Grade the Agent’s Answer

Marlowe Athletic has been piloting an internal analytics assistant called Ask Marlowe. Club general managers type a question in plain English; the assistant writes SQL, runs it, and returns a table plus a recommendation. GMs are not analysts.

Eight weeks ago the GM of Club 41 asked: “Which of my class formats bring in members who stay?”

Here is the schema Ask Marlowe has access to.

members(member_id, club_id, tier, joined_at, cancelled_at)
class_bookings(booking_id, member_id, class_id, booked_at, attended)
classes(class_id, club_id, format, start_at, capacity)

Here is the query it wrote.

SELECT c.format, COUNT(DISTINCT b.member_id) AS bookers, ROUND(100.0 * COUNT(DISTINCT CASE WHEN m.cancelled_at IS NULL THEN b.member_id END) / COUNT(DISTINCT b.member_id), 1) AS pct_still_active FROM class_bookings b JOIN classes c ON b.class_id = c.class_id JOIN members m ON m.member_id = b.member_id WHERE [c.club](https://c.club)

_id = 41 AND b.booked_at >= '2025-09-01' GROUP BY c.format ORDER BY pct_still_active DESC;

Here is what it returned.

And here is the recommendation it gave the GM:

Format	Bookers	pct_still_active
Aqua Fit	412	88.1
Cycle 45	1,933	81.4
Yoga Flow	1,610	79.8
Strength Lab	1,204	76.2
Anvil	688	58.3
Sunrise HIIT	501	55.9

“Aqua Fit and Cycle 45 drive the strongest member retention at your club. Anvil and Sunrise HIIT show the weakest retention among their attendees and are candidates for removal from the schedule.”

The GM cut Anvil and Sunrise HIIT from the Club 41 schedule six weeks ago. The regional director has now asked the data team to check the work.

You pull one more thing before you start — the member mix behind each format.

Format	Bookers	Median months since joining, at time of booking	% of bookings made in a member’s first 90 days
Aqua Fit	412	51	6%
Cycle 45	1,933	34	11%
Yoga Flow	1,610	29	14%
Strength Lab	1,204	22	19%
Anvil	688	5	61%
Sunrise HIIT	501	4	66%

And one club-wide baseline, which the data team already publishes: at Club 41, members in their first 90 days have a 12-month survival rate of 57%. Members past 24 months have a 12-month survival rate of 87%.

Write the regional director a memo.

Constraints to consider

• You have read-only access to the three tables shown. No new tables, no new columns, no additional event tracking, no survey.
• You cannot answer “we need more data” and stop. The regional director needs a position on the two cut formats, using what is in front of you.
• The two formats were cut six weeks ago. Whatever you recommend has to account for six weeks of members who have already lost them.
• Ask Marlowe stays on. It will keep answering GM questions across all 84 clubs regardless of what you find. Any fix you propose has to be a change to how the assistant behaves, not a request to switch it off.
• Thirty minutes. One corrected number is worth more to the regional director than five directional statements.

AI Usage Guidance

We expect you to use AI tools. We evaluate how you use them — not whether you use them. Evidence of iteration, redirection, and critical evaluation scores higher than a polished output with no process documentation.

The single highest-signal indicator: your video answer to the mandatory AI question. If you cannot name a specific moment where you redirected AI output, evaluators will assume you did not.

Mandatory AI question for your video: Walk me through one moment where you disagreed with, pushed back on, or redirected what the AI gave you — and what you did instead. Name the specific moment. Explain what the AI produced that didn’t meet the bar, what you did differently, and why.

Speak naturally. Communication is assessed on clarity of technical ideas and logical structure — not verbal polish, accent, or filler words.