Author: Andreas Triantafyllidis
Public administrations are increasingly automating how submitted evidence is processed. But the real question is not only which technology to use — it is how refined the underlying information needs to be, what functions that enables, and how reusable that information can become over time. In Pilot 7, the Document to Knowledge framework offers a way to make those choices explicit before technology or vendors are selected.
Public administrations automate the processing of submitted evidence for clear, immediate reasons: to work faster, to reduce costs and to make fewer errors. But these benefits depend on how refined the information needs to be. For example, sending a document to the right department needs much less refined information than checking whether an applicant meets the eligibility rules. Each project also determines, often without anyone deciding it explicitly, how widely the information can be reused later: by other services, in future reforms and, eventually, in once-only exchanges between administrations.
Both questions are about the information, not about the technology. They should be answered before any technology is chosen, because they define what the project must achieve: which functions it must support, which processes it must serve, and what it must keep possible for the future. Technologies change quickly, and new ones appear every year. The objectives tend to last longer.
Six levels of information maturity
In Pilot 7¹ we are developing a framework for this called Document to Knowledge (D2K)², a way to describe where the information of a service is today, and where it needs to be to meet its objectives. It has two parts: a set of levels that describe how refined information is, and a map of what each level makes possible, that is, which functions, use cases and processes it can support.
Part 1 of the D2K describes six levels of information maturity:
- Scanned image. A picture of a paper document. The computer can see it, but cannot read the words.
- Unstructured text. Software has turned the picture into text, but the structure of the document is lost.
- Structured text. The text is organised into parts such as headings, paragraphs and lists.
- Labelled information. Parts of the text are marked by what they mean — such as a name, a date or an amount — not only by where they appear.
- Modelled information. The important information is described as relationships between things — for example, this person lives at this address.
- Connected knowledge. That information is linked to trusted external sources, such as official registers.
What each level makes possible
The second part of D2K maps what each level makes possible: the functions a service can rely on, such as sorting documents by type, searching them, or checking them against rules. These functions do not switch on at one level. Most start to work at a lower level and become reliable at a higher one. For example, a computer can start sorting documents by type from a scanned image, but it does this reliably only once the text is structured.
Three questions for every service
D2K can be used to ask three questions about a service: where is its information today, where does it need to be for the current objectives, and where could it go later? The first two questions are about making the service work well now. The third is about the longer path. Higher levels make more automation possible and, over time, produce information that other services and administrations can reuse. This can bring parts of once-only closer.
Designing for reuse, not only today’s needs
The answers also help avoid a trap: a project that meets today's objectives but leaves the information no more reusable than before.
Discussing and answering these three questions makes the scope of a project clearer. It also helps define the requirements, and how they will be evaluated, before any technology or vendor is chosen.
_____
1 ― "Pilot 7 - AI4Evidence” is part of the 2nd period of the GovTech4All consortium program.
2 ― The "D2K framework" will be validated and tuned from feedback by relevant stakeholders.