Which Databases to Use for Industrial Technology Watch

In brief

1. Four families of databases answer four different questions. Patents say who is inventing what, regulatory references say what is binding, scientific literature gives the upstream state of the art, curated databases explain how it all applies.
2. Patent databases are free and the earliest signal. Espacenet exceeds 150 million documents, PATENTSCOPE holds 129.5 million, and publication 18 months after filing gives a one to three year lead on the market.
3. The budget tipping point is the standard, not the database. Patents and official regulation cost nothing, standards and indexed literature are paid.

The question of which database to use for industrial technology watch rarely has a single answer, and that is the first trap. A design office tracking a competitor, a quality department following a change in standards and an R&D team looking for the state of the art do not need the same tools. This landscape sorts databases by what they actually provide, with their volume, their access model and what they do not cover.

The landscape in one table

The table reads in three blocks. The first four rows are free and there is no reason to do without them. The standards and literature block is what weighs on the budget. The last row answers a question the others do not address, that of how knowledge is shaped for use.

Family 1: patents, the earliest source

A patent application is published 18 months after its filing date. That is the technical reason this family comes before all others in an industrial watch. The delay means a filing visible today describes work started a year and a half ago, and the resulting product will not reach the market for another one to three years. No other source offers that lead.

Espacenet, published by the European Patent Office, contains data on more than 150 million patent documents from around the world. Its strength is search by Cooperative Patent Classification, the CPC, which queries a technical field rather than a keyword. A keyword misses everything phrased differently, a CPC class misses nothing on the same subject.

PATENTSCOPE, published by the World Intellectual Property Organization, gives access to 129.5 million patent documents, including 5.5 million international applications filed under the PCT. Its distinctive feature is cross-lingual search, which automatically translates a query and searches collections in several languages. It is the tool that keeps an Asian filing from slipping past.

Google Patents and Data INPI round out the set, the first for its interface and machine translation, the second for French titles and the link to company data. Questel Orbit belongs to a different category, portfolio analysis and mapping, and is justified once patent watch becomes a function rather than a task.

The limit of this family is clear. A patent describes an invention, not a use. It says what was protected, never what was industrialised, and a significant share of filings never leads to a product.

Family 2: standards and regulatory references

This family answers a different question, that of what is binding. An industrial company monitoring its obligations is not looking for innovation, it is looking for the gap between what it does and what it should do.

Légifrance and EUR-Lex are free and are the only authoritative sources, one for national law, the other for European law. The distinction between a regulation and a directive is structural here, a European regulation applies directly whereas a directive must first be transposed.

AFNOR, ISO and IEC publish standards, and this is where the budget shifts. A standard is bought per unit, and only its text is authoritative. The question that comes up most often is whether a standard cited in a specification is actually binding. The answer is one figure, fewer than 1 % of AFNOR standards are made mandatory by ministerial order, and Légifrance publishes the list of those. The subject is developed in our landscape of documentation sources for industrial regulatory compliance.

Family 3: scientific and technical literature

Scopus and Web of Science are indexing and citation databases. They do not provide full text, they say what exists, who cites whom and which team works on a subject. This is the upstream scouting tool, before a technology is even patentable.

IEEE Xplore and ScienceDirect provide the text, the first on electronics, computing and energy, the second across a very broad Elsevier perimeter. Both require an institutional subscription and publish in English.

HAL and Google Scholar partially open the same corpus. HAL, the French open archive, hosts deposits from public laboratories and gives access to author versions of otherwise paywalled work. Google Scholar casts a wider net but without editorial control, which means checking the source of every result.

Family 4: curated technical documentation

The three preceding families deliver raw material. A patent is written to protect, not to explain. A standard states a requirement without saying how to meet it. A scientific paper describes a laboratory result without addressing scale-up. Between those sources and an engineer’s decision, a layer is missing.

That layer is occupied by curated documentation databases, of which Techniques de l’Ingénieur is the reference for the French-speaking field. The principle differs from the other three families, the papers are written and validated by domain experts, in French, and cover the whole of engineering, from materials to process engineering, construction and control systems. The question they address is not what has been discovered but how it applies.

This family replaces none of the other three. It provides neither the text of a standard, nor the detail of a filing, nor the raw results of a publication. It supplies the perspective the first three do not produce, and that is why it belongs in a watch setup.

How to combine these sources without spending all your time on it

The habit that costs the most in technology watch is accumulation. An organisation following fifteen feeds does not read them, it archives them. Three to five sources, one per family, handled seriously, produce more.

• One patent source on automatic alert, on one or two CPC classes rather than keywords, and on the applicants being tracked.
• One regulatory source, a Légifrance alert on the relevant headings, with standards monitoring added if the sector requires it.
• One scientific source, an alert on a Scopus query or on the tables of contents of two journals in the field.
• One synthesis source, consulted on demand rather than as a feed, when a subject surfaces and has to be understood quickly.
• One fixed weekly slot, assigned by name. A watch with nobody designated to read it does not exist.

The general method, independent of tools, is developed in our article on technology watch for engineers.