Asking Business Questions
of Your Own Data
Claude Code + Google Sheets
Claude Cowork for Everyone — Helsinki 2026
Quick intro
Hi, I'm Alejandra.
CCO (Chief Customer Officer) at Groweo · Helsinki
- ·Telecom engineer who fell into customer success ten years ago
- ·Five years at Supermetrics before this — global CS director, team of 27
- ·🇲🇽 Mexican by birth, 🇫🇮 Helsinki by choice
Started using Claude Code in January.
Found the Helsinki community at the meetup last month.
Héctor Tortosa asked if I'd share what I'd built with a business-focused crowd — and here we are. 🤓
The Opportunity
The questions that matter most
are rarely answered by a single dashboard.
The Opportunity
"How much revenue is at risk this quarter?"
"Do we have enough people to maintain the customer relationships we need?"
"Who should my team call first this week?"
These are cross-cutting questions
The data is there. The challenge is synthesizing it fast enough to act on it.
Use Case 1
Retention
Forecast
"Where are we headed? Where's the risk? What should we do about it?"
Use Case 1 — Retention Forecast
> What's our retention forecast looking like? How has it progressed over the last 3 months, where is the most risk coming from, and what should we address to reduce it?
◆ Reading "Billing Report" — Dashboard!A1:AK25
◆ Reading "Churn Notices" — Customer list!A1:AA4000
◆ Reading "Contracts" — All Customers!A1:I6000
◆ Reading "Churn Reasons" — Analysis!A1:AC50
Use Case 1 — Retention Forecast
Claude
● January: 91.2% → ● February: 89.4% → ● March: 87.7%
Trending down 1.7 pts/month. Worst case (all at-risk contracts churn): 54%
| Driver | Customers | ARR Impact | Trend |
| Low product adoption (<1% conversion) | 184 | €1.4M | ↑ growing |
| Failed onboarding (never launched) | 152 | €980K | steady |
| No CS engagement in 90+ days | 287 | €1.2M | ↑ growing |
| Pricing / ROI concerns (exit surveys) | 89 | €620K | ↓ declining |
1. Re-engage the 287 silent accounts — highest save potential before they give notice
2. Launch the 152 stuck onboardings — €980K ARR at risk from never seeing value
3. Run adoption reviews for the 184 low-conversion accounts — fastest-growing driver
It's not limited to Google Sheets
Claude has built-in connectors for the tools you already use —
Code, Chat, and Cowork all use them.
Use Case 2
Portfolio
Performance
"Are the portfolios balanced? Who's overloaded?"
Use Case 2 — Portfolio Performance
> Break down retention by CSM — customers per person, churn rate, and at-risk contracts.
Claude
| CSM | Customers | Churn Rate | At-Risk | Status |
| CSM A | 210 | 4.2% | 12 | Healthy |
| CSM B | 195 | 5.1% | 18 | Healthy |
| CSM C | 390 | 8.7% | 34 | Strained |
| CSM D | 430 | 12.3% | 51 | Overloaded |
| CSM E | 440 | 14.1% | 62 | Overloaded |
Avg churn rate for CSMs under 250 customers: 4.7%
Avg churn rate for CSMs over 350 customers: 11.7%
The leadership layer
This only becomes actionable if you've defined your contact cadence:
Monthly
High-value customers
Quarterly
Mid-tier customers
Automated
Long-tail customers
Without cadence, you're looking at numbers.
With it, you're looking at a staffing decision.
Use Case 3
Capacity Planning
& Hiring Budget
"How many people do we need? What does it cost?"
Use Case 3 — Capacity Planning
> Based on our customer segments — 600 automated, 500 quarterly, 400 monthly, rest in churn-rescue mode — how many CS specialists do we need? What does that cost as % of ARR?
Claude
| Role | People | Covers |
| CS Specialist (full lifecycle) | 7 | Quarterly, monthly & rescue customers |
| Automation / Ops | 1 | Tools & dashboards |
| CS Lead | 1 | Management & escalations |
| Total | 9 | |
At €20M ARR: ~3-4% of revenue
600 customers on automated — zero CS time needed
A business case for your CFO.
A number grounded in your actual data.
