What’s new in data strategy and why it matters now

The relationship between data and productivity has never been more discussed, and never harder to measure. Every week brings new platform announcements, new AI-assisted tools, and new promises about how analytics will help organizations do more with less. But the actual signals from the economy tell a more complicated story. Business formation is surging. AI adoption is accelerating. And yet survey after survey of data and technology leaders finds that the foundations required to make any of it work reliably are still catching up.

This issue tries to sit honestly with that tension. We look at what business formation data is showing right now and why the numbers deserve more attention than they typically get. We cover what is shifting in the analytics platform market, including a direction that matters more than any single product release. We unpack what several recent surveys of data leaders are actually saying about the gap between AI confidence and AI readiness. And we cover why the semantic layer, a concept that has lived inside data and analytics teams for years, is finally becoming a strategic priority now that AI tools have raised the cost of inconsistent metric definitions.

In the Actionable Insight section, we focus on a habit most analytics teams skip: looking back at what they built, whether it got used, and what to do differently. It does not require new tools or a reorganization. It requires a few honest questions asked in the right order.

As always, the goal is clarity over coverage. We are trying to help you focus on what matters and figure out what to do with it.

Business formation is running well above its historical pace — and that has real implications for planning

The United States has been creating new businesses at a rate that most historical benchmarks would not have predicted. According to the Census Bureau's Business Formation Statistics, 5.62 million business applications were filed in 2025, up 8.2 percent over 2024. The first several months of 2026 are running approximately 17 percent ahead of the same period last year, with April 2026 alone seeing more than 503,000 applications. For context, the annual average since 2005 has been roughly 3.47 million applications.

That caveat aside, the trend is real and durable. It reflects a combination of structural factors: remote work lowering the barriers to self-employment and independent business activity, platform infrastructure making it easier to operate a small business, and a labor market that has pushed more workers toward independent arrangements. Sun Belt states in particular have shown faster growth in new applications than many Midwest and Northeast states, reflecting broader regional economic shifts.

For business leaders, the planning implication is straightforward even if the data is often overlooked. A sustained surge in business formation over multiple years changes the competitive landscape in ways that aggregate market data can obscure. More businesses mean more competition for customers, more competition for employees, and more fragmentation in markets that may have previously been dominated by established players. It also signals where labor supply is moving and what kinds of business activity people are choosing to pursue on their own rather than through employers. Organizations that track business formation trends by region and sector have a more complete picture of the environment they are operating in than those that rely only on industry-level revenue or employment data. 

Source: U.S. Census Bureau, Business Formation Statistics, April 2026 release https://www.census.gov/newsroom/press-releases/2026/business-formation-statistics-may13.html

Tableau signals a shift from reporting to action at its annual conference

At Tableau Conference 2026 in May, Salesforce unveiled what it is calling an Agentic Analytics Platform spanning Tableau Cloud, Server, Desktop, and its newer Tableau Next product. The announcement framed the direction explicitly: the company is moving from helping users understand what happened to helping organizations orchestrate what happens next. Key additions include Tableau Next MCP, an Inspector tool accessible from within Slack, and a Tableau App for Microsoft 365. Separately, the April 2026 feature release added point-in-time metric capabilities through Tableau Pulse, richer filter and parameter actions, and
improvements to how data sources are managed across cloud and on-premises environments.

The theme across these announcements is not just feature expansion. It is a repositioning of what BI is supposed to do. For years, business intelligence was primarily a reporting and visualization layer. Teams built dashboards, leaders looked at charts, and analysts investigated anomalies. What Tableau is describing now is a platform where analytics is embedded in workflow: surfaced inside Slack, connected to agent-driven decisions, and capable of triggering responses rather than simply presenting information.

Most organizations are not yet operating at that level of integration, and the ambition of a vendor roadmap is not a reason to rebuild an analytics environment immediately. But the direction matters. If your current BI environment is essentially a report-delivery system, it is worth asking whether it is positioned to serve the business as expectations shift. The more immediate question is whether the foundations are in place: consistent metric definitions, trusted data sources, and clear ownership of what the numbers mean. Those fundamentals are prerequisites whether your organization ever uses agentic analytics or not.

