Field and support operations · Updated 2026-10-16

ISP operations on bad connections and cheap phones

A practical, vendor-neutral guide for support desks and installation crews working with patchy 3G, shared Android phones and expensive mobile data.

Start with the work that must survive a lost signal: identifying the job, recording what changed and telling the customer what happens next. A responsive screen alone does not prove that a save survives an interruption. For device-side preparation, read the TR-069 compatibility guide.

Where the data goes

Separate the staff phone's allowance from CPE-to-ACS traffic. A TR-069 Inform normally travels from the customer's device to the management server, not through the technician's phone; tethering can put both on the same data allowance.

ActivityWhat to countPractical control
DashboardsInitial HTML, scripts, fonts, charts and repeated refresh requestsOpen the queue you need; prefer a compact list and deliberate refresh.
MapsTiles, imagery and route changes while panning or zoomingPrepare directions on Wi-Fi; save a permitted map area or a written landmark description.
Installation photosOriginal files, previews, retries and copies sent to multiple placesCompress a readable copy once and attach it to the job.
TR-069 InformsThe whole management session in both directions, including transport overheadMeasure a representative session and agree an interval with the NOC.
Chat appsAutomatic media downloads, video, voice notes and cloud backupsTurn off unnecessary media downloads on mobile data; send a short job reference and text update.

There is no universal number for a dashboard, map or Inform. The examples below are planning arithmetic, not measured product benchmarks. We use decimal units: 1 kB = 1,000 bytes and 1 MB = 1,000,000 bytes.

A transparent example data budget

Assume a day with 20 cold page loads at 0.5 MB transferred per load, 6 jobs with 3 photos each at 200 kB per photo, and 2 MB reserved for text updates and small attachments. That is 10 MB + 3.6 MB + 2 MB = 15.6 MB. These are chosen inputs, not typical usage. This subtotal excludes maps, background polling, operating-system updates, retries and other apps; measure those separately before buying a bundle.

If the same 18 photos were 3 MB originals, their file bytes alone would total 54 MB. The comparison assumes one upload per file and excludes transport overhead. A failed upload retried from the beginning consumes more; do not budget only the final successful file.

Inform intervals: a separate network budget

For an illustrative fleet of 1,000 CPE, assume 20 kB per complete periodic session, including both directions and transport overhead, with no retries, event-triggered sessions, diagnostics or firmware downloads. With periodic Inform enabled, a 300-second interval produces 86,400 / 300 = 288 sessions per CPE per day, or 5,760 MB per fleet day. A 900-second interval produces 96 sessions and 1,920 MB under the same assumptions. The 20 kB is an input to replace with a measurement, not a protocol constant.

Longer intervals trade traffic for less frequent periodic evidence. Agree the freshness requirement first; do not change a whole fleet merely to match this example. Broadband Forum defines PeriodicInformInterval in seconds; it does not prescribe a fixed session size. See the existing TR-069 ACS integration scope for device-management context.

Keep work moving without a signal

Before departure, prepare a job sheet containing the job reference, customer-approved contact details, address and landmarks, equipment serials, agreed work, safety notes and dispatcher contact. Use a printed sheet or an approved local copy that opens in airplane mode. Save only the customer information needed for that visit; do not put router passwords or payment credentials on the sheet.

Use this handoff sequence with whichever tools your team has verified:

  1. Prepare: download the assigned jobs while connected and open the saved copy before leaving.
  2. Capture: record notes with the job reference and capture time. Keep photos attached to the same reference in an approved local store.
  3. Mark pending: distinguish locally saved, waiting to upload and office-confirmed records. If an app has no durable queue, use the prepared sheet and approved local files; do not assume a spinner means saved.
  4. Sync later: reconnect on a trusted connection, submit the pending changes and check the office record. Reconcile duplicates and conflicting edits with the dispatcher before closing the job.
  5. Clear the handoff: follow the operator's retention policy for local customer data, especially before a shared phone changes hands.

Test the actual device: enter a disposable test note, lose the connection, close and reopen the app, then reconnect. Check whether the note survives and whether a retry creates a duplicate. Keep a paper fallback until that behavior is proven. On shared phones, use individual staff accounts where supported, lock the screen between visits and agree who owns any unsynced work before signing out.

Photo targets for proof of installation

As a starting target to trial, export a JPEG copy with a 1,280-pixel long edge, using roughly 70–80% encoder quality, and aim for 150–300 kB per photo. These are proposed settings, not guaranteed output sizes: scene detail, noise and the encoder change the result. Measure the exported file. The earlier daily budget chooses 200 kB within this trial range.