When something changes — 50 new customers, one less team member —
just ask again with updated assumptions.
Why Claude Code
Why Code, not Chat or Cowork?
No preview overhead
Reads sheets programmatically. No file cards, no "reading file…" UI. Values in, synthesis out.
Composability
Chain tools: read a sheet → write a CSV → run a script → ask a follow-up. Chat can't do that.
Persistence
Local folders, files, project memory, git. Chat has Projects; Cowork has workspaces. Neither matches a filesystem.
One-off question? Chat. Shared team workflow? Cowork. Solo recurring analysis? Code.
A Fair Question
Could I do this in Cowork?
Yes — Cowork has the same Google Workspace connector. The answers would be similar.
Cowork is better if…
- • Multiple people on your team use the same analysis
- • You don't want to install anything on your computer
Code is better if…
- • You're running this personally, for yourself
- • You want to combine sheet data with local files or scripts
I picked Code because this is solo and recurring.
If your version is team-shared, Cowork is the right answer.
Claude Code
Anthropic's coding agent. It comes as a desktop app — no terminal, no command line needed for daily use.
(There's also a CLI, if you're more comfortable in a terminal.)
Think of it as Claude, but with jazz hands.
Project: customer-ops
Ask me anything about your data…
How to Set This Up
Two ways to connect Google Sheets
The easy path
Built-in Google Workspace connector
Three clicks. Uses your Google login. Right default for most people.
The advanced path
Service account + gws CLI
More setup. File-by-file scope. For personal data or team deployments.
Let's look at each.
Path 1 — The Easy Path
Built-in Google Workspace connector
- Open Claude Code settings → Connectors
- Click "Connect Google Workspace"
- Authorize in Google's consent screen
30 seconds. Uses your personal Google login. The right default for most people.
Claude sees whatever your Google account sees — all of Drive, optionally Gmail and Calendar if you enable those scopes.
Path 2 — The Advanced Path
Service account + gws CLI
When you want more control:
- ●File-by-file scope — Claude only sees sheets you explicitly share
- ●Personal Gmail and Calendar off-limits by construction
- ●Demonstrable data-minimization story for compliance or audit
> Install the Google Workspace CLI and its skills.
Claude
Installed gws via Homebrew and added the gws-sheets and gws-drive skills. Ready to use.
~10 min total. gws is community-maintained, open source, Apache-2.0.
How to Choose
Which path is for you?
Start here
Built-in connector
Personal workflow, your own Google account, data you're comfortable having Claude reach.
Upgrade if…
Service account + gws
You need exact file-by-file scope, your data policy requires it, or you want personal Gmail and Calendar untouchable.
Start simple. Upgrade if you need to.
Responsible Use
Know what your plan allows
- ●A DPA is required to process personal data. Anthropic provides a Data Processing Agreement on Team and Enterprise plans. Personal/Pro: no DPA.
- ●Data minimization still applies. Even with a DPA, only send what you actually need. Service-account scoping helps you prove it.
- ●Check training opt-out. Team and Enterprise: off by default. Verify before connecting sensitive data.
A DPA is the legal floor.
Scoping and minimization are the practical floor.
Responsible Use
What not to send to Claude
By default, keep personal data out. If you absolutely need to send it, anonymize it first.
Keep out (or anonymize if essential)
- • Full customer or employee names
- • Email addresses
- • Phone numbers
- • Postal addresses
- • Personal financial details tied to named individuals
- • Health or sensitive data (GDPR special category — keep out entirely)
You don't need names to answer "how much ARR is at risk."
Send the shape, not the identities.
Closing
Three things to take away
- 1.Cross-cutting business questions are now answerable in seconds. Your data already exists — you just needed a way to ask. No dashboards, no analyst, no code.
- 2.I picked Code for solo, recurring analysis. Composable with local files and scripts, file-by-file scope when I need it, persistent across sessions. Cowork wins for team-shared work.
- 3.Be mindful of what you send and from which plan. Personal data needs a DPA (Team or Enterprise). Limit the scope of what's reachable. Anonymize when in doubt.
Conversations with your own data —
within the scope you control.
Thank you.
Scan to connect —