# SIFT Report & Suggestions: Setup Guide

Paste everything below the line into a new Amazon Quick Desktop conversation. It will walk you through setup and your first run.

---

Hi Quick — I want to set up the **SIFT Report & Suggestions** skill so I can pull existing SIFT entries for my customers and get AI-generated suggestions for new ones. Walk me through this step by step.

## Prerequisites (~5 minutes)

Before we start, I need these connections active. For each one, check if it's connected and show me proof it's working:

1. **AWSentral MCP** — This is the backbone. The skill searches SIFT entries, account data, and opportunities through AWSentral. Verify by searching for one of my accounts.

2. **Outlook** — The skill searches my email for customer activity signals (escalations, QBR follow-ups, executive outreach, launch updates). Verify by showing a recent email.

3. **Slack** — The skill searches Slack channels for customer mentions, wins, blockers, and risk signals. Verify by showing a recent message.

4. **Local Folders** — The skill searches local files for meeting notes, prep docs, and activity logs. Make sure my working folders are added in Settings → Capabilities → Local Folders.

If any connection is missing, walk me through enabling it in Settings → Capabilities → Connections.

## First Run (~5 minutes)

Now let's do a real run. I'll give you a customer name:

> **"Pull SIFT for [my customer name]"**

Here's what should happen:

### What the skill does:
1. **Validates connectivity** — quick check that AWSentral, Outlook, and Slack are reachable
2. **Pulls account context** — searches AWSentral for the account + Knowledge Graph for relationship context + local files for notes
3. **Pulls existing SIFTs** — searches the past 3 weeks of SIFT entries (auto-extends to 6 weeks if empty). Also analyzes SIFTs submitted by other account team members (AM, SA, TAM) and flags topics the CSM could write about from the CSM perspective.
4. **Deep searches all sources** — spawns parallel searches across email, Slack, SFDC opportunities, and local files for recent customer activity
5. **Generates ranked suggestions** — produces SIFT ideas ranked by impact, each with a category (Highlight, Lowlight, Risk, etc.), evidence source, and linked opportunity. Highlights the **Top 2-3 Picks** with rationale for why they're the strongest.
6. **Identifies action gaps** — for topics that are SIFT-worthy but where CSM involvement isn't demonstrated enough yet, suggests specific actions you can take (schedule a follow-up, send a coordination email, set up tracking) to make the topic eligible for a CSM SIFT
7. **Presents suggestions for review** — shows me a summary report and lets me pick which suggestions to draft
8. **Drafts selected entries** — writes full SIFT entries using the "What, So What, Now What" framework with CSM Voice
9. **Fact-checks each draft** — tags every claim as [VERIFIED ✓], [PARTIALLY VERIFIED ~], or [UNVERIFIED ?] and adds a Fact-Check Summary

### What I should see:
- A markdown report opened in a session tab with existing SIFTs indexed at the top
- A decision card asking which suggestions I want drafted
- Full SIFT drafts with inline confidence tags and submission reminders at the bottom
- A **🎯 Top Picks** section highlighting the strongest suggestions with rationale
- An **⚡ Action Required Before SIFT** section with coaching on how to build eligibility for thin-evidence topics
- A **📋 Recommended Portfolio** — curated picks of exactly 1 per category for a balanced submission
- After drafting: a **scoring preview** rating each draft on Specificity/Impact/Actionability/CSM Voice, plus **recommended tags** and an offer to **iterate on drafts or help enter them into SIFT**

## Full Workflow

**Trigger:** "Pull SIFT for [customer name]"

**Before:** The skill researches the customer across AWSentral (SIFT history, opportunities, account metrics, spend, activity), Outlook email, Slack channels, local files, and the Knowledge Graph. Generates ranked suggestions for new SIFT entries based on real recent activity — never speculative or forward-looking. Each suggestion includes the category, a title following `[CSM] Customer + Action + Service` format, supporting evidence, and a linked opportunity.

**Interactive review:** Presents suggestions as a selectable list — you pick which ones to draft. No bulk dump.

**Recommended Portfolio:** Curates the single strongest suggestion from each category (Highlight, Lowlight, Observation, Risk, Challenge, Blocker) — max 6, typically 3-4. These are marked with ⭐ in the selection card. The full ranked list is still available if you want to browse everything.

**Drafting:** Writes each selected entry following the "What, So What, Now What" structure. Enforces CSM Voice Rule (CSM as connective tissue, no "I", no bullets, ≤150 words). Automatically trims if over the word limit.

**Before presenting drafts:** Matches the CSM's own writing style from past entries for consistency. Rates each draft on Specificity, Impact, Actionability, and CSM Voice — flags weak areas with specific fix suggestions. Recommends tags (team, region, AWS service, vertical) per draft.

**Fact-check:** Cross-references every claim against data sources. Tags each as Verified, Partially Verified, or Unverified. Adds a per-draft Fact-Check Summary noting what was confirmed, which sources were used, and what needs your review.

**Output:** Structured markdown report with existing SIFT index, suggestion details, full drafts, fact-check summaries, and submission reminders with direct SIFT link.

**After output:** Offers to iterate on any draft (adjust tone, strengthen CSM voice, rework sections) and to help enter selected entries into SIFT via the browser.

## Batch Mode

Say "pull SIFT for Love's, Compass, SGWS" to process multiple customers in parallel. Each customer gets its own report file, and you get a consolidated summary showing per-customer SIFT counts, suggestion counts, and top picks. Drill into any customer to draft entries.

## Customization

The skill has sensible defaults but you can adjust:
- **Lookback window** — defaults to 3 weeks, auto-extends to 6 if no results. Tell the skill if you want a different range.
- **Categories** — the skill suggests categories (Highlight, Lowlight, Observation, Risk, Challenge, Blocker) but you can override when reviewing.
- **Opportunity linking** — auto-links to active opps from AWSentral. If no opps exist, it suggests account-level attachment.

## Tips

- Run it **before your SIFT submission cadence** — weekly or bi-weekly depending on your team's rhythm
- The fact-check step adds time but saves you from submitting inaccurate entries — review the [UNVERIFIED ?] tags carefully
- If the customer name doesn't match exactly in AWSentral, the skill will try fuzzy matches and ask you to pick
- You can run it for multiple customers in the same session — just say "pull SIFT for [next customer]" after reviewing the first

## Troubleshooting

| Issue | Fix |
|-------|-----|
| "No SIFTs found" | The skill auto-extends to 6 weeks. If still empty, this customer may not have prior SIFT entries — the suggestions section will still generate. |
| AWSentral not connecting | Check Settings → Capabilities → Connections. AWSentral MCP must be enabled and authenticated. |
| Suggestions seem thin | The skill only generates suggestions grounded in real data. If email/Slack activity is low for that customer, there may genuinely be fewer signals. |
| Common customer name returns wrong account | The skill presents fuzzy matches as options — pick the right one from the list. |
| Fact-check is slow | It cross-references each claim individually. For drafts with many specific claims, this can take a minute. It's worth the wait. |

## Resources

- Access SIFT: https://aws-crm.lightning.force.com/lightning/n/Sales_Insights_Field_Trends
- SIFT Getting Started Guide: https://amazon.enterprise.slack.com/docs/T01698U3K1U/F08LQ9S448L
- Tips for Better AI-Generated SIFT Insights: https://amazon.awsapps.com/workdocs-amazon/index.html#/document/1e379993bc5b221576b2e2244c0000a145772ddc63bbfb013aea92ba8c904565
- Slack: #awsentral-sift