---
title: Client Sentiment Analysis Tool — Google Workspace Integration
type: article
created: '2026-02-18'
updated: '2026-02-18'
source_docs:
- raw/2026-02-18-mid-week-call-w-123489147.md
tags:
- ai-tools
- client-health
- google-workspace
- vector-database
- operations
layer: 2
client_source: null
industry_context: null
transferable: true
---

# Client Sentiment Analysis Tool — Google Workspace Integration

## Overview

An internally built AI tool that ingests all company communications via the Google Workspace API and uses a vector database to provide real-time client sentiment scoring and surface critical operational issues. The tool was demoed internally on 2026-02-18 and access is being rolled out to the broader team.

First discussed and demonstrated in [[meetings/2026-02-18-mid-week-call-asymmetric-internal]].

---

## How It Works

The tool connects to Google Workspace and pulls every communication artifact associated with a client:

- **Email** (all company inboxes)
- **Slack messages**
- **Call transcripts** (via Fathom recordings)
- **Google Drive documents**

All of this data is embedded into a **vector database**, enabling both structured filtering and semantic search across the full communication history. The system then runs analysis to produce:

1. **Sentiment scoring** — overall tone and trend across the relationship
2. **Critical issue flags** — operational problems surfaced from within the communication record
3. **Activity statistics** — volume of touchpoints over time, with date-range filtering
4. **Tone trend monitoring** — tracks changes in client responsiveness, professionalism, and engagement patterns

> "It's basically rooting through all your stuff and saying, here's what's going on." — Mark Hope

---

## Key Capabilities

### Sentiment Analysis
- Reads every document, email, and call transcript associated with a client
- Produces an overall sentiment label (e.g., Positive, Stable, Declining)
- Tracks tone trend over a configurable date range (e.g., 30 days, 90 days)
- Monitors response latency — flags if a client starts taking longer to reply

### Critical Issue Detection
- Surfaces operational problems mentioned or implied across communications
- Examples from the 2026-02-18 demo:
  - **Citrus America:** 32 inactive trade show geofencing campaigns still enabled and wasting budget; Google Ads remarketing inefficiency; conversion action misconfiguration
  - **Adavacare:** Task delivery delays; 3 of 10 locations missing pricing data; photo asset verification issues

### Search & Query Interface
- **Structured filters:** filter by client, person, content type (email, Drive doc, Slack), and date range
- **Semantic search:** natural language queries like "tell me about budget issues" or "project status"
- **AI Q&A:** ask freeform questions (e.g., "When was the last time we changed the budget?") and receive answers grounded in the actual communication record

### Activity Dashboard
- Per-client statistics card showing total email, Slack, Drive, and call volume
- Activity timeline to identify periods of high/low engagement
- Useful for spotting relationship drift before it becomes a churn signal

---

## Important Limitations

- **Sentiment reflects stated tone only** — if a client is being politely positive while privately dissatisfied, the tool will not detect it
- **Not a campaign analytics tool** — it does not analyze ad performance numbers, CRM data, or billing figures; it only processes communication content
- **Complementary to, not a replacement for, X-Ray** — X-Ray handles structured performance data; this tool handles unstructured communication data
- **Still in active development** — the AI Q&A ("Ask") feature is noted as not yet fully reliable; onboarding flow had minor login issues at launch

---

## Access & Rollout

- Built by Mark Hope
- Access being extended to Melissa Cusumano and others as of 2026-02-18
- Login is via invite; password reset flow had minor issues at initial rollout (Mark is aware and monitoring)
- Contact Mark to request access or report issues

---

## Example Outputs

### Citrus America (run: ~2026-02-18)
| Dimension | Result |
|---|---|
| Overall Sentiment | Positive and professional |
| Tone Trend (30-day) | Stable, consistently positive |
| Responsiveness | No declining patterns |
| Critical Issues | 3 flagged (geofencing waste, remarketing inefficiency, conversion misconfiguration) |

### Adavacare (run: ~2026-02-18)
| Dimension | Result |
|---|---|
| Overall Sentiment | Positive and stable |
| Tone Trend | No deterioration detected |
| Urgent Issues | 4 flagged (task delays, missing location pricing, photo asset verification, messaging adjustment needed for Fardale) |

---

## Strategic Value

The tool addresses a specific blind spot: **clients can appear healthy in calls and emails while operational problems accumulate underneath**. The Adavacare example is illustrative — the client's communication tone was positive and appreciative, yet the tool surfaced four urgent delivery issues that required attention.

This makes it particularly useful for:
- Pre-call preparation (know what's broken before the client brings it up)
- Account health monitoring at scale across the full client roster
- Early churn detection via tone trend and responsiveness monitoring
- Internal accountability (task delays become visible in the communication record)

---

## Related

- [[meetings/2026-02-18-mid-week-call-asymmetric-internal]]
- [[clients/citrus-america/_index]]
- [[clients/adavacare/_index]]
- [[knowledge/ai-tools/vector-database-architecture]]