Lists of automation framework types usually read like a museum tour: linear, modular, data-driven, keyword-driven, BDD, hybrid. Each gets a definition and a diagram, and you are left with no idea which one to pick. A more useful way to see them is through one question: who writes the test, and in what language? Every framework type is a different answer, and the answer decides who can contribute, how much engineering sits underneath, and where the maintenance cost lands.
Starting point: tests written in code by engineers
The simplest arrangement is a linear script: record or write the steps one after another, with locators and data inline. It is fast to produce and fine for a throwaway check, but every script duplicates everything, so it collapses beyond a few dozen tests.
A modular framework is the first real improvement. Repeated actions such as login, search or checkout become reusable functions or page objects, and tests call them. Tests are still written in a programming language by people comfortable with it, but a change in the login flow is now fixed once.
A data-driven layer then separates inputs from logic: the same test runs over many rows of data from a table, CSV or JSON file. This is less a separate framework than a capability most modern frameworks have built in, through features like parameterised tests in JUnit 5, TestNG data providers or pytest parametrisation.
Up to here, the authors are engineers and the language is code. The next two types change that.
Keyword-driven: tests as tables of actions
In a keyword-driven framework, a test is a sequence of high-level keywords with arguments, written in a table or a simple text format. A bank’s fund transfer test might read as rows like: Open Net Banking; Login As, retail customer; Add Beneficiary, account number, IFSC; Transfer Funds, ₹25,000; Verify Balance Reduced By, ₹25,000.
Each keyword is implemented once by an automation engineer, either in code or by composing lower-level keywords. Robot Framework is the best-known open-source example; several commercial tools use the same idea behind a visual editor.
The promise is that domain experts and manual testers can write tests without programming. That promise holds when two conditions are met: the keyword library is well designed and stable, and someone owns it. It breaks when keywords multiply without discipline. A library with Click Button, Click The Button, Press Button and Tap Submit is a sign that nobody is curating. Another failure is keywords that are too low-level, which turns test tables into programs written in a worse language, with no debugger, no type checking and weak refactoring tools.
BDD: tests as examples of behaviour
Behaviour-driven development, which grew out of Dan North’s work in the mid-2000s, writes tests as structured examples in near-English using the Gherkin syntax: Given a starting context, When an action happens, Then an expected outcome. Tools such as Cucumber, Behave and Reqnroll (the successor to SpecFlow) map each step to code.
On the surface BDD looks like keyword-driven testing with nicer grammar. The difference is intent. BDD is primarily a collaboration practice: product owners, developers and testers agree on concrete examples before code is written, often in a short workshop sometimes called the “three amigos”. The scenarios become a shared specification, and automation is the way that specification stays honest.
A scenario for loan eligibility might say: given an applicant aged 22 with a monthly income of ₹40,000, when they apply for a personal loan of ₹5 lakh, then the application is referred for manual review. A business analyst can read that, challenge it and sign it off. That readability is the whole point.
Where BDD goes wrong is when teams adopt Cucumber without the conversations. Testers then write Gherkin alone, after development, describing UI clicks: “When I click the element with id submit-btn”. This adds a translation layer that nobody outside QA reads, plus a step-definition layer to maintain, and delivers none of the collaboration benefit. If business people will never read or write the scenarios, plain code with good method names is usually cheaper.
Hybrid: what most real frameworks become
Almost every mature framework ends up hybrid. A typical stack has a core layer of driver management, configuration, logging and reporting; a domain layer of page objects, API clients and data builders; and a test layer on top, which might be plain code for most tests, data tables for validation-heavy cases and Gherkin for business-critical flows that stakeholders review.
“Hybrid” is not a type you choose on day one so much as a description of a framework that applies the right authoring style to each kind of test while sharing one engineering foundation underneath.
Comparing the trade-offs
| Type | Test authors | Main benefit | Main risk |
|---|---|---|---|
| Linear | Anyone with a recorder | Very fast to start | Duplication; breaks at scale |
| Modular | Engineers | Reuse, full language power | Non-programmers cannot contribute |
| Data-driven | Engineers, data by anyone | Broad input coverage cheaply | Tests become opaque if data hides intent |
| Keyword-driven | Domain testers | Wider authorship | Keyword sprawl, weak tooling |
| BDD | Whole team, together | Shared, readable specification | Becomes overhead without collaboration |
Choosing for your team
Start with people, not tools. A team of software development engineers in test who code daily gets little from a keyword layer and should write tests in code. A large services team in which many testers come from manual backgrounds may genuinely benefit from keywords, provided two or three engineers own the library. A product team where business analysts actively take part in refinement is where BDD earns its cost.
Also consider what happens in three years. Every abstraction layer adds something to maintain and something new joiners must learn. The question is not which type is most advanced but which one lets the people you actually have write and fix tests quickly, with the fewest layers between a failure and its cause.
Written by the RP Strativa Team. The banking and loan examples are illustrative.
Leave a Reply