Handling client data in analytics projects: access, copies and clean-up at the end

A data boutique finishes a customer churn project for a lender. Four months later, the client's security team asks where the extract is now. The analyst who built it has moved to another employer. A spreadsheet sits in a shared drive, a copy is in a chat thread, and a notebook still contains a few hundred real customer rows in its saved output.

Nobody was careless on purpose. The project simply had no plan for the end. This post sets out a plan covering the start, the middle and the close.

On day one, the project lead writes down the data the client will share. List the tables or files, say whether names, phone numbers or account details are included, and note whether the data is masked. Ask the client to mask what you do not need. Fewer sensitive columns mean fewer things to protect and fewer things to explain later.

Ask three questions in that first call. Who at the client owns this data? Who do we call if something goes wrong? Will the client want proof of deletion at the end? Write the answers into the kickoff note, because people change roles on both sides and the project may outlast their memory.

Read the contract and any data terms before work starts. Clauses on location of data, subcontractors, copies and deletion vary a lot. Have your own qualified legal adviser interpret them. Your job is to turn the agreed terms into daily habits that analysts can follow.

• List the people who need access and the reason for each. Keep it to the minimum.
• Use individual accounts, never a shared login.
• Give read access first. Grant write or export rights only to those who need them.
• Set an end date for each access at the start, and extend it only with the project lead's approval.
• Remove access on the day someone leaves the project or the company.

Agree one approved place where the data is stored and worked on. It might be the client's own environment, a locked cloud workspace you control, or a managed virtual desktop. Say plainly that data does not go to personal laptops, personal email, personal drives or chat threads. Where an extract really must leave the workspace, the analyst records it in a log with date, purpose, destination and the plan for its removal.

Notebooks and dashboards hide data in surprising places. Saved cell outputs, cached extracts, screenshots in slide decks and test fixtures can all hold real records. Build a habit of clearing outputs before sharing and checking decks for real names.

1. 1

   Kickoff note

   One page: data received, sensitivity, who has access, where it lives, what the contract says about copies and deletion, and who the client contact is for data questions.

2. 2

   Access setup

   Create named accounts with end dates. Brief everyone for ten minutes on the rules and ask them to confirm in writing.

3. 3

   Fortnightly check

   The lead reviews the extract log, checks the workspace for stray files and confirms that the access list is still right.

4. 4

   Incident route

   If data goes where it should not, the analyst tells the lead the same day. The lead informs the client as the contract asks and calls in your adviser.

5. 5

   Close-out

   Follow the checklist below, sign it, and send the client a short confirmation.

All individual access removed or handed back to the client.
Project workspace emptied or returned as agreed in the contract.
Extract log reviewed and every listed copy confirmed removed.
Laptops, drives, notebooks, chat threads and email searched for stray files.
Slides and reports checked for real names or identifiers.
Backups handled as the contract says, with the date noted.
Items the company must keep for a stated reason listed, with owner and retention decision.
Checklist signed by the project lead; written confirmation sent to the client.

Rules only hold if people understand why. Cover data handling in onboarding and again at the start of each project. Include it in your data privacy and security policy, written in plain sentences that an analyst can recall at the keyboard. For firms serving enterprise SaaS clients, expect security questionnaires and audits; having the extract log and close-out checklist ready makes these calmer.

Make the safe route the easy route. If the approved workspace is slow, analysts will copy data to a laptop for speed. Ask them what slows them down and fix it. A fast, well-provisioned workspace prevents more stray copies than any warning memo.

Resignation is the riskiest moment. Before the last day, the lead reviews what the person holds: files, notebooks, credentials, repository access. The person confirms in writing that nothing remains on personal devices. IT removes accounts the same day. Our guidance on exit and offboarding can help you build this into the leaver process.

Data handling also touches the tools analysts use. If your team uses assistants to write queries, read our note on rules for AI coding and writing tools, so client rows never go into a public tool. And recognise careful handling in reviews, not just speed.