A shared language
Customer, order, channel, revenue, margin, activity, and retention are defined consistently across P&L, CRM, product, marketing, and operations. The team debates decisions rather than versions of a number.
What a data strategy delivers
Data strategy connects business questions, definitions, sources, data products, and accountability for decisions.
Data priorities
The business collects what changes key decisions rather than everything possible. For every dataset, the user, frequency, required accuracy, and cost of error are clear.
From insight to action
A report ends with an owner, deadline, expected impact, and outcome check. Analytics becomes part of the management cycle rather than a parallel function.
Problems management analytics solves
The symptom often looks like a missing report, while the cause may be an unclear question or fragmented accountability.
No unified view of the business
Channels and functions show their own results, but the executive cannot see the shared customer, full cost, or the impact of initiatives on P&L.
Marketing is disconnected from sales
Last-click attribution overvalues the bottom of the funnel. Contacts, customers, orders, margin, and subsequent behavior must be connected.
Averages conceal causes
An aggregate metric masks differences among categories, segments, and cohorts. Detail must explain a decision rather than merely increase the number of cuts.
Decisions are made too late
Manual assembly and reconciliation take more time than the business has. Recurring data products must refresh at the speed of the decision and reveal deviations before the final loss.
Target data system
Architecture follows the management model and scale rather than fashion for a particular tool.
Sources and identifiers
Sales, products, customers, communications, digital events, and operations are connected through stable keys. Quality rules and change handling are defined before data marts are built.
DWH and definition layer
The DWH stores an agreed history and creates a shared metric layer. It separates transactional systems from analytical tasks and makes calculations reproducible instead of manually assembling a number.
Data products
A dashboard, segment, forecast, or signal is treated as a product with a user, owner, SLA, and measurable utility. An unused report is retired from support.
Governance
A specific role is accountable for definition, access, quality, and use. Authority is limited to the task, while sensitive data is protected at source, product, and process levels.
From business question to a working system
Implementation proceeds in short cycles where the decision and user are visible from the start.
1. Question and baseline
Define the decision that is made poorly today: pricing, assortment, retention, inventory, service, or investment. Record the current metric and cost of error.
2. Definitions and data
Agree entities, calculations, sources, historical depth, and acceptable quality. Unknowns are recorded as constraints rather than hidden behind a complex model.
3. Minimum product
The team creates the simplest report, data mart, or signal that can change an action. Manual steps are acceptable to test usefulness before large-scale automation.
4. Integration into the process
Define the moment of use, decision owner, procedure, permissions, and next step. The user understands what to do when the metric is normal, deviates, or lacks data.
5. Impact and scaling
Measure the change in the decision and business metric. After usefulness is proven, expand the audience, data depth, and automation while preserving rollback capability.
End-to-end metrics and attribution
An end-to-end model connects the contact, customer, order, margin, and repeat behavior. It shows the role of communications, the store, app, contact center, and service—not only the last-click channel. The practical result is an ability to reallocate budget according to actual contribution, find cannibalization, and evaluate omnichannel scenarios. The method must be explainable to P&L owners and resilient to changes in data sources.
Predictive analytics and management decisions
A forecast of demand, churn, availability, or workload is useful only together with an action: change replenishment, contact priority, price, capacity, or service routing. Predictive analytics is therefore designed together with the management response. Accuracy is compared with the current decision method and assessed through the cost of error. Sometimes a simple forecast with a clear response creates more value than a complex model the process cannot use.
Economics and data quality
System cost includes integrations, storage, calculations, support, quality control, and users’ time. Priority goes to a product where expected impact and decision frequency justify TCO. Quality is measured through completeness, timeliness, consistency, and the impact of error on a specific action rather than abstract cleanliness. Critical metrics need owners, monitoring, and a clear recovery scenario.
It is useful to calculate the cost of manual work, decision delay, and wrong action as well as platform cost. Sometimes the first investment should fix why source data appears late or without the required identifier rather than add a new system. This choice lowers recurring costs and builds user trust more effectively than another visualization layer.
When AI is needed, and when data comes first
If the business cannot define customer, order, or margin consistently, it first needs a factual foundation and management process. AI is useful where the task already has data, an action, a quality criterion, and economics. Use-case selection, pilots, human-in-the-loop, and scaling are covered separately on the page AI for business.
A separate model will not fix fragmented accountability across functions or create an action missing from the process. The executive first defines the decision and owner; the team then chooses a sufficient analytical or AI tool.
Practical cases
Metrics belong to specific companies and demonstrate management experience rather than abstract technology expertise.
ORTEKA
Customer analytics, attribution, product, and commercial initiatives formed one portfolio with an impact of +RUB 165m. This is the impact of the full change portfolio, not a single analytics or AI mechanism.
Kenguru
CRM campaigns delivered RUB 40m+, while personal scenarios relied on customer, behavioral, and assortment data. AI personalization has its own separately measured result.
SUNLIGHT
A unified operating system supported e-commerce growth from RUB 5.6 bn to RUB 13 bn and moved inquiry resolution from 20% to 98%.
Questions about data strategy
Brief answers on priorities, platforms, and impact.
Where should a data strategy begin?
With a list of decisions and losses, not a target platform. Initial data products are chosen by frequency, value, and the team’s ability to change an action.
Is a unified DWH necessary?
A shared analytical layer is usually needed for end-to-end history and agreed metrics. Its scale and technology depend on sources, users, speed, and quality requirements.
How can the value of analytics be proven?
Record the decision baseline, owner, action, and business metric. Utility is demonstrated by a change in outcome or speed, not the number of reports and users.
Related areas
Discuss a data challenge
Describe the decision the business makes too slowly or inaccurately, the available sources, and the cost of error. At the first meeting, we will determine whether a data audit, analytics product, or broader management system is needed.