TIME SERIES MODELS IN FORECASTING DIGITAL PERFORMANCE INDICATORS OF ENTERPRISES

Abstract

The rapid expansion of digital business operations has established a distinct class of management metrics, digital performance indicators (digital KPIs), whose volume, frequency, and structural complexity differ fundamentally from traditional financial metrics. These data streams exhibit specific characteristics: high observation frequency, multi-level seasonality (intra-day, weekly, and annual cycles), heightened volatility, and flow irregularity with gaps and outliers. The article systematizes classical time series models, including trend models, moving averages, exponential smoothing (ETS), and ARIMA/SARIMA, evaluating their applicability to digital KPIs. Methodological analysis indicates that Holt-Winters triple exponential smoothing is optimal for indicators with trend-seasonal patterns, which characterize most commercial digital time series, whereas SARIMA models are appropriate for stochastic series with sufficient observation volume. To resolve the fragmented application of forecasting techniques, the study develops a methodological selection matrix linking seven enterprise digital indicators (web traffic, online sales, conversion rates, social media engagement, email marketing metrics, advertising spend/ROAS, and lead acquisition) to recommended forecasting models, forecast horizons, and minimum implementation conditions. Unlike conventional approaches that evaluate models in isolation, the proposed matrix integrates indicator types, time series attributes, model specifications, and practical constraints into a cohesive decision-support framework. The results demonstrate that classical time series models remain robust tools for digital performance forecasting, providing enterprises with practical instruments to enhance managerial decision-making and advance data-driven governance.