A semantic layer sits between raw tables and the people using them. It records how tables relate (orders belong to customers, invoices to branches), what each column means in plain words, how key metrics are calculated (“gross margin is revenue minus cost of goods, excluding VAT”), which words mean the same thing (“turnover” is revenue, “debtors” are receivables) and which tables hold the official figures.
It matters because without it every report re-decides the definitions. One dashboard counts refunds in revenue, another does not; the AI assistant picks a third way. A semantic layer makes the definition a shared asset: change it once and every chart, scheduled report and answer follows.
A worked example: a group with an ERP, a CRM and a POS defines “net sales” once, relates the three by customer and store codes, and adds the synonyms its managers actually use. Now the sales dashboard, the monthly board pack and a manager asking “what were net sales in Doha last week?” all return the same number. Rule of thumb: define the ten metrics people argue about first; the rest can wait.
In Klayara, the semantic layer holds relationships, saved metrics, synonyms, plain-language descriptions (which Klayara AI can draft for a person to approve) and certified tables, and Klayara AI uses it to answer questions the way finance would. See the semantic layer.