How can I answer a question about a client when the details are spread across several systems?

Map the sources once. For each spreadsheet, folder or export, record what it covers and how the client is named in it, because the same client is often a full legal name in one file, a short name in another and a code in a third. After that you type one command and a name, and get a single table with the source on every line. Octavia Niekoop built this in the 3rd Space AI Circle and went from ten to fifteen minutes per question to under a minute.

What happens when a client is named differently in each file?

That is the thing that breaks a lookup, and it is why the naming is recorded per source at setup with two real examples each. Some sources do not hold the client's name at all — a device export keyed by serial number, a portal keyed by contract number — so the map also records which other source holds the key, and the lookup reads that one first.

Is it safe to point an AI at client data?

This workflow is read only: it never edits, moves or deletes anything in any source, and it keeps nothing about the client after the answer. It runs on your own accounts and your own files, never anyone else's. Only connect sources your company allows you to use with AI tools. When two plausible matches turn up, it stops and asks rather than guessing.

3rd Space Bucharest Events AI Circle Partners Blog

3rd Space Bucharest · AI Workflows Card 008 Shipped Oct 2026

One name in, every system asked, one table outClient Snapshot

Type a client’s name. Claude checks every place a piece of the answer lives and hands back one table — fast enough to answer your boss in the middle of a meeting.

CreatorOctavia Niekoop

RunsEvery time someone asks

Sources5+

Answer inunder a minute

One question about one client, by hand 10–15 min

~7 min — opening five systems and finding the client in each ~4 min — follow-up questions, each one back into another file ~2 min — writing the answer back

The real change isn’t the minutes. It’s that the answer arrives while the question is still in the room. “I’ll get back to you” became “here it is.”

“Now if he asks something, even during a meeting, I pull out my phone, write the command and simply tell him.”

Octavia Niekoop · made this one · Head of Operations & Transformation Strategy

Overview

What this does

Every business has a client that lives in five places. Payments in one file. The account manager in the CRM. The technical history in a shared folder. The support log somewhere else. Each system knows one piece, and nobody knows the whole picture without opening all of them.

So when your manager asks “what’s going on with this client?”, the honest answer is “hmm, let me check…” and fifteen minutes of tabs.

This skill gives the answer in one. You type a short command and a client’s name. Claude goes through every source you’ve mapped, finds the client in each one — even where they’re called something slightly different — and returns one table, with the source written next to every line.

Octavia built it for a company with five large data sources. Her colleagues were, in her words, utterly shocked. Nothing from those sources appears on this page: the data stays where it is. What’s shared here is the method.

Before

What answering one question used to take

A manager, a client name, and five tabs.

01Five systems for one answer. CRM, contracts, project files, support log, receivables — each opened, searched, and read separately.

02Every answer brought a follow-up. “Hmm, let me check… Ah, you want the invoice status too? Okay, let me find that…” Each question sent you back into another file.

03Each colleague held one piece. Sales knew the deal, the project manager knew the build, finance knew the payments. The whole picture lived nowhere.

After

How it works, in three stages

The first stage is done once. The other two take less than a minute, every time.

1once

Map your sourcesa conversation with Claude

Claude asks, you answer: where each piece of information lives, and how the client is named there.

Every source: the app, the spreadsheet, the folder.
How the client is named in each one. “Acme Ltd” here, “ACME” there, a client code somewhere else.
Which columns to pull from each, and the table you want back.

Only you have this Knowing where things really live — including the file everyone uses but nobody documented.

2seconds

Type a command and a namefrom your laptop or your phone

One short command, one client. That’s the whole input.

Claude searches every mapped source for that client.
Two possible matches? It asks which one. It never picks.
It only reads. Nothing in any source is changed.

Works in a meeting On the phone, under the table, while the question is still being asked.

3under a minute

Get one tablewith the source on every line

Payments, people, history, open issues — in one place, each line saying where it came from.

Anything missing reads “not found in [source]”. Never a guess.
The source column lets you check any line in seconds.
Nothing about the client is kept after the answer.

You decide What to say to your boss. The table gives you the facts; the judgement is still yours.

Example

The same method, in a different business

A property manager, an apartment, and five places that each know one piece. Nothing to do with the original — that’s the point.

The command: snapshot Ap. 14

Source	The apartment is called	What it pulls
Lease contracts Drive folder	“Ap. 14, Str. Exemplu 3”	Tenant, start date, rent
Payments Spreadsheet	“AP14-E3”	Last payment, amount due
Repairs Shared log	“Exemplu 3 / 14”	Open jobs, last repair, contractor
Meters Utilities export	Meter number only	Last reading, date
Contacts CRM	Tenant’s name	Phone, email

Four sources, four different names for the same apartment. And the meters don’t know the apartment at all: the lease holds the meter number, so the snapshot reads the lease first and uses that number to search the export. Chains like that are why the question couldn’t be answered in a meeting before.

