Skip to main content

Better data does not guarantee better decisions: 5 lessons from the Data & Analytics Summit

Photo from event

Retail and consumer packaged goods organizations have invested heavily in data platforms, analytics, AI, and sophisticated dashboards. Yet one message surfaced repeatedly at the 2026 Data & Analytics Summit in Chicago: Better data alone does not create better decisions.

The gap between insight and action remains stubbornly wide. A model can identify an opportunity, but a sales team may not trust it. A dashboard can show a change in customer behavior, but different departments may interpret the metric differently. An AI agent can retrieve the latest information, but without historical context, it may misunderstand why an earlier decision was made.

That challenge is becoming more urgent as customer journeys cross digital and physical channels. Research presented at the summit found that online sales contributed more than 70% of total grocery sales growth in 2025. At the same time, 94% of consumers who purchased groceries shopped both online and in stores. The distinction between a digital customer and an in-store customer has become increasingly artificial.

The lesson for leaders is clear: Your data strategy must support a decision system, not simply a reporting system. Five capabilities make that possible.

1. Build trustworthy measurement around the decision

Organizations often search for one authoritative metric or measurement method. In practice, no single method can answer every business question.

Deterministic measurement can provide fast, detailed insight when you have consented identity data. Marketing mix modeling can help you understand broader performance and guide budget allocation without relying entirely on individual-level tracking. Controlled experiments can determine whether an intervention caused an outcome rather than merely occurring alongside it.

Start with the decision, then select the evidence:

  • Use observed behavioral data to understand what customers did.
  • Use modeled evidence to estimate performance when complete observation is unavailable.
  • Use experiments to establish whether a campaign, promotion, or experience created incremental value.
  • Use operational metrics to determine whether teams executed the intended change.
  • Use customer and employee signals to understand why the outcome occurred.

This combination matters in an omnichannel environment. A customer may discover a product through retail media, compare options in an app, purchase it in a store, and contact customer service through chat. If each team measures only its own interaction, every dashboard may be technically accurate while the broader interpretation remains wrong.

Trustworthy measurement also requires transparency. Document what each method can show, where it has limitations, which assumptions it uses, and what decision it should inform. Instead of asking whether a metric is universally correct, ask whether it is sufficiently reliable for the decision at hand.

2. Create shared definitions before creating more dashboards

Many data disputes are definition disputes disguised as technical problems.

Marketing may define an active customer by recent engagement. Sales may use account activity. Customer service may focus on recent interactions. Finance may recognize a customer only after revenue is recorded. Each definition can be valid within its function, but combining them without reconciliation creates conflicting reports and weakens confidence.

A request for a “single source of truth” rarely means that every team must use one application. It means the organization needs consistent rules for interpreting key concepts across applications.

Create shared definitions for common terms and assign an owner to each definition. Record its calculation, source systems, update frequency, and approved uses. If teams need different versions, make the distinction explicit rather than allowing multiple unlabeled definitions to circulate.

This work may feel less exciting than deploying a new analytics tool, but it directly affects adoption. When two teams debate whose numbers are correct during every business review, leaders stop using the analysis to make decisions. Shared definitions move the conversation from “Which report should we believe?” to “What action should we take?”

3. Give analytics and AI a governed history

Traditional analytics usually emphasize the current state: the latest customer profile, sales forecast, inventory position, or model output. Decision-making often requires a different question: What did the organization know when it made the decision?

That distinction becomes critical when data changes after the fact. A retailer may restate sales information. A forecast may be revised. A customer may update consent. An account may move between territories. If the system retains only the newest value, an analyst or AI agent can inadvertently evaluate an earlier decision using information that was unavailable at the time.

A governed history should capture two timelines: When the information was true and when the organization learned or recorded it.

This “two-clock” model supports point-in-time analysis. It allows you to reproduce a prior calculation, understand the context behind an action, and determine whether a decision was reasonable based on the evidence then available.

