Interest moving elsewhere — or the problem was solved |
| Low and flat | Quiet — either neglected or uncommercial |
Low and flat is ambiguous and worth investigating. Some quiet classes are quiet because nobody needs the technology; others because nobody has looked.
A sharp recent rise is the most actionable pattern, because the field is commercially live but the prior art has not yet accumulated.
The eighteen-month lag
| Applications publish at | 18 months from earliest priority |
| What you can count | Filed at least 18 months ago |
| What is invisible | Everything filed since |
| Implication | Counts describe the past, not the present |
Trend data is retrospective by construction. A class showing rapid growth describes where money was going eighteen months ago, and considerably more has been filed that you cannot see.
Which means you are never early by reading the data. You are informed, which is different.
Crowding cuts both ways
| Effect of a crowded field | Consequence |
|---|---|
| Dense prior art | Narrower claims |
| Well-funded incumbents filing | Hard to obtain broad coverage |
| More potential infringers | If you get a claim, it matters |
| More rejections, more rounds | Higher prosecution cost |
| Established vocabulary | Easier to search |
A narrow patent in a crowded field costs the same as a broad one. Same filing fees, same prosecution, same $14,470 in maintenance fees for a large entity across its life.
And most of them do not survive. Only 41.4% of US utility patents reach full term — see the patent survival curve.
Where an individual can realistically win
| Look for | Why |
|---|---|
| A specific unsolved problem inside a growing area | Narrow enough that incumbents skipped it |
| Problems you encounter in your own work | Domain knowledge is the real asset |
| Adjacent applications of a maturing technology | Different field, different prior art |
| Manufacturing and servicing problems | Under-filed relative to product features |
| Cost or simplicity, not capability | Incumbents optimise for capability |
Domain knowledge beats trend data. Patentable inventions come from understanding a problem well enough to see why existing solutions fall short, and no classification count supplies that.
Trend data tells you where to look. It does not tell you what to build.
Filing is belief, not outcome
| Ipiry Patent Survival Curve v1.0 | Rate |
|---|---|
| Survive the 3.5-year fee (2022 cohort) | 85.8% |
| Survive the 7.5-year fee (2018 cohort) | 64.6% |
| Reach full term (2014 cohort) | 41.4% |
| Abandoned before full term | 58.6% |
| Lost at the first fee | 14.2% |
| Lost at the third fee | 23.2 points |
Computed from 27,273,654 USPTO maintenance fee records covering 8,262,336 US utility patents.
Nearly three in five patents are abandoned by their owners. Filing counts measure how many people believed something was worth protecting; the survival curve measures how often that belief held up.
Read them together. A class with rising filings and high abandonment is a class where a lot of money is being spent on things that did not work out.
Worked example: reading one class
An inventor with maintenance and servicing experience looks at a growing area.
| Step | Action | Finding |
|---|---|---|
| 1 | Count filings in the CPC group, 2016–2024 | Rising 3× over the period |
| 2 | Identify top assignees | 6 large companies, 71% of filings |
| 3 | Read 30 recent applications | All focused on capability and performance |
| 4 | Look for what nobody filed on | Field servicing and calibration |
| 5 | Count filings on the servicing sub-problem | Very few |
| 6 | Free prior art search on the specific idea | Two close references |
| 7 | Read them properly | Both require a fixed reference standard |
| 8 | Reformulate around self-calibration | Novel over both |
Why this worked
The growing class was not the opportunity. Six well-funded incumbents were already filing heavily on the obvious problems.
The gap was adjacent. Field servicing is a real cost, and incumbents optimising for capability had not addressed it.
Step seven produced the invention. Reading the close prior art showed what had been tried and where it fell short, which is where the reformulation came from.
Steps one to eight cost nothing except time. See invention ideas.
Novel is not enough
| Requirement | Test |
|---|---|
| Novelty (§102) | Not disclosed in a single prior art reference |
| Non-obviousness (§103) | Not obvious to a skilled person |
| Eligibility (§101) | Not an abstract idea, natural law or phenomenon |
| Utility | Has a specific, credible use |
| Enablement (§112) | Can be made and used from the description |
Non-obviousness fails more applications than novelty. Combining known elements in a predictable way is obvious even when nobody has combined them before.
Eligibility matters most for software and service inventions, where a method of organising human activity is an abstract idea regardless of novelty. See what can be patented.
Search before anything else
| Source | Why |
|---|---|
| Google Patents | Full text worldwide, translations |
| USPTO Patent Public Search | Classification browsing |
| Espacenet | Foreign families |
| Product listings and marketplaces | A product on sale is prior art |
| Trade publications | Where examiners search least |
| Academic and conference papers | Frequently missed |
Search by function in five vocabularies. What you call one thing, the prior art may call another entirely.
Market searching matters as much as database searching. No patent database holds a product that was sold but never patented, and it is prior art all the same.
The search is free and it eliminates most inventions, which makes it the highest- value step in the process.
What trend data cannot do
| Cannot | Why |
|---|---|
| Tell you what will be valuable | Filing is belief, not outcome |
| Show current activity | 18-month publication lag |
| Identify unfilled gaps directly | Absence is ambiguous |
| Replace domain knowledge | The invention comes from understanding |
| Predict eligibility outcomes | A legal question, not a market one |
Absence of filings is ambiguous. A quiet class may be neglected or it may be uncommercial, and the data alone cannot distinguish them.
New invention ideas: the checklist
- Count CPC filings over time rather than speculating. The tools are free.
- Treat rising activity as crowding as well as opportunity.
- Remember the eighteen-month lag. You are seeing the past.
- Look for adjacent gaps, not the growing area itself.
- Check what the top assignees are optimising for — and what they are not.
- Start from a problem you actually understand. Domain knowledge is the asset.
- Read filing counts alongside the survival curve. 58.6% are abandoned.
- Test for non-obviousness, not just novelty. It fails more applications.
- Search products as well as patents. Anything on sale is prior art.
- Read the close references properly. The reformulation is usually in them.