The behavioural round: scoring and preparation

A behavioural round asks two or three questions that begin “Tell me about a time when…”, and each answer has a few minutes to prove something about you with one story. Amazon’s preparation pages say each interviewer typically asks two or three such questions, because “past behavior is an indicator of future success”.

Preparation decides these rounds more than talent: a story never said aloud runs long, loses its numbers and turns into “we”. What follows is what gets scored and how, STAR used without sounding rehearsed, and one story told three ways and measured. The full method, story by story, is in the Behavioral Interviews track.

What a behavioural round scores

Three employers publish what their interviewers listen for. Meta’s guide names five focus areas: resolving conflict, growing continuously, embracing ambiguity, driving results and communicating effectively. Amazon assesses its 16 Leadership Principles, with different interviewers assigned different ones. Google’s tips ask how you handle ambiguity, how you work with people and teams, and how you lead, and say its questions want specific examples rather than “I always do X”.

Whatever the list, an answer is judged on what the interviewer can write down from it:

Amazon’s page asks for a detailed account of one situation per question, suggests “I” instead of “we”, and welcomes numbers or specific outcomes. Failure stories count too: Amazon says not every story has to be a success, and Meta’s guide says being open about imperfect projects won’t be counted against you.

Each published list is also a map for your preparation. A story inventory is a set of real stories chosen to cover the list of the employer you are meeting, so that every question of a loop gets a story of its own.

STAR, without sounding scripted

STARSTAR method: A way to structure an answer to a behavioural question in four parts: the situation, your task, the actions you took and the result. Employers publish small variations of it. names four parts: the situation, your task, the actions you took and the result. Employers use it with small differences. Amazon calls it one way to structure an answer but not the only way, Meta’s guide asks for one or two sentences of situation, and Google’s version calls the A the Activity. An answer sounds scripted because of memorised sentences, not because of the structure: Amazon’s page says it does not expect rehearsed scripts. Learn each story as facts and numbers, and say it in fresh words each time. Step through one answer:

Here is that story told three ways, measured part by part at Anaya’s own pace, 150 words read aloud in 66 seconds:

"""One story told three ways, measured part by part at the speaker's own pace."""

# Anaya read a 150-word passage aloud, at her interview pace, in 66 seconds.
RATE = 150 / 66 * 60  # words a minute
STOCK = ("team player", "hard worker", "passionate", "above and beyond", "ownership")

RAMBLING = {
    "situation": "So this was at my last company, which is a food delivery startup, we had "
    "about forty engineers then, maybe more, and our team owned the order service and "
    "also notifications, and there was a lot going on that quarter because of a big "
    "sale, and the product team had also changed the app, and honestly the on-call "
    "rota was a bit of a mess, and our manager had just left, so it was a hard time. "
    "On the sale evening things started going wrong with notifications and people got "
    "messages late, and support was getting a lot of tickets, and we were all on a call.",
    "task": "",
    "action": "We looked at the logs and the dashboards and we tried a few things, we "
    "restarted some workers and we talked to the vendor, and eventually we fixed it.",
    "result": "It was fine after that and everyone was happy.",
}
SCRIPTED = {
    "situation": "In my previous role I faced a challenging situation that tested my ownership.",
    "task": "My task was to go above and beyond to deliver results for the customer.",
    "action": "As a passionate team player and a hard worker, I took ownership, I dove "
    "deep, I invented and simplified, and I disagreed and committed with my team.",
    "result": "As a result the customer was delighted, and I learned a great deal.",
}
STAR = {
    "situation": "On a sale evening at a food delivery startup, order-status messages "
    "were reaching customers up to 14 minutes late.",
    "task": "I was on call and owned the notification service, so finding the cause was mine.",
    "action": "I compared the queue's depth with the provider's error rate and saw that "
    "every failed send was retried at once, so failures multiplied. I added a "
    "backoff to the retries and a limit on the queue, tested both on a copy of the "
    "evening's traffic, and shipped them behind a switch within the hour. Then I "
    "wrote a load test for the retry path so the next sale would be covered.",
    "result": "Delays fell from 14 minutes to under one minute at the peak, and the next "
    "two sales ran without an incident. I learned to load-test failure paths, not "
    "only the happy path.",
}


def words(text: str) -> int:
    return len(text.split())


