Client story  /  AI training data

Toloka

toloka.ai
Toloka
The offering
Human-labelled data for machine learning at scale: the annotation, classification and quality checking that a model needs before it is worth anything.
The task
Find out whether Gulf companies were building anything that needed training data yet.
21
Meetings booked
20
Companies introduced
UAE
Market
11 months from 2022
Engagement
Industries we sold into
Retail and groceryE-commerceTechnology and IT services
A number we left out
The share of these meetings that went ahead was below the bar we publish at, so it is not on this page. The client and the companies are, because the work was real.
More stories
Yango Ads Advertising technology859 booked Go Mobile Mobile marketing30 booked LemonAI Predictive marketing analytics25 booked

In 2022 we were hired to sell the ingredient a year before the market started cooking.

This engagement ran through 2022 and into 2023, which in the history of artificial intelligence is the year before everything. Our reps were explaining why a model needs labelled data to retailers and marketplaces who, for the most part, did not yet have a model.

Some markets are not slow. They are early. The distinction matters, because the tactics that fix a slow market make an early one worse: more calls into an audience with no project produce more polite conversations and no pipeline.

What we did

Eleven months across the UAE, in retail, e-commerce, mobility and technology services. 21 meetings booked with 20 companies.

IKEA, Sharaf DG, Dubizzle, ekar, Lenskart, 6thStreet, Max Fashion, Apparel Group, BFL, Eros Group, elGrocer, FNP, Momos and the Roads and Transport Authority, plus regional AI consultancies including ADDO AI.

What an early market actually looks like

The signature of an early market is a good meeting that leads nowhere, repeatedly. The prospect is interested, asks intelligent questions, is glad you called, and has no budget line because the project that would consume the product does not exist yet.

An outbound team can read that pattern in about six weeks if it is looking for it. The tell is the second meeting: in a real market the buyer brings a colleague, and in an early one he brings enthusiasm. Counting second meetings rather than first ones is the cheapest early-market detector we know, and we now build it into the reporting on every new category we take on.

The postscript belongs to the client rather than to us. The company is now part of Nebius and works with some of the largest names in artificial intelligence, and the demand it was looking for here arrived a couple of years later. Being early is not the same as being wrong. It is expensive in a way that being wrong is not, because everything you learn stays valid.

What it shows

Distinguish an early market from a slow one, quickly, and say which one you are in. The measurement that tells you is the second meeting. When a category has not arrived, the honest recommendation is to spend less and wait, and an agency that cannot say that is charging a client to be patient.

Every figure counted from the activity tracker.
book_a_call