For AI agents, that history becomes operational memory. A reliable agent may need access to:

  • A versioned record of state changes
  • A log of events and actions
  • Previous decisions and their rationale
  • Stable identities for customers, products, locations, and accounts
  • Consent, access, and retention policies
  • Approval records for actions that affected business systems

Governance must extend to anything an agent can change. Define which systems the agent may update, which actions require human approval, how the organization records those actions, and whether a reviewer can reconstruct what happened. An AI system that produces an answer is useful. One that can show which information, policies, and prior decisions informed the answer is much easier to trust.

4. Treat cross-functional alignment as part of the product

An analytics initiative can meet every technical requirement and still fail if the people expected to use it do not see their work reflected in it.

This often happens when business teams serve only as requestors and end users. They describe a need, an analytics team builds a solution, and the organization treats delivery as completion. If the result does not match the user’s workflow, terminology, timing, or decision rights, adoption slows. Employees return to spreadsheets, local reports, and workarounds.

Break that pattern by treating business teams as development partners.

For each use case:

  1. Define the decision. Identify who makes it, how often, and what happens next.
  2. Map the workflow. Determine where the insight must appear to be useful.
  3. Agree on success. Establish adoption, execution, and business-outcome measures.
  4. Test with real users. Observe where people hesitate, reinterpret, or ignore the recommendation.
  5. Create a feedback loop. Give users a clear way to report missing context and incorrect signals.
  6. Assign ownership. Specify who improves the solution after launch.

The goal is not to let every stakeholder redesign the solution. The goal is to combine analytical rigor with operational reality. Analytics teams contribute methodology, data knowledge, and product discipline. Business teams contribute customer context, decision logic, and an understanding of how work gets done.

That partnership builds trust over time. It also changes the definition of a finished product. A decision tool is not complete simply because it launches. It must evolve as customer expectations, market conditions, data sources, and workflows change.

5. Design adoption as a continuing business capability

Training at launch is not an adoption strategy.

Organizations frequently focus on the technical path to production while underestimating the behavioral change required afterward. A new tool may alter how employees prioritize accounts, evaluate promotions, plan inventory, respond to customer needs, or approve an AI-recommended action. Those changes require reinforcement, support, and visible leadership.

Build adoption into the roadmap from the beginning:

  • Assess current data literacy and workflow readiness.
  • Identify the roles most affected by the change.
  • Explain how the new capability improves a specific decision.
  • Use pilots to demonstrate value in a bounded environment.
  • Recruit champions from the teams expected to use the solution.
  • Provide office hours, role-based training, and communities of practice.
  • Measure usage, overrides, response times, and actions taken.
  • Review where employees still rely on manual workarounds.
  • Continue improving the product after deployment.

Avoid using logins as the sole measure of success. A stronger adoption scorecard asks whether people trusted the insight, acted on it, and achieved a better outcome. For an AI recommendation, for example, track how often employees accept, modify, or reject it, why they make that choice, and whether the resulting action improves revenue, efficiency, or customer experience.

Assess your decision infrastructure

The Summit’s most important takeaway was that organizations need to connect their data investments to the conditions that produce confident action.

Before adding another dashboard, model, data source, or AI agent, assess your decision infrastructure:

  1. Do leaders know which measurement method fits each major decision?
  2. Do departments use shared definitions for essential business concepts?
  3. Can you reconstruct what the organization knew when it made a decision?
  4. Do business and analytics teams share ownership of analytical products?
  5. Can you show that employees act on insights and improve outcomes?

Weakness in any one area can limit the value of the others. Better measurement cannot resolve conflicting definitions. Shared definitions cannot compensate for missing history. Governed history will not create value if teams do not trust the tool. Alignment will fade if the organization does not support adoption.

Better data remains essential. But the advantage comes from building a system in which people and technology can interpret the data consistently, understand its history, evaluate it appropriately, and turn it into action.

That is the difference between collecting more intelligence and making better decisions.

 

Photo courtesy of My Digital Shelf.