Cutting Weekly Campaign Reporting from 3–4 Hours to 15–30 Minutes

Weekly reporting inside VK Ads

VK Ads is a digital advertising platform for campaigns across VK properties. While the platform serves both internal and external users, this initiative focused on managers inside the organisation.

Every manager prepared a weekly report on key campaign metrics. No reporting product existed when I started; I proposed it as a new internal tool.

I initiated the product and owned its roadmap

As Product Manager, I identified the opportunity and owned the product from zero: user research, problem definition, workflow design, MVP scope, prioritisation and roadmap.

I worked directly with one developer. I set the product boundaries and validation approach, and each round of testing shaped the next release.

Three to four hours of manual work for every weekly report

Managers spent 3–4 hours each week manually transferring campaign data, calculating metrics and assembling a report. Because each person also formatted the output differently, the process was difficult to repeat consistently.

Errors created a second workload: teams had to find mistakes, correct calculations and rebuild parts of a report before using it. We did not need to automate judgement; we needed to remove the repetitive work that came before it.

Security, a two-person team and a firm boundary for AI

The workflow involved confidential campaign data, so this public case excludes source data, internal screens and implementation detail. Off-the-shelf tools could not meet internal security requirements or adapt to the reporting workflow.

A two-person delivery team meant the first version had to focus tightly on the highest-friction steps.

We could validate calculations and structured data transfer against reference outputs. Generated conclusions were less dependable, so they could assist review but not determine the final action.

Start with the people preparing the reports

I interviewed the managers doing the work. Instead of asking which features they wanted, I asked why reports took so long and where mistakes entered the process.

I broke the workflow into data transfer, calculations, report population, presentation and interpretation. The MVP automated the core transfer and population loop, left several supporting steps manual and was tested with one manager before wider rollout.

Automate preparation, keep decisions with managers

The new workflow separated repeatable processing from context-dependent judgement. Managers received a consistent starting point and remained responsible for interpretation and action.

A standard report built from structured campaign data

The workflow processed structured campaign interactions such as impressions, clicks, retention and button activity. It then calculated metrics including spend, CPC and CTR, and populated a consistent campaign report.

The MVP proved the core data-transfer and report-filling loop while leaving supporting steps manual. Later versions expanded parsing and calculation coverage. In v2.0, we added AI-assisted highlights; v2.3, launched in December 2025, enabled one-action population of the reporting database.

Five decisions kept the workflow useful and safe

15–30 minutes per report, adopted across the department

Weekly report preparation fell from 3–4 hours to approximately 15–30 minutes, and every manager in the department adopted the workflow. We reviewed calculation errors and discrepancies quarterly. In the latest retained window, which covered four months, approximately 1% of reports contained an error or discrepancy.

For each release, we checked calculations against 100 human-verified reference reports. After corrections, the outputs matched the reference set. AI-generated highlights remained advisory and were deliberately excluded from the accuracy claim.

The department added the workflow to its operating instructions, and other teams used those instructions as guidance.

Measure AI assistance separately from calculation accuracy

Interviewing the people who produced reports every week gave us a clear initial scope and helped us choose the right sequence of work.

Next time, I would evaluate assisted highlights separately from calculation accuracy. Manager feedback on the highlights was encouraging, but a task-specific usefulness measure would make that part of the product as testable as the reporting calculations.