Natural language filtering in Dynamics 365 Sales lets a rep type a question — "deals closing this quarter over $100k in healthcare" — and Copilot converts it into a Dataverse query behind the view. It replaces the advanced-find dialogue, not the analyst.

What the modern grid gives you

The Power Apps grid control is the component underneath modern views, and three of its capabilities change how pipeline review actually runs:

  • Inline editing. Correct a close date or an amount in the list, without opening and closing twenty records.
  • Column aggregation. Totals at the foot of the column, so the number you are discussing is on screen.
  • Nested child grids. Expand an opportunity to its lines in place.

Together these remove the main reason pipeline review happens in a spreadsheet export, which is the reason pipeline data is usually a week old.

What the prompt becomes

OpenHealthcareOpportunities.fetchxml
<fetch version="1.0" output-format="xml-platform" mapping="logical" distinct="true">
  <entity name="opportunity">
    <attribute name="name" />
    <attribute name="estimatedvalue" />
    <attribute name="closeprobability" />
    <attribute name="estimatedclosedate" />
    <attribute name="stepname" />
    <order attribute="estimatedvalue" descending="true" />
    <filter type="and">
      <condition attribute="statecode" operator="eq" value="0" /> <!-- Open Opportunities -->
      <condition attribute="estimatedvalue" operator="ge" value="100000" />
      <condition attribute="estimatedclosedate" operator="this-quarter" />
    </filter>
    <link-entity name="account" from="accountid" to="parentaccountid" alias="acc">
      <attribute name="name" />
      <attribute name="industrycode" />
      <filter type="and">
        <condition attribute="industrycode" operator="eq" value="1" /> <!-- Healthcare -->
      </filter>
    </link-entity>
  </entity>
</fetch>

This is the useful mental model: the prompt is a faster way to write that. It is not a different kind of question, and it can only see what Dataverse holds.

Where it stops being reliable

SituationWhat happensWhat to do
The field is not populatedConfidently empty or partial results.Check field completeness before trusting a filter on it.
Ambiguous business termIt picks an interpretation without telling you which.Read the generated query, not just the result count.
The same question, rewordedA slightly different query, and a different number.Save it as a view once it is right.
Cross-entity or time-series analysisBeyond what a view can express.Power BI. This is not the tool for it.
The risk is not a wrong answer. It is a plausible answer to a question slightly different from the one you asked, arriving with no indication that the substitution happened.

Using it well

  1. Let reps explore with prompts — that is where the time saving is real.
  2. Promote anything recurring to a saved view, so the number is stable week to week.
  3. Read the FetchXML for anything that will be quoted in a forecast meeting.
  4. Fix the data quality the prompts expose. That finding is often worth more than the feature.