The advent of bidirectional programming, in recent years, has led to the development of a vast number of approaches from various computer science disciplines. These are often based on domain-specific languages in which a program can be read both as a forward and a backward transformation that satisfy some desirable consistency properties. Despite the high demand and recognized potential of intrinsically bidirectional languages, they have still not matured to the point of mainstream adoption. This dissertation contemplates some usually disregarded features of bidirectional transformation languages that are vital for deployment at a larger scale. The first concerns efficiency. Most of these languages provide a rich set of primitive combinators that can be composed to build more sophisticated transformations. Although convenient, such compositional languages are plagued by inefficiency and their optimization is mandatory for a serious application. The second relates to configurability. As update translation is inherently ambiguous, users shall be allowed to control the choice of a suitable strategy. The third regards genericity. Writing a bidirectional transformation typically implies describing the concrete steps that convert values in a source schema to values a target schema, making it impractical to express very complex transformations, and practical tools shall support concise and generic coding patterns. We first define a point-free language of bidirectional transformations (called lenses), characterized by a powerful set of algebraic laws. Then, we tailor it to consider additional parameters that describe updates, and use them to refine the behavior of intricate lenses between arbitrary data structures. On top, we propose the Multifocal framework for the evolution of XML schemas. A Multifocal program describes a generic schema-level transformation, and has a value-level semantics defined using the point-free lens language. Its optimization employs the novel algebraic lens calculus.