Review the copy on the receiving phone. Can support read the serial label and see the cable route and mounting detail? If not, take a closer, steadier photo or retain more detail; evidence quality wins over the target. Keep an original when your evidence policy requires it, and schedule its larger upload separately. Avoid repeated recompression through chat forwards. Do not upload customer faces, household details or location metadata unless needed and authorized for the job.

SMS fallback when mobile data is unreliable

SMS can reach a customer whose packet-data service is unavailable, provided the handset and mobile network support SMS and have service. It is not a remedy for no cellular coverage, a switched-off phone or a failed SMS route. Agree who sends the fallback, when a message expires and how to prevent a late chat message and an SMS from creating contradictory instructions. A delivery receipt does not prove the customer read it.

Encoding changes the segment count

Under the common SMS concatenation scheme, a GSM-7 message holds 160 septets in a single segment or 153 per segment when concatenated. UCS-2 holds 70 code units singly or 67 per concatenated segment. These limits assume the usual concatenation header, no national-language shift tables and no special route limits; confirm your gateway's rules. See the SMS length reference.

French accents do not all force UCS-2: é, è, à and ù are in the basic GSM-7 alphabet; ê is not. Curly quotes can also trigger a switch. GSM-7 extension characters such as € consume two septets, so visible character count is not always the billable count. See the GSM-7 character map.

For a 100-character message made entirely of basic GSM-7 characters, assume one septet per character: it fits one segment. Replace a character with ê, keeping 100 characters in the basic multilingual plane, with no transliteration: it needs ceil(100 / 67) = 2 segments. If the route charges the same price per segment, that doubles the segment charge. It does not prove every French message costs twice as much: length, substituted characters, provider fees and route rules matter. Emoji can consume more than one code unit.

Check the final rendered message after inserting the customer's name, job reference and support number. Prefer clear short wording; do not remove meaningful accents blindly. Keep personal account details out of notifications that can appear on a shared phone's lock screen.

Choose a support tool by work completed and bytes transferred

Compare the same task on the same low-cost phone: open a customer, read the last note, add a test note and confirm it saved. Measure the first visit with an empty cache and a repeat visit separately. In browser network tools, record transferred bytes, not just the decoded resource size; include requests made while completing the task and while idle. Chrome documents the distinction in its Network panel reference.

Record the phone model, browser, network latency and throughput, cache state and test steps. Try the real route your staff travel as well as a repeatable throttled lab run. Check long lists, keyboard use, readable labels, upload recovery and what happens when the signal drops during a save. A small page that loses a job note is not a successful tool.

A Lighthouse result measures a specific page under stated conditions. It does not establish that every authenticated CRM page is fast, or that local saves and later synchronization work.

A low-data operations day checklist

  • Prepare assigned job sheets, directions and essential contacts on Wi-Fi.
  • Confirm that saved information opens without a connection and that the battery has enough charge for the route.
  • Disable unnecessary media auto-downloads and defer large app updates.
  • Capture readable photo copies; check their file sizes before uploading.
  • Mark unsent notes and evidence as pending with a job reference.
  • Check SMS encoding and the final wording before using the agreed fallback.
  • Reconcile pending work and confirm the office received it before closing jobs or handing over the phone.

Where ISPAgents fits: planned and guided-pilot boundaries

ISPAgents low-bandwidth mode is planned; no measured CRM performance claim is made here. The O5 acceptance gate for CRM pages has not yet been measured; a result for this public guide cannot establish product performance.

An offline-tolerant technician view in ISPAgents is planned, not shipped. Field-service foundations are a guided-pilot scope; the full technician PWA and offline synchronization remain roadmap work.

ISPAgents notifications are a guided-pilot foundation; automatic SMS fallback is not a finished customer workflow. Treat the checklist above as an operating procedure to validate with your existing tools, not a claim that the platform implements it.

For the next planning step, discuss your crew's operating constraints. Continue with the device compatibility guide or subscriber and device identity guide.

Frequently asked questions

Does every French accent make an SMS more expensive?

No. Characters such as é and à are in GSM-7. Others, such as ê, can switch the entire message to UCS-2. Check the final text, including names and punctuation, with your provider before sending.

What should a crew do when a save cannot be confirmed?

Keep the local note and job reference, mark it pending, and reconcile with the dispatcher when connectivity returns. A pending upload is not proof that the office received the evidence.

Is the low-data workflow available in ISPAgents?

ISPAgents low-bandwidth mode and an offline-tolerant technician view are planned, not shipped promises. Notifications are a guided-pilot foundation; automatic SMS fallback is not a finished customer workflow.