Best Practice
A best practice is an approach that has repeatedly proven itself in practice and is therefore considered a recommended standard. It is not a rule, but an experience-based recommendation that can change as new insights emerge.
When many people solve the same task, very different paths to the solution emerge over time. Some of these regularly lead to good results, others often go wrong. The path that has proven itself over many attempts is called a best practice. The English term literally means “best procedure” and refers to a recommendation, not a law. Those who follow it don’t have to make every mistake themselves all over again. Those who deliberately deviate from it should be able to give a good reason for doing so.
Why other people’s experience saves time and money
In technology, mistakes are often costly. A poorly secured password system, unclearly documented software, an AI model without testing before launch: such problems cost companies millions and users their data. Best practices are an attempt to bundle this knowledge. Instead of every team having to learn the same lesson anew, it is written down once and passed on.
A second reason is comparability. When all developers in a company write their code according to the same rules, everyone can read each other’s work. New employees find their way around more quickly. It also helps during external audits to be able to refer to recognized procedures.
It is important to distinguish this from a standard and from a law. A standard is formally established, for example by a standardization organization. A law is binding, and a violation has legal consequences. A best practice is neither of these: it only carries as much weight as the professional community grants it. Nevertheless, it is sometimes used in court as a benchmark when examining whether someone acted with due diligence.
How experience turns into a recommendation
It usually starts with a concrete problem and a solution that works surprisingly well. Others try it out and report on it. Gradually, experience reports, studies, and measurements accumulate. If the procedure proves successful in many different situations, it solidifies into a recommendation.
This is recorded in manuals, guidelines, or internal rulebooks. Large providers publish such collections for their products. Authorities do the same, for example for protecting computer systems from attacks. Often, not just the rule but also the reasoning behind it is included.
A common misconception is that a best practice is right forever. The opposite is true: it applies to a certain state of the art. It used to be considered exemplary to have users change their password every three months. Today, experts advise against this because people then tend to choose easily guessable variants. So anyone who continues an old recommendation without checking it does not automatically do everything right.
Best practices in dealing with AI systems
In the AI industry, the term is currently used particularly often. It concerns questions such as: What data may a model use for learning? How does one test before release whether it gives dangerous answers? How does one label images that were generated by a machine? For all of this, there are not yet firm laws in every country, but growing collections of recommendations.
Rules of thumb have also become established when it comes to formulating instructions for a chatbot. Clear task description, an example of the desired result, specification of the desired length: such tips circulate as best practices for dealing with language models.
In corporate announcements and stock market news, the expression often appears when something has gone wrong. After a data leak, it is then said that the company will in future adhere to “industry-standard best practices.” Such statements are easy to make and hard to verify. It is worth asking exactly what concrete measures are actually meant.