What does it collect?
Name every field, upload, message, export, and connection.
BrainIT Consulting · Free Field Guide No. 3
Collect only what the first process needs. Decide who may use it, keep consequential actions human-controlled, and learn safely before real work depends on the app.
A prototype crosses a line the moment you type in a real name, copy a customer message, save an internal note, or connect it to another business system.
That does not make the app automatically unsafe. It means the experiment now has consequences. A wrong color or awkward button is inconvenient. A private note shown to the wrong person, a lost record, or an automatic promise sent to a customer is different.
Name every field, upload, message, export, and connection.
Connect the field to the first process, not a vague future possibility.
Separate viewing, changing, approving, exporting, and deleting.
Know who notices a mistake and how the business returns to a safe state.
Security is not one switch a developer turns on at the end. You do not need to implement every technical control, but you must describe the business meaning those controls protect.
Follow one fictional inquiry through every place its information travels.
Include exports, email, spreadsheets, integrations, backups, logs, and printed copies. Information is not protected merely because the main screen requires a password.
For each stage, name a person who owns the decision. “The system handles it” is not an owner. Someone must know who may correct a record, approve a connection, restore a backup, and decide when information should be removed.
The easiest sensitive field to protect is the one the first version never collects.
Do not disguise speculation as a requirement. “We may want it one day” belongs in Nice Later. If a field is needed only for a rare exception, consider handling that exception outside the first version.
Collect for a reason, keep for a reason, and remove when that reason and any applicable obligation have ended. This guide does not invent a universal retention period.
Begin with invented records. Move toward real information only when lower-risk tests have passed.
| Kind | Meaning | Use during the first pilot |
|---|---|---|
| Fictional | Invented people, requests, addresses, and amounts. | Begin here |
| Copied and changed | A realistic shape from which names and identifying details have been removed. | Use carefully |
| Live | Current information about a real customer, employee, supplier, or transaction. | Introduce deliberately |
Changing one name may not make a record safe. A distinctive request, address, photograph, account number, or combination of details may still identify someone or reveal confidential business information.
Mark every invented record visibly, such as TEST — Rosa's Bicycle Repair. Never paste passwords, secret keys, full payment-card data, identity documents, or regulated records into a general-purpose coding or AI tool merely to make a test realistic.
“Everyone on the team” is rarely a useful access rule.
| Role | May view | May change | May approve |
|---|---|---|---|
| Owner or administrator | All inquiry records and access settings | Records, assignments, configuration | Users, exports, deletion, live launch |
| Assigned staff member | Records assigned to them | Status, work notes, next action | A reviewed reply within policy |
| Occasional helper | Minimum records needed for the task | The specific step they own | Nothing outside that task |
| App or automated job | Fields required for its narrow function | Drafts, flags, reversible changes | Nothing consequential without a person |
Give each person a separate account. Remove access when a role ends. Use multifactor authentication where the service supports it, especially for administrative access. Avoid shared administrator credentials and keep secret keys out of notes, source files, screenshots, and email.
Sign in as a limited user and confirm the account cannot see another person's records, change settings, or reach an administrator screen by editing a web address.
Let the app prepare work without giving it permission to complete every action.
Collect consistent fields, show missing actions, prepare a draft, flag waiting work, and display history.
Messages, promises, prices, payments, deletion, exports, permissions, and sensitive exceptions.
If AI is involved, begin in draft-for-review mode. Human review is a real control only when the reviewer has enough context, time, and authority to stop the action.
If the app is wrong, who notices, how quickly, can the action be reversed, and what evidence will show what happened?
A message that says “backup complete” is not the same as a proven recovery.
Write down what is backed up, where it exists, who controls the account, how often it is created, who notices a failed backup, and how to restore one record or the whole app.
Do not damage live information to prove recovery. A developer or hosting provider may need to create an isolated restore environment. Uploaded files may use different storage from text records and need their own recovery plan.
GitHub can preserve application code and change history. It does not automatically back up a production database, uploaded documents, or third-party service configuration.
Run the app beside the current process while a person remains in control.
Let at least one person who did not build the app perform realistic tasks. Observe rather than coaching every click. The goal is to discover what the app makes unclear.
Fill it in here, copy a blank template, or print this card.
| Case | Expected safe result | Observed result | Status |
|---|---|---|---|
| Normal | |||
| Missing information | |||
| Duplicate | |||
| Sensitive exception | |||
| Mistake and recovery |
Write down observed friction instead of collecting an unlimited wish list.
Prioritize anything that prevents a record from reaching “done,” reveals information to the wrong person, creates an unreviewed commitment, or makes recovery uncertain. Cosmetic wishes can wait.
A change is not verified because code was edited or a developer says it is fixed. Repeat the affected case and keep the observed result.
Review the card with the person responsible for the work and make one deliberate decision.
If live use begins, start with a small number of records and named users. Monitor access, errors, backups, unusual cases, and external actions. Review the boundary whenever the app gains a new integration, user role, data type, automatic action, or public page.
The goal is not merely “we launched software.” The business should understand what the app knows, what it may do, who remains responsible, and how to recover.
Nothing here requires a sales call. The Data Boundary and Safe Pilot Card is yours to copy, print, and use. If you would like a second pair of eyes, I can review the boundary with you, help run the five cases, or perform a focused security scan before live information is introduced.
Emile du Toit · BrainIT Consulting
brainitconsulting.com
Sources and limits. This guide draws on official guidance from the FTC, NIST, SBA, Digital.gov, and CISA. The worksheet and five-case pilot are BrainIT teaching devices, not official government checklists.
FTC: Start with Security · FTC: Protecting Personal Information · NIST: CSF 2.0 Small Business Quick-Start Guide · SBA: AI for small business · NIST: AI RMF Core · Digital.gov: Usability testing. Accessed July 28, 2026.
This is general educational guidance, not legal advice, a compliance certification, or a guarantee against loss. Requirements vary by industry, jurisdiction, contract, and the kind of information involved.