Four Ways of Saying “Customer” (5/5): Three Theses and a Caveat
Part 5 of 5. Part 4 compared RDF and LPG.
From this path I draw three theses. The first concerns accountability. The connection between the assistant and company systems is an infrastructure problem, and infrastructure today can be bought.
The meaning of the data instead requires someone who writes down what the company's words mean, indicates which system is authoritative for each, and answers for what they have written.
The second concerns the order of work. Content readiness is assessed before choosing the tool. The usual order, buying first and fixing the content afterward, turns a measurable problem into a risk. The third concerns technique.
The graph model is derived from the questions: whoever writes the competency questions has already made half the choice, whoever starts from the vendor postpones it until after the purchase.
To these three theses I add a caveat, which I owe once more to Eco. In The Search for the Perfect Language (Eco 1993) he reconstructed the long European dream of a language able to mirror the structure of the world without residue, from the language of Adam to the philosophical projects of the seventeenth century.
Among these the most ambitious was that of John Wilkins, who in 1668 proposed arranging every existing thing in a system of genera and differences and giving each a name that revealed its position (Wilkins 1668).
Jorge Luis Borges drew from it a famous essay, in which Wilkins's classification ends up beside an imaginary Chinese encyclopedia that divides animals into categories such as those belonging to the emperor and those that from a distance look like flies (Borges 1952).
A semantic layer project that claims to describe once and for all the meaning of the entire company repeats Wilkins's mistake.
No organization has a single meaning of customer, and every attempt to impose one produces a taxonomy that departments soon learn to get around. The semantic layer works when it remains a local dictionary: written for a declared set of questions, dated, signed by whoever answers for it, revisable.
The closest model is the land register, the record that establishes who owns what, within which boundaries and since when. It is updated at every subdivision and every transfer of ownership, and no one ever considers it finished.
A company's semantic layer should have the same nature: state which word designates which thing, who answers for it, from which system it may legitimately be drawn, and record every change. Frontinus is remembered for having measured the aqueducts and for having written down what flowed through them.
Almost two thousand years later we still read his report, while of the pipes he described mostly ruins remain.
I must, however, state the limits of this work. I took the technical ideas, as I said, from a single firm, which sells what it recommends. Its examples illustrate a principle, and Espie herself presents the Acme case as adapted from a real situation. On the economic weight of these choices I have no evidence.
The three theses also rest on reasons independent of that firm, drawn from archival science, semiotics and Frontinus's method, but they remain to be tested with measurements gathered in Italian companies.
If I had to start tomorrow, I would choose a question the company actually asks, such as the one about open commitments to a customer. I would write the related competency questions, indicating for each who asks it.
I would fix the terms that appear in them, customer, job, deadline, giving each a meaning for every context, an owner and an authoritative system. I would add to the assets involved the business metadata needed to tell them apart.
Finally I would measure, before and after, the number of calls, the reasoning steps and the manual corrections. It is the step most often skipped, and the only one that can turn the three theses into conclusions, or prove them wrong.
I wrote this text with the assistance of Claude (Anthropic).
Sources for this part
- Eco, Umberto (1993). La ricerca della lingua perfetta nella cultura europea. Rome-Bari: Laterza. English edition: The Search for the Perfect Language.
- Wilkins, John (1668). An Essay towards a Real Character, and a Philosophical Language. London: Royal Society.
- Borges, Jorge Luis (1952). «El idioma analítico de John Wilkins». In Otras inquisiciones. Buenos Aires: Sur. English translation: “The Analytical Language of John Wilkins”.