Migrating Subscribers From an Old System Without Losing Anyone

6 min read

The hardest part of changing your management system isn't the new system. It's the old data.

Hundreds of subscribers, each with a plan, a quota, a renewal date and usage in flight. Any mistake in the move means a paying customer who can't get online — and they will call you, not the system.

This guide is about that part specifically. Everything in it is practical: every trap listed here has been met in real export files.

The golden rule: don't touch the username

Before anything else.

Your subscriber's credentials are saved in their router or device exactly as they are. Any "cleanup" you apply to that column — stripping a space, normalising a format, adding a missing leading zero — disconnects that person.

Real export files contain:

  • Usernames with a space inside the number: 010 62408022
  • Mixed lengths for the same type: 10, 11 and 12 digits in one file
  • Usernames with no relationship to the phone column at all (old typos)

Import them byte for byte. Cleanup — if you need it — belongs on the phone column you send messages from, not on the one that authenticates. That distinction is the difference between a clean migration and a day of angry calls.

Before exporting: agree on the cutover moment

Migration isn't "export and import". It's a moment that has to be planned:

  1. Export a snapshot at a known time
  2. Prepare and test on the new system — without switching any router
  3. Freeze changes on the old one (no renewals during the move)
  4. Export a second snapshot with just the delta
  5. Switch the routers
  6. Watch

Step 4 is the one people skip. One to three days pass between the first export and the switch, and renewals and payments happen inside that window. A second snapshot taken immediately before cutover catches them.

Try ten first. Ten real subscribers, on the new system, actually working — before you think about the rest. Every trap below shows up in those ten.

What to export

The minimum that makes a migration possible:

FieldWhy
UsernameAuthentication. Byte for byte
PasswordIn cleartext where available
Plan nameTo map onto the new system's plans
Start and end dateThe current cycle
Total quotaThe limit
Consumed quotaWhat's already been spent this cycle
PhoneContact and notifications

The traps that corrupt data silently

These are the ones that pass a "successful" import and surface a week later.

1. A byte is not always 2³⁰

Most systems write 57G and mean 57,000,000,000 bytes — decimal. If the new system reads that as binary gigabytes (57 × 1024³ = 61.2 billion), every subscriber quietly gets 7% more than they paid for.

The reverse is worse: a subscriber gets less than they paid for, and is cut off early.

Verify the unit by arithmetic, not by the label. Take a subscriber whose quota you know, divide the file's number by 10⁹ and by 2³⁰, and see which one produces the right answer.

2. The plan name lies

This is the trap that catches the most people.

In real exports, the same plan name carries different quotas per subscriber. A plan called 57G-4M will have subscribers on 57, 58 and 60 GB. A plan called 140G-10M turns out to carry an actual quota of 87 GB — the name says one thing, the number says another.

The cause is mundane: old promotions, customer-specific exceptions, and manual adjustments accumulated over years.

So: read the quota from the subscriber's row, and treat the name as a label. Any import that says "everyone on this plan gets the plan's limit" overwrites real agreements with real customers.

3. Usage: carry the remainder, not the consumption

A subscriber on 100 GB has used 60. If the new system starts them at zero, you have given away 40 GB. Get it wrong the other way and you cut off someone who has paid.

The remainder is the number that moves — and confirm that consumed ≤ total on every row before importing. Rows that break that rule are corrupt data, to be resolved by hand rather than imported.

4. The dates aren't the same type

The start column is usually a datetime (2026-08-15 14:32:00) and the end column a date only (2026-09-14). If the new system reads the second as midnight, the subscriber loses a day.

Across hundreds of subscribers, that shows up as a wave of calls on a single day.

5. Cards usually have no password

In most systems a hotspot card's code is the username, and the password column is entirely empty.

If the new system requires a password, you cannot import cards at all. Ask about this before you start.

6. Duplicates and deleted plans

You will find subscribers on a plan named --- (deleted long ago), or names containing commas that break the CSV. Decide what happens to them before the import, not during it.

After cutover, watch these specifically

The first 48 hours:

  • Failed authentications. A sudden rise is a username or password problem, not coincidence
  • Subscribers cut off unexpectedly — usually a miscalculated quota (trap 1 or 3)
  • Subscribers with no speed limit — the plan didn't map, or the speed is empty. And note: zero in a speed limit means unlimited, not cut off — so a plan misconfigured to zero grants open speed
  • Sessions that never close — accounting isn't being closed; check the accounting configuration

Leave a way back

Don't delete the old system on cutover day. Leave it running (and frozen) for at least a week.

Not because you'll roll back — you probably won't. But because it is the only reference you have when a subscriber says "I had 30 GB left" and you need to check.

Quick checklist

  • Usernames imported byte for byte, no cleanup
  • Byte unit (decimal or binary) verified by arithmetic
  • Quota read from the subscriber row, not the plan name
  • Remainder calculated correctly; consumed ≤ total
  • Dates converted correctly (especially the end date)
  • Passwordless cards have a path
  • Trial run on 10 real subscribers before the rest
  • Second snapshot immediately before cutover
  • Old system left running for a week afterwards

If you're planning a migration and have a question about a specific export file, get in touch — data differs between systems, and the fastest way to know whether a migration will go smoothly is to look at the file itself.

Read next