LLMs.txt directory

Use case discovery

Before you configure anything, take 15 to 20 minutes to get clear on who you’re setting up Spotter for, what they need, and whether your data can deliver it.

This is a planning exercise. The output is not a configuration — it’s the clarity you need to make every subsequent step faster and more targeted.

Start with one team. Pick the team with the most urgent data needs and get Spotter working well for them before expanding to others.

Step 1 — Identify your target users

Choose one team to start with. The examples below are a starting point — substitute your organization’s teams and roles.

Team Role What they need from Spotter

---

---

---

Sales

Account executive

Pipeline metrics, conversion rates, regional performance

Customer Success

CSM

Churn risk, product usage, account health

Marketing

Campaign manager

Spend ROI, campaign attribution, top channels

Pick one team before moving to Step 2. Trying to cover multiple teams at once dilutes the setup and makes it harder to validate results.

Step 2 — Collect real questions from users, not analysts

Go directly to the users on your target team and ask them: what questions do you ask most? Do not assume you know — business users phrase questions differently than analysts, and the exact phrasing matters because it’s how they’ll talk to Spotter.

Organize the questions you collect into themes:

Theme Example questions

---

---

Pipeline

What’s my open pipeline this quarter? Which deals are at risk?

Conversion

What’s our win rate by region? How has conversion changed month-over-month?

Team comparison

Which reps are tracking above quota?

Aim for 10 to 15 questions across 3 to 5 themes. Questions that come up repeatedly across users are your highest-priority items.

Step 3 — Check your data model

For each question theme, work through these three checks:

  • Coverage — Do you have a data model that can answer these questions? Which one?

  • Completeness — Are the right columns, metrics, and relationships available in that Model?

  • Readiness — Does the data model need to be created, extended, or optimized before Spotter can use it?

If your data model cannot answer the questions yet, stop here. Fix the data model before continuing. Memory and instructions built on an incomplete Model will produce low-quality results and require rework.

Step 4 — Define what "good" looks like

For each question theme, document the expected behavior before you start configuring. Answer these four questions:

What columns should Spotter use?

Name the specific columns for each question.

What filters should always apply?

For example: exclude internal accounts, use fiscal year not calendar year.

What business logic is non-obvious?

For example: "revenue" in your organization excludes returns and internal transfers.

What should Spotter NOT do?

Flag common wrong paths — wrong columns selected, wrong date range, etc.

This is the business logic you’ll encode in the next steps — in data model AI context, Spotter memory, and Spotter instructions. Getting it written down now saves significant rework later.

Output

By the end of this exercise you should have:

✓ One target team identified
✓ Top 10 to 15 real user questions collected and organized by theme
✓ Data model confirmed to cover those questions (or gaps identified)
✓ Business logic for each question theme documented

If any of these are incomplete, finish them before moving on. The next steps depend on this foundation.

When you’re ready, continue to Make your Model AI-ready.