Microsoft Completed Pre-Training a 2.8 Trillion-Parameter Model That Claims 4–12x the Compute Efficiency of DeepSeek. MAI-Base-2 Hasn't Shipped Yet, But the Numbers Signal a Strategic Shift.

Salesforce signed a definitive agreement on October 12, 2026 to acquire Loops, the Israeli AI analytics startup founded in 2020 by CEO Tom Laufer and CTO Ido Adiv. The deal is expected to close in early November within Salesforce's fiscal year 2027. The price was not officially disclosed. Calcalist, the Israeli business newspaper, estimated approximately $30 million — a figure neither company confirmed. The technology will be folded entirely into Tableau.

On the surface, a $30 million acquisition is a rounding error for Salesforce, which paid $15.7 billion for Tableau in 2019. But the Loops deal is not about the price. It is about a capability that Tableau has never had, that every enterprise analytics platform is racing to build, and that represents the single largest unfulfilled promise in business intelligence: explaining not just what happened to your metrics, but why.

The $30 Million Bet on the Question Every Dashboard Can't Answer

The history of business intelligence is a history of progressively better answers to one question: what happened? The spreadsheet told you. The data warehouse told you faster. Tableau told you visually. Tableau Pulse, introduced in 2023, started telling you proactively — sending digests when metrics moved rather than waiting for someone to open a dashboard and notice.

But in every iteration of this evolution, the user still had to ask. And the answer was always descriptive: this number went up, that number went down, here is the trend line for the past 90 days. The question every business decision-maker actually needs answered — why did retention drop 4 points last month? — remained outside the scope of any BI platform. It was delegated to an analyst, who would spend hours running SQL queries, building hypotheses, testing correlations, and eventually producing a report that was typically obsolete by the time it was ready.

The Death of the Dashboard covered the structural forces behind this shift — the migration from static visualization toward natural language and AI-driven investigation. Loops is the product that makes that migration concrete at the enterprise analytics layer.

What Loops built, and what Salesforce is buying, is a causal inference engine wrapped in an agentic product interface. The Scout agent does not wait to be asked. It monitors your KPIs continuously, detects material changes, identifies the variables most likely to have caused those changes, ranks them by estimated impact magnitude, and delivers that explanation to Slack, Teams, or email before a human analyst has opened the dashboard to notice the anomaly.

What Traditional BI Actually Does — and Where It Stops

To understand why this acquisition is more significant than its price suggests, it helps to be precise about what BI tools have always done and where they have always stopped.

A business intelligence platform is, at its core, a system for aggregating, storing, and visualizing historical data. It can show you that daily active users declined from 142,000 to 136,000 over a two-week period. It can show you that the decline happened more sharply in mobile than desktop. It can overlay acquisition channel data to show that users acquired through paid social performed worse than users acquired through organic search in that cohort.

What it cannot do is tell you whether the paid social users performed worse because of something intrinsic to the acquisition channel, because of a product change that happened to coincide with when paid social users hit their first meaningful engagement milestone, because of a seasonal pattern that suppresses paid social user quality in October, or because of an interaction between all three. Answering that question requires counterfactual reasoning: what would have happened to retention in the paid social cohort if the product change had not occurred? BI tools do not reason counterfactually. They show correlations.

The enterprise analytics burden this creates is substantial. A typical enterprise product team monitors dozens of KPIs. A growth team might track hundreds of metrics across acquisition channels, funnel stages, feature adoption rates, and retention cohorts. When any of those metrics moves materially, someone has to explain why — to the product manager, to the exec team, to the board. That explanation work is the primary time sink for most analytics teams. It is repetitive, high-stakes, and frustratingly manual given how much other data infrastructure has been automated.

Most enterprises manage this problem by accepting that most metric movements will never be fully explained. The analyst capacity simply runs out before the investigation queue is cleared. Teams learn to investigate only the most urgent and most visible movements and let everything else scroll by. The result is a systematic blindspot in business decision-making: the majority of signals in the data are never converted into explanations, which means they are never converted into decisions.

