Demand Forecasting for Massive Request Seasonality

TL;DR

Big-tech demand repeats on the client’s own business cycle, so request volume by language is predictable a year out – which is what removes the need for emergency scaling when a seasonal burst lands.
The deliverable was a process, not a campaign: reporting between Marketing and BizDev with the planned schedule against the actual one, and business margin known by language and request type.
Fulfillment on key client requests held at 100% and the acquisition budget for major language segments fell 22% with revenue unchanged.

Situation

Toloka is a crowdsourcing platform built inside Yandex – spun out in 2025 with Nebius Group (ex-Yandex) as its majority shareholder – that connects businesses and researchers with a distributed crowd to complete various micro-tasks, such as data labeling, image recognition, and content moderation.

Toloka's revenue is connected with client requests' fulfillment, which workers perform. The number of completed tasks per day and the number of days the worker stays active on the platform are critical variables for business margins.

As Senior Marketing Manager, B2C, I was responsible for acquiring, engaging, and retaining these workers.

Task

For the Toloka Business Development team, it is critical to fulfill key clients' requests (big tech) entirely and on time, regardless of internal processes and optimization issues. These requests usually have a significant volume and obvious seasonality (meaning that their requests don't come gradually within a period of one year). As a result, a massive burst in the number of workers is needed immediately, significantly affecting the business margins and net margins of these requests.

Tasks like these happen in different types of businesses that deal with substantial seasonal sales during holidays or other events. The common feature of this particular ask is that these sales have strict time limits and KPIs.

The task was in handling big requests, handling standardization, and connected costs minimization, which are related to:

excessively rapid ad campaigns' scale;
acquiring too many workers in case of a gap between requests.

Action

To meet the challenges, I've decided to focus on the following areas:

find the metrics to evaluate worker's efficiency and calculate workers' unit economy;
increase workers' retention to key projects;
forecasting and planning upcoming requests fulfillment.

Here, we will dwell more on actions and results in area 3. We'll review areas 1 and 2 in separate cases.

During the research phase, it became apparent that it's crucial to build the processes of key requests' forecasting and workers' acquisition optimization via:

understanding of the request volume distribution among different languages, since the requests' repetition and relatively stable annual demand are driven by clients' business nature (in other words, we can, indeed, predict with considerable accuracy the request volume distribution among the languages);
understanding of the request types and other clients' business requirements;
regular communication and planning, as well as building the reporting system between Marketing and BizDev teams.

Result

The user acquisition budget for major language segments decreased by 22%, while Toloka's revenue remained unchanged.
Maintained the key requests' fulfillment success rate of 100%.
Built a transparent collaboration process between Marketing and BizDev teams, in particular:
- built communication and planning processes regarding terms of order fulfillment and required business margins between Marketing and BizDev teams;
- built the reporting system, which indicated the planned as well as actual schedule of the clients' requests fulfillment, which let us adapt the plan and understand what we should adjust to meet the schedule and business margins;
- understanding the business margins by language and request types, which lets the BizDev team prioritize and make more margin deals.

Results in numbers

Metric	Target	Result	How measured
Key-request fulfillment rate	–	100% maintained	Success rate on big-tech client requests with strict deadlines
Acquisition budget, major language segments	–	−22%	Budget against revenue; revenue held flat
Marketing and BizDev planning	–	Reporting system, plan against actual schedule	Planned and actual fulfillment schedule, business margin by language and request type

Table 5. Outcomes as stated in this case. No numeric hypothesis was set before this work – the deliverable was a process, and the constraint was a hard fulfillment KPI.

"Maria achieved great results in the acquisition, retention, and engagement of ML data makers thanks to her strategic thinking, data analysis skills, project management skills, understanding of marketing tools and technologies, strong work ethic and attention to detail, and continuous improvement mindset."
– Dmitry Stepanov, Founder & GP at AAL VC · Forbes 30 Under 30 · Senior to Maria at Toloka

Written by Maria Atamanova – Senior Growth & Performance Marketing Manager and applied mathematician; twelve years growing businesses – apps, e-commerce, EdTech, AI data platforms – on numbers, not gut feeling. This is the planning half of the Toloka work; the efficiency metric behind the budget cuts is in the unit economy case.

More cases

How Marketing Can Help Find Product-Market Fit – Toloka

65% retention increase · 26% budget cut · revenue unchanged
How to Build a Unit Economy Model to Improve Media Buying Strategy – Toloka

44% acquisition budget cut · revenue unchanged