Monitor delivery per tenant, not per platform. A platform-wide delivery rate hides the one customer whose texts are being filtered, because the healthy majority averages the problem away. Track each sub-account on its own, compare it to its own history, and alert on change.

This post is for CTOs and heads of product at platforms that text on behalf of customers. It describes a practice, not a benchmark. We don't quote an industry delivery rate here, because carriers publish none, and the right threshold for your traffic is whatever your own baseline says.

What should you measure?

Measure outcomes by tenant, number and carrier, over a rolling window. Three counts do most of the work: texts sent, texts reported delivered, and texts reported failed or undelivered. Add opt-outs and inbound replies, because they tell you how recipients received the traffic.

SignalWhere it comes fromWhat it tells you
Delivered shareStatus events on each messageWhether texts reach handsets
Failure codesError code on failed messagesWhy: filtering, invalid number, landline
Opt-out shareInbound STOP messagesWhether recipients wanted the text
Reply shareInbound messagesWhether the conversation is real
Refused before sendYour pre-send gatesProblems caught before they cost anything

Be honest about one limit: a delivered status means the downstream network accepted the text, and a handset can still not show it. Treat delivered share as a trend line, not an exact truth. Silent filtering is the hard case, since some filtered texts get no clear failure at all. That's why reply and opt-out shares matter alongside the status counts.

Why per tenant and not one number?

Because carrier reputation attaches to a brand and campaign, and each customer should have its own. If you register every customer separately, as we describe in 10DLC for SaaS platforms, one tenant's bad list damages that tenant's registration and not everyone's.

Your monitoring should mirror that boundary. Group by sub-account, then by sending number inside it. A single bad number can pull a tenant's average down while the others are fine, and that's a different fix: retire or rest the number instead of rewriting the campaign.

How do you separate bad content from bad data?

Read the failure codes before you guess. A cluster of invalid or unreachable number errors is a data problem, usually an old list. A cluster of filtering errors is a content or reputation problem. The two need opposite responses, and mixing them is how teams rewrite perfectly good templates.

A quick triage order:

  1. Check the failure code mix for the tenant over the last few hours.
  2. Check whether the drop is one number, one carrier or everything.
  3. Look at what changed: a new template, a new list, a new link, a volume jump.
  4. Check the registration status for the tenant.

For codes, keep the SMS API error codes guide open. For content causes, carrier filtering and public URL shorteners and link in the first message cover the usual suspects.

How do you set thresholds?

Set them relative to each tenant's own baseline, and use two levels. A warning level opens a ticket for your team. A critical level pauses or throttles that tenant's sending. Both should need a minimum volume, or a tenant that sent four texts and had one fail will page someone at 3 AM.

As an example with stated assumptions: say a tenant's rolling seven-day delivered share has been steady, and you've picked a warning when the last hour falls well below it, and a critical when it stays low for several hours with at least a few hundred texts sent. Those exact numbers are placeholders. Pick yours from a few weeks of your own data, and revisit them after any change in registration or volume.

Two patterns keep alerts useful:

  • Compare to self. A dental office and a promo-heavy retailer have different healthy ranges.
  • Alert on change, not level. A sudden drop matters more than a steady lower number.

Volume matters too. Throughput is set per campaign, roughly 1 to 4 segments per second for a fresh standard campaign with no vetting, and carriers cap daily volume by trust score for T-Mobile. A tenant hitting those limits sees delays or throttling that looks like a delivery problem. Our post on campaigns approved then throttled covers the difference.

What should happen when an alert fires?

Decide the response before the alert exists. Someone at 2 AM shouldn't be inventing policy. Write down who is notified, what the platform does automatically, and what the tenant is told.

A sensible default ladder:

  • Warning: notify your support team and flag the tenant's dashboard.
  • Critical: stop campaign sends for that tenant while leaving one-to-one replies open, and email the account owner with the likely cause.
  • Recovery: resume gradually, not all at once, and keep watching.

Keep your tenants informed in plain words. A message like "Your texts to one carrier are being filtered, here's what we think caused it" is far better than a silent pause. Showing pre-send refusals clearly also saves your support team from guessing, as we describe in pre-send refusal error UX.

What can a messaging provider do for you?

It can stop some problems before they reach a carrier. Ready refuses opted-out numbers and numbers on your ban list up front, without charge, and refuses numbers its own sends have already seen as landlines. Registration is per sub-account, and unregistered traffic gets carrier-filtered, which is why registration status belongs on your monitoring dashboard.

Ready doesn't fail over between providers, so don't design a monitor that expects traffic to reroute itself. Your levers are the tenant, the number, the content and the list. You don't have to compute delivered share yourself: GET /sms/my-delivery-rate returns each account's delivery rate for the last 24 hours, 7 days and 30 days with a count per carrier, and compares two windows. Send it with X-Sub-Account-Id to read one sub-account (docs). Ready also checks each account every 30 minutes and posts a "Delivery rate dropped" notice in the app when more than 30% of at least 50 texts failed in the past hour. For reply and opt-out shares, read messages through GET /v1/messages, check registration with GET /v1/registration, and keep that history in your own tables. The data model post shows one way to store status transitions so these counts are cheap to compute.

The practical takeaway

Watch each tenant against its own baseline, split failures by cause, require a minimum volume before alerting, and write the response ladder in advance. You'll catch a bad list or a filtered template within hours instead of when a customer complains.

If you want to see how sub-accounts and registration are set up per customer, the sub-accounts docs are the place to start.