def report(name: str, story: dict[str, str]) -> None:
    total = sum(words(part) for part in story.values())
    text = " ".join(story.values()).lower()
    print(
        f"{name}: {total} words, {total / RATE * 60:.0f} s at {RATE:.0f} words a minute"
    )
    for part, said in story.items():
        share = words(said) / total
        print(
            f"  {part:<10}{words(said):>4} words {share:>4.0%}{'  MISSING' if not said else ''}"
        )
    numbers = sum(ch.isdigit() for ch in story["result"]) > 0
    stock = sum(text.count(phrase) for phrase in STOCK)
    print(
        f"  result has a number: {'yes' if numbers else 'no'};  stock phrases: {stock}"
    )


for name, story in (("Rambling", RAMBLING), ("Scripted", SCRIPTED), ("STAR", STAR)):
    report(name, story)

Rambling: 145 words, 64 s at 136 words a minute
  situation  107 words  74%
  task         0 words   0%  MISSING
  action      29 words  20%
  result       9 words   6%
  result has a number: no;  stock phrases: 0
Scripted: 67 words, 29 s at 136 words a minute
  situation   13 words  19%
  task        14 words  21%
  action      27 words  40%
  result      13 words  19%
  result has a number: no;  stock phrases: 6
STAR: 137 words, 60 s at 136 words a minute
  situation   19 words  14%
  task        15 words  11%
  action      71 words  52%
  result      32 words  23%
  result has a number: yes;  stock phrases: 0

The rambling version spends 74 % of its words on the situation and never says what was hers. The scripted one is short but gives nothing to check: no number, and six stock phrases, two of them principles recited by name. The STAR version gives half its words to actions in the first person and ends with a number and a lesson, in about a minute, which leaves room for questions.

Write the checker those measurements come from, then run it on two stories of your own:

"""Check a behavioural story written in four labelled parts."""

PARTS = ("situation", "task", "action", "result")


def check_story(story, rate):
    """Return the problems of a story, in the order the prompt lists them.

    story: a dict from part names to text; rate: speaking words a minute.

    >>> check_story({"situation": "s", "task": "t", "action": "a b c", "result": "2x"}, 136)
    []
    """
    # Write your solution here.
    return []

from star_check import check_story

GOOD = {
    "situation": "On a sale evening messages were 14 minutes late.",
    "task": "I owned the notification service.",
    "action": "I compared the queue with the error rate and added a backoff and a limit.",
    "result": "Delays fell from 14 minutes to under 1.",
}


def test_a_complete_story_has_no_problems():
    """a story with every part, a long action and a number passes"""
    assert check_story(GOOD, 136) == []


def test_missing_parts_are_named_in_order():
    """each absent or empty part is reported, in the order situation, task, action, result"""
    story = {"situation": "", "action": "I did a lot of careful work here.", "result": "It took 2 days."}
    assert check_story(story, 136) == ["missing: situation", "missing: task"]


def test_action_must_be_the_longest_part():
    """a situation as long as the action is a problem"""
    story = dict(GOOD, situation=" ".join(["word"] * 15))
    assert check_story(story, 136) == ["action is not the longest part"]


def test_result_needs_a_number_or_a_lesson():
    """a result with neither a digit nor learned is a problem"""
    assert check_story(dict(GOOD, result="It went well."), 136) == ["result has no number or lesson"]


def test_a_lesson_is_enough():
    """a result that says what was learned passes without a number"""
    assert check_story(dict(GOOD, result="I learned to test the retry path."), 136) == []


def test_three_minutes_or_more_is_too_long():
    """408 words at 136 words a minute take exactly 180 seconds"""
    story = dict(GOOD, action=" ".join(["word"] * 386))
    assert check_story(story, 136) == ["too long: 180 s"]


def test_just_under_three_minutes_is_fine():
    """407 words at 136 words a minute take 179 seconds"""
    story = dict(GOOD, action=" ".join(["word"] * 385))
    assert check_story(story, 136) == []

A story inventory of eight to ten stories, mapped to published values Pro: opens with a Pro, Premium or Ultimate pass

Delivering an answer: length, follow-up questions and failure stories Pro: opens with a Pro, Premium or Ultimate pass

Practise: what is scored, and STAR

Check yourself on the scoring and on STAR, starting with the idea you have answered least well:

Interview questions

Key takeaways

• Behavioural rounds score evidence from one past situation: what happened, what you did, the result and the lesson.
• STAR structures a story; memorised sentences, stock phrases and recited values make it sound scripted.
• One specific story, in the first person, with a number or a lesson, fits in about a minute and leaves room for questions.