Scout: The Agent That Replaces the Analyst's Root Cause Question

According to TechTarget's coverage of the acquisition, Scout is Loops' core product — an analytics agent that autonomously monitors KPIs, analyzes root causes of changes, and identifies opportunities to improve business performance. The product's key differentiator is its causal inference engine, which attempts to distinguish genuine causal drivers of metric changes from variables that merely correlate with those changes.

In practice, a Scout deployment looks like this. A product team defines a set of KPIs they want monitored — 30-day retention, activation rate, feature adoption, time-to-value, and similar product health metrics. Scout connects to the relevant data sources and begins continuous monitoring. When a metric moves outside its expected range — defined by historical baseline and statistical significance thresholds — Scout does not just send an alert. It begins an automated investigation.

The investigation queries across the full set of variables available in the connected data: user acquisition channels, cohort characteristics, product usage patterns, feature flags that were active for different user segments, geographic distributions, device types, and any other dimensions available in the data. Scout's causal model applies statistical methods that go beyond correlation analysis to estimate which variables are most likely to have caused the observed change, ranked by estimated impact magnitude.

The output is an explanation, not a correlation matrix. Instead of showing a product manager a table of correlations and asking them to interpret it, Scout delivers a ranked list: "The 3.2-point drop in 30-day retention was driven primarily by lower activation in the new onboarding flow introduced October 5 (estimated impact: -2.1 points), partially offset by improved retention in users who completed the setup wizard (estimated impact: +0.4 points), with residual effect from seasonal patterns in the enterprise cohort (estimated impact: -0.7 points)."

Scout can also deliver findings proactively through Slack, Microsoft Teams, or email. Teams do not need to log into Tableau to receive explanations; the explanations arrive in the workflow tools they already use. This delivery mechanism matters because the value of an explanation degrades rapidly with time — a root cause finding delivered hours after an anomaly is detected enables faster intervention than one delivered days later.

A notable additional capability is impact measurement without formal A/B tests. Most impact measurement requires splitting users into control and treatment groups before a change is made. Scout can estimate the causal impact of changes that were not set up as controlled experiments, using observational causal methods to control for confounding variables. For product teams with insufficient traffic volumes for conventional experimentation or where running a controlled test was not logistically possible, this capability provides a path to impact measurement that did not previously exist in commercial BI tooling.

Causal Inference Is the Hard Part — Here's Why

It is worth being specific about why causal inference is technically difficult, because the difficulty explains both why this capability is genuinely valuable and why any vendor claiming to offer it needs careful scrutiny.

Correlation between two variables — say, users who adopted Feature X retain at higher rates than those who did not — is easy to measure and nearly useless as a causal claim. Users who adopt Feature X are probably already more engaged than users who do not, and higher engagement predicts better retention independently of Feature X. The correlation overstates the feature's actual causal effect on retention. If a product team acts on that correlation as if it were causal, they may optimize for Feature X adoption among users who would have retained anyway, wasting engineering and growth resources on a non-problem.

Identifying genuine causal effects requires either randomized controlled experiments — where the random assignment breaks the correlation between user characteristics and feature exposure — or observational causal methods that explicitly model and adjust for confounding variables. The latter, which is what Scout applies in non-experimental settings, includes methods such as difference-in-differences, instrumental variables, synthetic control, and various regression discontinuity designs. These methods are well-established in economics and epidemiology but are substantially more complex to implement correctly than the correlation analysis that most BI tools perform.

The failure mode for causal inference systems is not null results — it is incorrect results stated confidently. A BI tool that says "users in segment A retained at 8 points higher than users in segment B" is making a measurable, verifiable descriptive claim. A causal inference system that says "Feature X caused a 4-point retention improvement" is making an inference that requires validation. If the inference is wrong, teams act on false information — potentially worse than no information at all.

This is the bar that any vendor claiming causal inference capability needs to meet, and it is a bar that the Loops acquisition cannot fully validate from the outside. Salesforce is betting that Loops' causal methods are sound enough to deliver correct explanations in enterprise production environments. The integration into Tableau's data infrastructure will be the test.

