Quickstart Guide

Quickstart #4: Structured Data Analysis (SDA)

View as Markdown

Background & Objectives

This guide will walk you through how to build Agent Studio plugins that return large volumes of data to the AI Assistant.

In this quickstart guide, you will

  • Build a plugin that fetches 500 accounts from a mock CRM
  • Set up the plugin to work alongside our Code Interpreter plugin

At the end, your AI Assistant will be able to handle requests like this

Prerequisites

Define your plugin

  1. Go to Agent Studio. In the Plugins tab, select Create

  2. Define your plugin.

    📘 SDA Tip!

    It’s important to be descriptive about the data that an SDA plugin returns.

    AttributeValue
    Plugin TitleFetch Accounts
    Plugin DescriptionReturns all Accounts with ID, name, industry, territory, created/modified, last contact & renewal, ARR & MRR, employee count, opportunities (total/open/won/lost), avg deal size, churn rate, NPS, tickets, offices, market share %, growth rate & funding.

Create a trigger

  1. In the Plugin Editor, add a Conversational Trigger and fill it out


Build your process

Connect to the mock CRM

We need an HTTP action that will fetch the large volume of data from the external system.

  1. Click ⊕ in the tab bar and create a new HTTP Action

  2. Select the Import option (next to the Test button)

  3. Import the following API call

    $curl --location 'https://marketplace.moveworks.com/api/purple-suite/crm/accounts?count=500'

  4. Hit Test

  5. Name your HTTP Action. It should look something like this.


  6. Save your HTTP action

Attach a Conversational Process

Now let’s connect the HTTP action to your plugin & prepare it for analysis

  1. In the Plugin Editor, add a Conversational Process & create a new one.
  2. In the Conversational Process Editor add an action-based activity.
  3. Choose the HTTP Action you just built.
  4. In the activity configuration, under Output Mapper, we will rename the fields to make them more friendly for SDA.

    📘 SDA Tip!

    It’s important to name fields succinctly & clearly to maximize the performance of SDA against various queries.

    1MAP():
    2  items: response.accounts
    3  converter:
    4    AccountID: item.Id
    5    AccountName: item.Name
    6    AnnualRecurringRevenue: item.ARR__c
    7    AverageDealSize: item.AverageDealSize__c
    8    ChurnRate: item.ChurnRate__c
    9    ClosedLostOpportunities: item.ClosedLostOpportunities__c
    10    ClosedWonOpportunities: item.ClosedWonOpportunities__c
    11    DateCreated: item.CreatedDate
    12    DateLastModified: item.LastModifiedDate
    13    Employees: item.NumberOfEmployees
    14    FundingAmount: item.FundingAmount__c
    15    GrowthRate: item.GrowthRate__c
    16    Industry: item.Industry
    17    LastContactDate: item.LastContactDate__c
    18    MarketShare: item.MarketSharePct__c
    19    MonthlyRecurringRevenue: item.MRR__c
    20    NetPromoterScore: item.NPS_Score__c
    21    NextRenewalDate: item.NextRenewalDate__c
    22    OfficeCount: item.OfficesCount__c
    23    OpenOpportunities: item.OpenOpportunities__c
    24    SupportTickets: item.SupportTickets__c
    25    Territory: item.Territory__c
    26    TotalOpportunities: item.TotalOpportunities__c

Launch it

Under Audience Settings you can launch it to yourself for testing.

Test it!

Try it out! You should see something like the following.

🚧 Not seeing the right results?

Verify that SDA is running by checking for the Running analysis progress update

Here are some other questions you could try:

  • What’s the total ARR of customers in the Tech industry?
  • Which industry is performing best in Europe?
  • Which accounts have the most support tickets are are up for renewal soon?

Next Steps

Start building your own plugins for Structured Data Analysis!