Skip to content
Agile Defense

Curriculum / Foundry Foundations

ADVANCEDUnit 7day-07

AIP Analyst Investigation

not started~360 min
Use AIP Analyst (now GA) to investigate supplier exposure in natural language and produce a cited brief.

Introduction

Scenario: Leadership asks which suppliers drive the highest projected inventory exposure this week, and why.

Day 7 puts AIP Analyst at the center of the disruption-response problem. Leadership has asked a pointed question: which suppliers drive the highest projected inventory exposure this week, and why? AIP Analyst is generally available in Foundry and lets you use natural language to perform ad-hoc analyses across your Ontology. Rather than hand-building a workflow, you pose the question and the analyst answers by autonomously searching your Ontology, creating object sets, and transforming data before generating summaries and visualizations. Today you learn to drive that investigation deliberately so the answer is trustworthy, not just fast.

This matters because disruption response lives or dies on speed plus defensibility. AIP Analyst can discover the right object types, apply filters and search-arounds across object sets, and run aggregations (count, sum, average, min, max, percentile) or even SQL against object sets and datasets to quantify exposure. Because it shows its analytical process as visual workflow steps, you can see exactly which object type search ran and which filter was applied, turning an opaque answer into an auditable query trail you can hand to leadership.

Before any of this works, AIP must be enabled for your enrollment, and the right model families accepted in Control Panel, so part of today is understanding those guardrails. You will run the investigation inside a Workshop module using the AIP Analyst widget, scoped to your supply-chain Ontology, and then extract the analyst's object set, charts, and summary into Workshop variables so the findings become a reusable, shareable brief rather than a throwaway chat.

Capability focus: AIP Analyst — ontology-first natural-language analysis; Workshop widget · Artifact: Saved investigation brief

Key concepts

  • AIP Analyst: a generally available agent interface that uses natural language to run ad-hoc analyses across your Ontology, autonomously searching object types, building object sets, transforming data, and generating summaries and visualizations.
  • Object sets and search-arounds: the analyst discovers data via object type and dataset search, then narrows it with filters and search-arounds (e.g., from a supplier to its affected SKUs and open orders) to assemble the exact population it reasons over.
  • Aggregations and SQL: exposure is quantified with count, sum, average, min, max, and percentile aggregations, or SQL queries against object sets and datasets when the calculation is more complex.
  • Query trail / reasoning steps: the interface exposes its work as visual workflow steps and an outline sidebar, so you can inspect which object type search ran and which filter was applied before trusting the answer.
  • Enabling AIP: AIP is on by default for new enrollments, but model families must be enabled and their terms accepted in Control Panel > AIP settings, with access scoped to specific groups or organizations.
  • Output reuse in Workshop: the AIP Analyst widget can extract the most recent object set, Vega chart specs, and message history into Workshop variables, and findings export as PDF for downstream widgets and reporting.

Companion video

Speedrun: Your First AIP Workflow · open on YouTube

Hands-on activity

each step validates · the unit completes when all steps pass
  1. 1

    Formulate investigation questions

    Translate leadership's request into precise, answerable prompts the analyst can act on, because AIP Analyst answers natural-language questions by searching your Ontology and may ask clarifying questions when a query is ambiguous. Start broad ("Which suppliers have the highest projected inventory exposure this week?") then prepare follow-ups that force search-arounds ("For the top suppliers, which SKUs and open purchase orders drive that exposure?"). Use the widget's suggested prompts that appear when the chat is empty as a starting scaffold, and confirm the module is scoped to your supply-chain Ontology so the agent searches the right object types.

    not startedself-attested

    Self-attested: questions are specific and operational.

  2. 2

    Quantify total inventory exposure

    Have the analyst build the supplier-to-SKU-to-order object set with search-arounds, then quantify exposure using aggregations such as sum and percentile across projected on-hand and at-risk inventory values. Where the math is layered (for example, weighting exposure by lead-time risk), prompt it to run a SQL query against the object set or underlying datasets rather than a simple count. Inspect the resulting Vega chart and text summary to confirm the ranked list of suppliers and the dollar (or unit) exposure each one drives this week.

    not startedinstance check

    Confirms the exposure figure your brief cites reproduces via an aggregation (works on the non-AIP Object Explorer fallback too).

  3. 3

    Capture the query trail

    Open the outline sidebar and step through the visual workflow the analyst produced, verifying which object type search located the suppliers and exactly which filters and search-arounds shaped the object set before any aggregation ran. This is the "why" leadership asked for: a defensible chain from raw Ontology objects to the exposure figure. Because AIP Analyst does not retain conversation history after a session closes, extract the most recent object set, the chart Vega specs, and the message history into Workshop variables now so the trail survives the session.

    not startedself-attested

    Self-attested: the brief preserves queries, tables, and charts.

  4. 4

    Write the investigation brief

    Assemble the captured outputs into a concise brief that answers both halves of leadership's question: the ranked suppliers by projected exposure, and the drivers behind each (specific SKUs, late orders, lead-time risk) pulled from the query trail. Use the widget's PDF export and the extracted Workshop variables to feed the ranked object set and charts into downstream widgets so the brief is reusable, not a one-off chat. State the data scope and any clarifying assumptions the analyst surfaced so reviewers can judge confidence at a glance.

    not startedself-attested

    Self-attested: the brief separates evidence, interpretation, and recommended action.