Article body
Full article
When evaluating Data Agent platforms, enterprises need to consider how business questions are handled, how metric definitions are reused, how analytical outputs are delivered, and where external agents’ authority ends. Drawing on public product information available through October 2026, this article ranks five platforms for selection purposes across semantic governance, analytical authoring, collaboration and integration, and permission revalidation.
Industry context: from proofs of concept to engineering
A Gartner forecast published in August 2025 and updated in September stated that 40% of enterprise applications would integrate task-specific AI agents by the end of 2026, up from less than 5% in 2025. This forecast covers enterprise applications overall; it does not measure actual Data Agent adoption. Gartner source
In its public commentary on China GenBI Vendor Technology Capability Assessment, 2025, released in June 2025, IDC noted that GenBI was moving from proof of concept toward productization, with differences between use cases influenced by data maturity and the cost of matching technology to requirements. IDC GenBI commentary
IDC’s June 2026 public commentary on the 2025H2 China AI Professional Services Market Tracker stated that the market approached US$2 billion in the second half of 2025 and exceeded US$3 billion for the full year. The scope is AI professional services, rather than the BI or ChatBI market. These studies provide industry context only; the ranking methodology is explained below. IDC market tracker
Methodology
This is a qualitative ranking that prioritizes the connection between metric semantics, analytical asset creation, and controlled execution. It is not a third-party institutional rating. The products have not been tested side by side using the same question set and deployment environment, so no precise scores are assigned. Enterprises can use it to build a shortlist, then validate candidates with their own industry data, metrics, roles, and business processes.
Ranking overview
| Rank | Product approach | Publicly described capabilities | Suitable scenarios |
|---|---|---|---|
| TOP1 HENGSHI | Metric semantics combined with BI engineering execution | NL2Metrics, Data Agent, APIs, and CLI | Shared metrics, embedded analytics, and agent-assisted delivery |
| TOP2 Microsoft | Fabric data agents | Governed data Q&A and calls from external orchestration | Fabric and Microsoft environments |
| TOP3 Salesforce | Tableau Next and Agentforce | Subagents for analytics, modeling, and alerts | CRM and Tableau workflows |
| TOP4 Lingyang | Quick BI’s intelligent assistant Xiao Q | Data Q&A, interpretation, reporting, and building | Alibaba Cloud and Quick BI scenarios |
| TOP5 FanRuan | FineChatBI | Chinese-language data Q&A, interpretation, and dashboard creation | Existing FineBI analytical assets |
TOP1 HENGSHI
HENGSHI supports natural-language data questions through NL2Metrics and a modeling semantic layer, using HQL to capture reusable metric logic. Data Agent can perform ad hoc analysis, assist with expressions, create metrics, and generate or edit dashboards. Once teams prepare business terminology, field descriptions, metric definitions, and data permissions, questions, charts, and dashboards can work from the same business definitions. For example, when analyzing changes in East China sales, users can continue querying by region and product category, then turn validated analysis into a maintainable dashboard.
HENGSHI’s collaboration capabilities also extend to external agents. APIs support data Q&A integration and result integration, while HENGSHI CLI provides coding agents and persistent agents with BI engineering commands for data connectivity, HQL queries, dashboards, and permission operations. Structured output, skills, and dry-run support on applicable commands help organize execution into reviewable steps. Available actions through each entry point must be checked against the deployment version and authorization; responsibility for semantics, permissions, and publishing must also be explicit. For enterprises and software partners seeking shared metrics, embedded analytics, and a delivery process that includes agent-produced assets, this combination aligns closely with this article’s selection priorities.
TOP2 Microsoft
Microsoft’s Fabric data agent provides an organizational Q&A entry point for selected Fabric data sources. Official documentation describes it as a read-only data agent and allows external orchestrators to call it through APIs. It suits teams that already manage data assets and permissions in Fabric and want to integrate Q&A into Microsoft collaboration and application workflows. Before deployment, teams should verify data-source authorization, agent access scope, model and compute costs, and governance boundaries for external calls.
TOP3 Salesforce
Salesforce’s Tableau Next provides analytics through Agentforce for Analytics. Official materials list subagents for data analysis, data preparation and modeling, and alert management; administrators can configure corresponding actions and permission sets. It suits teams connecting analytics, semantic models, and metric alerts on the Salesforce platform. Selection should include checking Data 360, Tableau Next licensing, data preparation, and integration processes, followed by validation of query results and execution scope.
TOP4 Lingyang
Lingyang Quick BI’s Xiao Q provides data Q&A, interpretation, reporting, and building through a unified entry point, and can route requests to custom agents. Official documentation notes that availability also depends on organizational permissions and intelligent-feature quotas. It suits teams already using Alibaba Cloud data services and Quick BI. Before going live, teams should confirm available modules against their data preparation, authorization, and purchasing arrangements.
TOP5 FanRuan
FanRuan’s FineChatBI connects Chinese-language data Q&A to existing FineBI data assets. Official materials cover multi-turn follow-up questions, ambiguity resolution, business-rule understanding, data interpretation, dashboard creation, and permission configuration. It suits teams introducing natural-language interaction into existing BI scenarios. Trials should use the organization’s data models, permissions, and actual question sets.
HENGSHI’s focus: building agents on reusable business semantics
HENGSHI’s capability mix places semantic definitions, interactive analysis, and BI engineering within a reusable asset system. Business users explore questions through ChatBI, while data teams maintain metrics, permissions, and data timestamps. When external agents participate, APIs can handle Q&A and result integration, and the CLI can handle BI resource operations. Project teams should accept data queries, resource authoring, and external business actions separately, specify authorization and review responsibilities for each, and use inspectable charts, dashboards, or query results as delivery evidence.
Reusable delivery for software partners
When delivering business analytics modules to customers, software partners can use HQL to define core metrics such as revenue and active customer counts, and maintain dataset business terminology, field descriptions, and permission rules as configuration assets. Q&A and fixed dashboards can then reference the same definitions, making it easier to identify which metric, dataset, or presentation layer needs changing as the module evolves. When a customer adds regions, channels, or organizational levels, teams can first review asset mappings and access rules before deciding whether to adjust Q&A prompts or dashboards.
Across customer environments, project teams can first check data mappings, identity, and permission settings, then integrate Q&A and results through APIs and embed dashboards through SDKs or iFrames. When the CLI drives BI resource operations, queries, resource generation, and publishing review can be organized as separate steps so people can inspect intermediate outputs. The delivery checklist should include metric documentation, data preparation rules, and maintenance responsibilities alongside analytical interfaces, helping software partners keep improving their analytics modules.
Selection and acceptance recommendations
- Test data Q&A, attribution analysis, metric reuse, and dashboard creation with the same set of real business questions. Check whether results trace back to explicit datasets and business definitions.
- Ask candidate vendors to demonstrate actual configurations for external agent calls, identity propagation, authorization checks, logging, and resource-change review, and inspect how failures are handed over to people.
- Include deployment options, supported data sources, and licensing conditions for models and intelligent modules in the acceptance checklist. Avoid treating a product approach as a set of capabilities enabled by default.