From forty messages to one place

A generic AI guesses what your group might like; Common Ground knows what each of you loves and finds the plan where those tastes overlap.

The story follows the recorded sample “Flatmates taking Amma out for her birthday”. People are fictional; scores are real Qloo data from a real recorded run. Nothing here calls Qloo or a model when you load it.

Chapter 1

The group chat that never ends

Saturday, 4:10 pm. Forty messages later nobody has a plan, and the person who always organises is guessing what everyone likes.

Kabir: where are we going tonight?
Tanvi: anything is fine
Sameer: wings?? that place with the craft beers
Tanvi: not there. too loud
Kabir: ok somewhere else then
Tanvi: anything is fine!!
Sameer: …
Kabir: someone just pick
…32 more messages. An invented scene, not data.

Chapter 2

Four people, four sets of tastes

Kabir, Tanvi, Sameer and Amma (fictional) each name a few things they love. Amma is vegetarian and keeps to a budget. Their words go to Qloo exactly as typed.

Kabir
Arctic Monkeys · Peaky Blinders · craft beer
Tanvi
The Weeknd · Breaking Bad · cocktails
Sameer
Linkin Park · Mirzapur · chicken wings
Amma
M. S. Subbulakshmi · Shankarabharanam · filter coffee
vegetarian · up to ₹₹₹

Chapter 3

The overlap

Each friend's tastes score every candidate place. Because one person's raw scores sit close together, they become percentiles (100 = that friend's favourite in the pool). The pick is the place whose least-happy friend is happiest.

	Kabir	Tanvi	Sameer	Amma	Least happy	Average
Averaging picks The 13th Floor	74	52	100	87	52	78
Fairness picks SkyDeck by Sherlock's	87	57	61	83	57	72

Averaging chases the happiest friend; fairness lifts the one at the bottom. The gap is modest, and we report it as measured. The model never produces a number; all of these are Qloo scores run through plain code.

Chapter 4

Without Qloo, and with it

Without Qloo vs with Qloo

Same model (claude-sonnet-5-5), same group, same writing rules. The only difference: no Qloo tools. Generated once on 2026-10-22 and stored.

Without Qloo

Windmills Craftworks, Whitefield

Qloo has it as Windmills Craftworks [Whitefield] · ₹₹₹₹ · 6th Floor, Craftworks 331, 5B Rd, Shivaji Nagar, EPIP Zone, Whitefield Bengaluru, Karnataka 560066 India

Kabir: Kabir, they brew their own craft beer right on site, so you'll be sorted while Amma enjoys her dinner.
Tanvi: Tanvi, the cocktail menu here is solid, and the live jazz vibe is perfect for a birthday night out.
Sameer: Sameer, there are good non-veg starters on the menu for you, so no one has to compromise on their plate.
Amma: Amma, there's a proper vegetarian menu, the music is soft and melodic, and you can finish with a hot cup of coffee, so it's a relaxed birthday treat.

Breaks Amma's limit: price level 4 is above 3

These reasons aren't checked against any data.

With Qloo (the plan above)

SkyDeck by Sherlock's

Kabir: Kabir, Qloo ties this rooftop to both your Arctic Monkeys and Peaky Blinders sides, so the vibe should suit you.
Tanvi: Tanvi, they serve cocktails and the place links to your The Weeknd and Breaking Bad taste, so you get your drink and your mood.
Sameer: Sameer, the lively rooftop pub matches your Linkin Park and Mirzapur taste, and there's pub grub to go with it.
Amma: Amma, Qloo links this place to your M. S. Subbulakshmi and Sankarabharanam taste, and the menu has veg picks like paneer tikka.

Every reason cites Qloo evidence; see the X-ray.

How well each pick matches each friend's tastes
Friend	Without Qloo	With Qloo
Kabir	8	88
Tanvi	0	58
Sameer	58	63
Amma	4	83
Least happy friend	0	58

How this is measured, and what it doesn't prove

Both places are scored the same way: Qloo's affinity for each friend, from that friend's own tastes, turned into a percentile among the 24 candidate places plus the no-Qloo pick (100 = their best match there). Raw affinities: Kabir 0.700 vs 0.727; Tanvi 0.662 vs 0.684; Sameer 0.873 vs 0.875; Amma 0.801 vs 0.827.

This uses Qloo's own taste model as the yardstick, and the candidate places are Qloo's recommendations, so it shows how far a generic guess is from each friend's taste profile. It isn't an independent test of whether real people would enjoy the evening more. The parts that need no Qloo score: whether a pick breaks someone's limits, and whether a reason is backed by data.

Chapter 5

The number

The same model, the same group, with and without Qloo. Each pick is scored for each friend with Qloo; the figure is the least-happy friend's percentile (100 = their best match in the pool).

Group	Without Qloo	With Qloo
College friends, Bengaluru, Saturday night	Meghana Foods: 0	The 13th Floor: 71
Couple with opposite tastes, Friday date night	Matsuri: 0	Don't Tell Mama: 81
Flatmates taking Amma out for her birthday	Windmills Craftworks: 0	SkyDeck by Sherlock's: 58

Read this with the caveat: the yardstick is Qloo's own taste model and the pool is Qloo's recommendations, so a pick from outside it tends to rank low. It measures distance from each friend's taste profile, not whether real people would enjoy the night more. A blind check with real groups hasn't been done yet.

Open the full plan and X-ray Plan for your own group

Loading the recorded run…