Why Tableau, Not Agentforce

The strategic decision to integrate Loops into Tableau rather than into Agentforce — Salesforce's AI agent platform that crossed $1.2 billion ARR in Q1 FY27 — deserves analysis. On the surface, Agentforce seems like the natural home for an analytics agent. The product is explicitly an AI agent platform; Scout is an AI agent; the integration would be straightforward from a product positioning perspective.

Salesforce chose Tableau for reasons of data gravity. Tableau is where enterprise analytics data already lives. The visualizations, the data sources, the calculated fields, the metrics definitions, the dashboards — all of that infrastructure exists in Tableau deployments. Integrating causal analytics into the platform where the data already lives is categorically easier than asking organizations to connect their analytics infrastructure to a separate platform.

There is also a differentiation argument. Tableau has been under competitive pressure from Snowflake, Databricks, Microsoft Fabric, and Google Looker — all of which have been building analytics capabilities that increasingly overlap with Tableau's core visualization and reporting functions. Tableau Pulse was the first meaningful product response: proactive monitoring in place of passive visualization. Loops adds the next layer — explanatory analytics on top of the monitoring — and gives Tableau a capability that none of its direct competitors currently offers in a production-ready commercial product.

The Anthropic-Salesforce CRM integration also plays into this. Salesforce's CRM data is the most complete record of customer behavior available to most enterprise sales and success teams. Tableau already connects to that CRM data. A causal analytics layer operating on CRM-native KPIs — contract renewal rates, expansion revenue, support ticket volume, customer health scores — creates an analytics capability that no standalone product can replicate without the CRM data context.

Tableau's $15.7 billion price tag has been criticized as a difficult acquisition to justify given the subsequent rise of cloud-native analytics platforms. The Loops integration is part of Salesforce's answer to that criticism: turning Tableau from a visualization tool into an explanatory analytics platform that no competitor can match because no competitor has simultaneous access to the CRM data, the analytics platform data, and the causal inference engine to explain relationships between them.

The Agentic Analytics Race: Every Data Platform Is Building This Layer

The Salesforce-Loops deal does not exist in isolation. Every major data platform is building toward the same capability layer — explanatory and prescriptive analytics on top of monitoring.

Snowflake Cortex Analyst is Snowflake's natural language query interface for warehouse data. It answers questions about data in plain English but operates primarily in the descriptive layer — it retrieves and aggregates data in response to queries rather than proactively diagnosing causal relationships in metric movements. Snowflake has not shipped a production causal inference layer as of October 2026.

Databricks AI/BI combines Databricks' unified data lakehouse with Genie, an AI assistant that can query data and produce visualizations. Genie's root cause analysis capabilities are at an early stage; the product primarily helps analysts run queries faster rather than autonomously diagnosing why metrics moved.

Microsoft Fabric Copilot integrates with Power BI and other Fabric components to provide AI-assisted analytics. Copilot's current analytical capabilities focus on generating insights from data and producing visualizations on demand rather than autonomous causal investigation.

Google Looker with Gemini adds Gemini's language model capabilities to Looker's semantic layer, enabling natural language exploration of data. The causal inference layer is not a production feature.

What this competitive survey suggests is that Loops has a narrow but meaningful head start in causal inference at the commercial product level. Tableau's integration of that technology puts Salesforce ahead of the immediate competitive set in explanatory analytics capability. The lead is not permanent — Snowflake, Databricks, and Microsoft have the engineering resources to build or acquire equivalent capability within 12–24 months. But the head start matters for enterprise sales cycles, where platform decisions have long tails.

What Product and Retention Teams Should Actually Do With This

The practical implication for product and retention teams is not to wait for the Tableau integration roadmap. The market signal from the Salesforce-Loops acquisition is that causal analytics at the KPI monitoring layer is now a standard enterprise expectation, not a differentiated capability. Teams should audit their current analytics practice against this expectation immediately.

The audit question is simple: When a retention or activation metric moves materially, how long does it take your team to produce a credible root cause explanation? And what percentage of anomalies are explained versus rolling past uninvestigated?

The answer most teams will find is that investigation time runs from several hours to several days, and that the majority of metric movements are never formally explained. That gap is the opportunity that Scout and systems like it are designed to close.