Source: Tableau Conference 2026 announcements https://www.salesforce.com/news/stories/tableau-agentic-analytics-platform-announcement/

The AI readiness gap is wider than most organizations want to admit

Several independent surveys published in the first half of 2026 point to the same finding, and the consistency across sources makes it worth taking seriously. A study from Cloudera and Harvard Business Review Analytic Services found that only 7 percent of enterprises say their data is completely ready for AI. A separate Cloudera survey of nearly 1,300 global IT leaders found that while 96 percent of organizations report integrating AI into core business processes, approximately 80 percent say their AI and data initiatives are still constrained by limited data access across environments. Grant Thornton's 2026 AI Impact Survey found that insufficient data readiness is the third-leading cause of AI underperformance, and that 55 percent of CIOs and CTOs report fewer than half of their core applications are AI-ready.

The pattern that emerges across these studies is not that organizations are failing to adopt AI. It is that adoption is running well ahead of the data infrastructure required to make it reliable. Organizations are deploying AI tools into environments where the underlying data is inconsistently defined, difficult to access across systems, and only conditionally trusted by the people who use it. AI tools amplify whatever is already in the data. Inconsistency at the input level does not disappear; it scales.

For leaders outside the data function, this is worth understanding because it reframes what AI investment actually requires. The technology is rarely the binding constraint. The binding constraint is usually the quality, consistency, and governance of the data the technology runs on. Organizations that have invested in those foundations are moving faster and with more confidence. Organizations that have not are discovering that even sophisticated AI tools cannot compensate for data environments that were not designed with this kind of use in mind. The gap between those two groups is widening, and it is showing up in measurable differences in AI return on investment.

Sources:

• Cloudera and Harvard Business Review Analytic Services, Taming the Complexity of AI Data Readiness (March 2026) https://www.cloudera.com/about/news-and-blogs/press-releases/2026-03-05-only-7-percent-of-enterprises-say-their-data-is-completely-ready-for-ai-according-to-new-report-from-cloudera-and-harvard-business-review-analytic-services-reveals.html

• Cloudera, The Data Readiness Index (April 2026) https://www.cloudera.com/about/news-and-blogs/press-releases/2026-04-14-nearly-80-percent-of-enterprises-say-ai-is-held-back-by-data-access-challenges-cloudera-report-finds.html

• Grant Thornton, 2026 AI Impact Survey (April 2026) https://www.grantthornton.com/services/advisory-services/artificial-intelligence/2026-ai-impact-survey

The semantic layer is becoming a strategic conversation

For most of its history, the semantic layer, which is the part of a data environment where raw data gets translated into consistent, business-defined terms like revenue, customer, or churn, was a concern that lived almost entirely within data engineering teams. That is changing, and the shift is happening faster than most business leaders have noticed.

In January 2026, an open-source industry specification for semantic interoperability reached version 1.0, backed by dbt Labs, Snowflake, Databricks, and Microsoft. The practical effect of that alignment is significant: the major data platforms are now converging on a shared standard for how business metrics get defined, stored, and made available across tools. Snowflake Semantic Views reached general availability in early 2026. dbt's MetricFlow engine, which powers its semantic layer, was open-sourced and updated with a simplified specification in January. Microsoft acknowledged dbt as a standard for analytics engineering and announced first-class support for it within Microsoft Fabric. When competing platforms invest in the same open specification within the same quarter, it signals that the underlying problem, inconsistent metric definitions across tools and teams, has become too costly to ignore.

The reason this matters beyond data teams is that AI-assisted analytics has raised the stakes for metric consistency considerably. When a human analyst pulls a revenue number from a dashboard and it does not match the number in a different report, the friction is annoying and trust-eroding, but it is manageable. A person can ask a question, track down the discrepancy, and reconcile the definitions. When an AI tool generates analysis or takes action based on a metric definition it retrieved from raw tables, there is no such reconciliation step. The tool uses whatever definition it finds, and if that definition is inconsistent or ambiguous, the output is wrong in ways that may not be immediately visible.

For leaders outside the data function, the practical implication is this: the organizations that have invested in consistent, well-governed metric definitions are significantly better positioned to get reliable output from AI tools than those that have not. The semantic layer is not a new technology. It has been discussed in data circles for years. What is new is that it is now a prerequisite for a category of capability, AI-assisted analytics, that most organizations are actively trying to deploy. The conversation belongs at the leadership level, not just the architecture level, because the question of what "revenue" or "customer" or "utilization" means in your organization is a business question before it is a technical one.

Source: Unwind Data, "Semantic Layer Industry Standard" (March 27, 2026) https://unwinddata.com/semantic-layer-industry-standard

Running an analytics retrospective: a habit most teams skip and most programs need

Analytics teams spend most of their time building forward. A request comes in, work begins, something gets shipped, and attention moves to the next request. It is a natural rhythm, and in most organizations it is relentless enough that there is rarely a pause to look back. That is a problem, because the questions that would most improve an analytics program are almost always backward-looking ones. What did we build over the past six months? Who used it? What decisions did it actually support? What are we maintaining that nobody is asking for?

