, I’ve the chance to strive new AI-powered analytical instruments, together with Microsoft Cloth’s information agent. That’s why I need to share what I’ve realized, clarify what a knowledge agent is, and spotlight the distinction between it and a “customary” AI agent.
So, with out additional ado, right here is my definition of a knowledge agent:
A knowledge agent is a report you possibly can discuss to.
For these of us in analytics, this implies two long-held needs may lastly turn out to be a actuality:
#1: Analysts spend manner much less time constructing visualisations.
#2: Self-service insights come nearer to enterprise customers.
Let me elaborate on every of those factors a bit extra.
Fewer visualisations, not fewer insights
I actually get pleasure from a very good report that may inform me “what’s up” with the metrics I’m presently involved in. However being skilled in analytics, I understand how stories can generally forged metrics within the improper gentle, main enterprise customers to incessantly ask analysts for KPIs interpretation, normally 10 minutes earlier than necessary conferences.
And that’s one of many causes we frequently find yourself in a vicious cycle of getting dashboards nobody is utilizing, and stakeholders continuously wanting “the quantity” served advert hoc or by way of spreadsheets.
On the brilliant facet, visualisations and spreadsheets aren’t going wherever, however serving the insights has a brand new manner with a Cloth information agent.
As an alternative of wrapping queries in graphs, you possibly can wrap them in prompts and directions paired with the consumption-ready ruled information property in Cloth, i.e., in a lakehouse, warehouse, Energy BI semantic fashions, KQL database, and even an ontology. This suggests the underlying information nonetheless must be ready and modelled to reply enterprise questions similar to “What was the income this week in comparison with final week?”
Nonetheless, from a design perspective, fairly than making a scoped visible report back to reply this enterprise query, you now create a scoped information agent to offer this, and different subsets of solutions derived from the underlying information mannequin(s).
Extra exactly, the input-output stream goes as follows:
(1) a stakeholder asks a query, (2) the agent, powered by Azure OpenAI Assistant API, interprets the query and “decides” which of knowledge sources is almost definitely to have the reply primarily based on supply schemas and agent directions, (3) generates the suitable question (SQL, DAX, or KQL relying on the supply sort), (4) validates it, (5) executes it below the stakeholder’s credentials, and (6) returns the outcome as a textual content or a desk, not (but) as a visible.
In sum, a stakeholder interplay with insights by way of the information agent is a Q&A session on prime of the curated dataset, and drill-down visuals will be changed with follow-up questions, similar to “Are you able to additionally break the income out by section?”
With that, it’s clear how analysts’ work now not must be re-expressed solely by way of dashboards, aka the long-known tangible proof that the work of capturing the enterprise logic inside information fashions was delivered.
Now, let’s speak about…
Self-service insights, nearer to the place enterprise customers “dwell”
I discussed earlier than that stories can generally misrepresent metrics, however that’s not the one purpose why “If you build it, they will come” not often works for them or analytics basically. The reality is, the information barrier is commonly too excessive to know the underlying semantic fashions and use BI instruments to create visuals on prime.
Though this factors to information literacy, which is a change-management downside, it’s a proven fact that the focused enterprise viewers, who must be report shoppers, usually has an excessive amount of on their plate to trouble studying BI instruments for self-service analytics.
That’s why it’s necessary to carry insights nearer to the place finish customers “dwell”, which these days factors in direction of AI-powered instruments like M365 Copilot.
With the likelihood to show insights by way of information brokers exterior of Cloth, analysts can now deal with the analytical logic behind self-service information brokers, and finish customers can entry insights in the identical AI-powered instruments that help their different day by day duties, with out the complexity of switching to a different platform.
I’ve to notice this isn’t the one method to combine Cloth information brokers within the workflows, and no matter whether or not you’re a developer or a client, it’s good to know…
The distinction between information and an AI agent
We’ve realized thus far that the Cloth information agent is an analytical agent targeted on read-only, ruled information entry, able to translating pure language prompts into complicated database queries that unlock insights, even exterior the Cloth tenant.
On the opposite facet, an AI agent is outlined as a system that permits Massive Language Fashions (LLMs) to do issues, not simply reply to prompts, on behalf of customers or different methods by accessing instruments and information.
That means, the entire magic is within the AI agent setup, the place you need to use a Cloth information agent as a specialised device or information supply.
I’ll illustrate this with one easy instance.
Think about an authorised consumer requests the AI agent to “Draft an electronic mail to the workforce summarising final week’s income by section.” To get this work carried out, the AI agent would, amongst different issues, want to arrange income insights from the enterprise database. So, in an purpose to scale back errors in income calculation, the developer would design an agentic workflow to route the enter immediate to the Cloth information agent device, which might deal with the heavy lifting of figuring out the schema, writing the question, executing it, and returning the exact figures. Lastly, the AI agent would then use these figures to complete its broader workflow and write the e-mail.
What’s the distinction between these two, then? It’s that an AI agent acts, whereas the information agent grounds.
Thanks for studying.
Should you discovered this put up helpful, be happy to share it together with your community. 👏
Need to be taught extra about information brokers?
If that’s the case, take a look at the next assets:
Fabric data agent creation – Microsoft Fabric
Learn how to create a Fabric data agent that can answer questions about data.learn.microsoft.com
Implement Microsoft Fabric Data Agents – Training
Implement Microsoft Fabric Data Agents (chat with your data)learn.microsoft.com