Estimates based on published benchmarks from analytics teams and product organizations; your mileage will vary by data infrastructure maturity.

The PLG-to-enterprise handoff creates a specific context where causal analytics is highest value: when a product-led growth motion is transitioning to sales-assisted, the activation and retention patterns of different customer segments become critical inputs to both the sales qualification model and the product roadmap. Teams that can explain why certain segments activate poorly — by channel, by company size, by onboarding flow variant — can address the root causes rather than applying broad-based product changes that may fix the problem for some segments while creating new problems for others.

The three most immediate applications for causal analytics in a product and retention context are:

1. Activation diagnostics. When activation rates drop, causal analytics should tell you whether the drop is driven by a specific step in the onboarding flow, a cohort characteristic (acquisition channel, company size, user role), or an external factor (time of week, product version). This replaces hours of funnel analysis with an automated explanation that product teams can act on immediately.

2. Churn signal investigation. When churn rate increases, the question is never simply "churn went up." The question is which segment churned, whether the segment characteristics predict that their churn was avoidable, and whether there was a detectable product or service trigger. Causal analytics applied to churn data can differentiate between structural churn (users who were going to leave regardless) and addressable churn (users who churned because of something the product team can fix).

3. Feature impact measurement. For teams shipping features without controlled experiments — either because traffic volume doesn't support A/B testing or because the feature was shipped to all users — observational causal methods can estimate the feature's impact on downstream metrics. This is not a substitute for experimental rigor when experiments are feasible, but it provides signal in the large portion of product decisions that happen without formal experiments.

The Open Questions Salesforce Hasn't Answered

Futurum Group's analysis of the acquisition characterized the deal positively as a strategic move to extend Tableau's analytics lead. Beri.net's more critical take pointed to several questions the announcement left unanswered:

Existing Loops customers without Tableau. Loops had its own customer base before the acquisition. Those customers are getting a product that is being wound down as a standalone offering and integrated into a platform they may not be running. Salesforce's announcement said nothing about migration paths or continued support for non-Tableau Loops users.

Pricing model. Will Scout's capabilities become a standard Tableau feature available to all subscribers, or will they be gated behind an add-on SKU? The answer materially affects the adoption trajectory. Enterprise BI platform features that require separate purchasing decisions often take multiple budget cycles to reach broad deployment even when the underlying technology is compelling.

Integration timeline. The deal closes in early November 2026. When will Scout's root cause analysis capabilities be generally available inside Tableau? Salesforce has not announced a roadmap. Tuck-in acquisitions often take 12–24 months to reach full product integration.

Rationalization with Tableau Pulse. As ISG's David Menninger noted in SalesforceBen's coverage, Salesforce will need to rationalize Tableau Pulse's existing monitoring and insight capabilities with the overlapping capabilities Loops brings. The risk is user confusion — two overlapping monitoring and insight systems in the same platform create complexity that enterprise administrators have to manage. The integration roadmap needs to produce a single coherent product experience, not two parallel feature sets.

These open questions are not reasons to dismiss the acquisition's strategic significance. They are the normal uncertainties of a tuck-in deal. But they are the right questions for enterprise teams to ask in vendor conversations before making analytics platform decisions based on Loops capabilities that are not yet in production.

Takeaway

Takeaway: The Salesforce-Loops acquisition is the clearest enterprise validation yet that the analytics market is moving from descriptive to explanatory as its competitive table stakes. Dashboards that show what happened are no longer sufficient. Enterprise buyers — and particularly the product, growth, and retention teams that live in activation and churn metrics — are demanding systems that explain why metrics moved and what to do about it. Loops' Scout agent is the first commercial product integrated into a major enterprise analytics platform that attempts to deliver causal explanations at scale. The ~$30 million price tag understates the strategic value: Salesforce is buying the capability that Tableau has been missing since its $15.7 billion acquisition, and it is buying a head start in the causal analytics race that every data platform will be running in 2027. For enterprise teams: audit your current root cause investigation process, ask your BI vendor when they plan to ship causal analytics, and treat any vendor claiming this capability as requiring the same scrutiny you would apply to any statistical method — correlation-based explanations dressed as causal inference are worse than no explanation at all.