A deliberate analytics retrospective is a structured answer to those questions. It is not a project postmortem, which typically focuses on a single initiative and what went wrong. It is a periodic review of the full portfolio of analytics work: what was built, what is being used, what is being ignored, and what that pattern reveals about where the program is and is not working. Done well, it takes a half-day and produces three things: a clearer picture of where analytical capacity is actually going, a list of things that can be retired or simplified, and an honest conversation about whether the right work is getting prioritized.

Most analytics teams that try this for the first time discover the same few things. A meaningful share of active reports and dashboards have not been opened in months. Several reports that get opened regularly are not connected to any identifiable decision. Some of the most-used outputs are informal: a spreadsheet someone built three years ago, a Slack message with a screenshot, a monthly email with a few numbers. And some of the work the data team is most proud of is the work stakeholders understand the least.

None of that is failure. It is information. The retrospective is valuable precisely because it surfaces these patterns and gives the team something concrete to act on. 

The structure that tends to work best involves five questions, each examined across the full inventory of active work. These are some of the same questions our consultants ask when we are helping an organization formalize their data strategy. https://consulting.sva.com/solution/data-strategy

The first is simply: what exists? Most analytics teams do not have a clean, current list of everything they maintain. Reports accumulate. Dashboards get created for a specific purpose and never retired. Before any evaluation is possible, an honest inventory is necessary. This step alone is often surprising. Teams discover they are actively maintaining more than they realized, and that a significant portion of it was created for a purpose that has since changed or been resolved.

The second question is who is actually using this, and how often? Usage data is usually available. Most BI platforms log view counts, last-opened dates, and user activity. Pulling that data and matching it against the active inventory is often the most illuminating part of the exercise. The goal is not to shame low-viewed reports but to identify which outputs have genuine ongoing demand and which are maintained by habit rather than need. A report that was opened once in the last 90 days is probably not worth the maintenance burden it carries.

The third question is what decision does each output support? This is harder than it sounds, and the difficulty itself is informative. If the team cannot answer this question for a given report, either because the purpose was never clearly defined or because the original need no longer exists, that report is a candidate for retirement. If the answer is that it supports a decision but nobody on the business side is making that decision based on the data, that points to a different problem: a gap between what the data team is providing and what stakeholders are actually using to decide.

The fourth question is what is missing? The retrospective is not only about reducing; it is also about identifying where demand exists that the current portfolio does not serve. Stakeholders who are working around the analytics environment, using spreadsheets, pulling one-off data requests, or making decisions without data, are signaling unmet need. A structured conversation with business partners about what they are not getting is as valuable as an audit of what is being ignored.

The fifth question is what should change as a result? A retrospective with no output is just a review. The output should be a short list of concrete decisions: reports to retire, outputs to simplify, definitions to reconcile, or gaps to prioritize. The list does not need to be long. Three to five clear actions, owned by specific people and attached to a timeline, is more useful than a comprehensive improvement plan that never gets executed.

The full exercise works best as a shared effort between the data team and at least a small group of business stakeholders. A retrospective conducted entirely within the data team will miss the perspective of the people who are or are not using the outputs. A retrospective conducted once is useful. One conducted every six months becomes a governance habit that keeps the analytics portfolio aligned with what the business actually needs rather than what it needed at some point in the past.

A reasonable place to start: pull usage data for every active dashboard and report in your primary BI environment for the past 90 days. Sort by last-opened date. Look at the bottom third of that list. Ask whether those outputs are being maintained because they are needed or because nobody has stopped to ask the question. That audit, done in an afternoon, is usually enough to open a conversation worth having.

CLOSING

The thread running through this edition is the gap between what organizations have built and what is actually being used. Business formation data shows more economic activity than most planning models account for. AI adoption is running ahead of the data foundations that would make it reliable. The platforms organizations depend on are converging on metric consistency as a prerequisite for AI, not an optional enhancement. And analytics programs accumulate outputs that outlast the questions they were designed to answer.

None of these are problems that require a new platform or a larger team. They are problems that require a clearer look at what is already in place and an honest conversation about what is actually working. That is harder than buying something new, and usually more useful.

If any of this edition's topics connect to challenges your organization is working through, we are glad to continue the conversation.

data@work is published by SVA Consulting. Each issue is designed to help business leaders understand what is changing in the data and analytics landscape, why it matters, and what to do with it.