Tools

What you need

Claude · Cowork Your sources, connected Spreadsheets (Excel, Sheets) Cloud folders Your CRM Claude on your phone

The hard part

A blank answer beats a confident wrong one

The one to design for

The client is called something else in the payments file.

In the CRM the client is “Acme Trading Ltd”. In the receivables file it’s “ACME-TRD”. Ask for the snapshot and a careless setup finds the CRM record, finds nothing in receivables — and reports no outstanding payments.

That’s the dangerous answer. Not an error, not a gap: a clean, confident line that’s wrong. And it gets said out loud, in a meeting, to the person who decides things.

Nothing found is not the same as nothing owed.

Map the names, not just the sources. For each source, write down how the client appears there.
Make “not found” visible. Every gap reads “not found in [source]”, never an empty cell or a zero.
Put the source on every line. So the one line that matters can be checked in seconds.
Test on clients you know by heart, including the one with the awkward name.

Read only, always

A lookup that can also edit is a lookup that can break your records. Give it reading rights and nothing more.

It asks. It never changes.

The data stays home

Client data is sensitive: money, contracts, problems. The skill reads it where it lives and keeps nothing afterwards. Check your company’s rules on AI tools before you connect anything.

Share the method, never the data.

Two matches? Ask

Two clients with similar names is normal. Picking one silently is how you brief your boss on the wrong company.

“Did you mean A or B?”

Three more things worth knowing

Some sources don’t know the client’s name

An export from a device or a portal might only know a serial number. Write down the chain: the contract holds the number, the number finds the data.

Map the links, not only the files.

Start with the questions you actually get

Write down the last three things your manager asked. The table’s columns come from those, not from everything the files contain.

Three real questions, one table.

It works far beyond sales

A property manager and an apartment. A school and a student. An agency and an account. Same method, different shelves.

See the second tab.

Your turn

Copy this and use it today

Two documents. The first builds your snapshot. The second is a filled-in example from a completely different business.

Step 1

Copy the template

Pick a tab below and press Copy.

Step 2

Map your sources with Claude

Paste it into Cowork with your files connected and say: “Interview me through this, one question at a time.”

Step 3

Test on clients you know

Three you know by heart, then the one with the awkward name. Check every line against its source.

Step 4

Make it a skill

Say “turn this into a skill”. Next time, it’s one word and a name.

Or install the ready-made skill: client-snapshot. It maps your sources on the first run.

client-snapshot-template.md

# Client Snapshot — template

Paste this into a new Claude chat (Cowork, with your files or apps connected) and say: "Interview me through this, one question at a time. Then turn it into a skill."

---

## 1. The question you keep getting
Who asks you about a client, and what do they want to know? Write the last three questions exactly as they were asked.

> Example: "Has X paid?" "Who's handling X?" "What went wrong at X last month?"

## 2. Your sources
List every place a piece of the answer lives: an app, a spreadsheet, a shared folder.
For each one:
- where it is (the app, or the file path)
- how the client is named there (full company name? short name? a code?)
- which columns you need from it

> The naming is the part that breaks. The same client is "Acme Ltd" in one place, "ACME" in another, and "C-0142" in a third.

## 3. The table you want back
Which columns, in what order? Put the source next to every line.

| Topic | Answer | Source |
|---|---|---|
| Last payment | … | Invoices file |
| Account manager | … | CRM |
| Open issues | … | Support log |

## 4. The rules
- **Not found is an answer.** Write "not found in [source]". Never fill a gap with a guess.
- **Two possible matches?** Ask me which one. Never pick.
- **Read only.** Change nothing in any source.
- **Don't keep anything.** No notes about the client after the answer.

## 5. The command
What will you type? One word and a name.

> Example: `snapshot Acme`

## 6. Test it
Run it on three clients you know well. Check every cell against the source. Then try one with a tricky name.

# Example source map — a property manager

A filled-in section 2, for a business that has nothing to do with the original. Same method, different shelves.

Command: `snapshot <apartment>`

| Source | Where | The apartment is called | Pull |
|---|---|---|---|
| Lease contracts | Drive › Leases | "Ap. 14, Str. Exemplu 3" | tenant, start date, rent |
| Payments | Payments.xlsx | "AP14-E3" | last payment, amount due |
| Repairs | Repairs log (Sheets) | "Exemplu 3 / 14" | open jobs, last repair, contractor |
| Meters | Utilities portal export | meter number only | last reading, date |
| Contacts | CRM | tenant's name | phone, email |

Note the meters row: the apartment isn't named there at all. The lease holds the meter number, so the snapshot reads the lease first and uses the meter number to search the export. Write chains like this down. They're the reason you couldn't answer in a meeting before.

Built by Octavia Niekoop Head of Operations & Transformation Strategy · AI Circle, October 2026 · Credential AIC-2026102

More cards in 3rd Space AI Workflows — each one a real workflow from a real member, with the template attached.