The CPaaS Company
Insights
Independent, source-cited writing on cloud communications — original research alongside the best work published elsewhere.
38 pieces
The state of business messaging, 2026
Four changes that landed between 2025 and 2026 — per-message billing, mandatory registration, RCS on every modern handset, and AI disclosure — and what each one does to a programme built before them.
What CPaaS actually is, and what it is not
The category is defined more clearly by what it replaced than by any vendor's feature list. A working definition, and the three questions that separate a real platform from a reseller.
Channels are not interchangeable, and treating them as if they are is the most common early mistake
SMS, WhatsApp, RCS and email differ in who may start a conversation, what may be said, and what happens when the recipient ignores it. Those differences are not cosmetic.
The 24-hour window: the rule that breaks most first integrations
Why your test message worked and your production message did not, explained without reference to any particular vendor's documentation.
A2P, P2A, P2P: why the direction of a message decides everything about it
The same 160 characters are priced, routed, filtered and regulated differently depending on whether a person or a program pressed send.
Omnichannel is an identity problem, not a channel count
The difference between multichannel and omnichannel is not how many channels you support. It is whether the customer has to repeat themselves.
Consent is a data model, not a checkbox
A boolean cannot answer the question a regulator asks, and cannot answer the question your own send path needs answered either.
How to read a market forecast without being misled by one
Three reputable firms size this market at figures that differ by more than a factor of two. None of them is wrong.
What RCS changed, and what it did not
Apple's adoption in iOS 18 turned a long-stalled standard into something with reach. Reach is not the same as a business case.
Delivery rate is not a result
It is the metric most often reported and the least often useful. What to measure instead, and why the funnel is steeper than anyone's dashboard suggests.
The five layers of a communications stack, and which one you are actually shopping for
Most evaluations compare transport, which is the commodity layer. The four above it are where programmes succeed or fail.
Video without a vendor: what peer-to-peer actually removes
For small calls, the media server is optional. Understanding when it stops being optional is the whole decision.
Conversational marketing, minus the hype
A useful definition has to exclude something. This one excludes most of what gets sold under the name.
WebRTC in plain terms
What the browser gives you for free, what it does not, and the three acronyms you cannot avoid.
Choosing a channel per message, not per programme
The channel decision belongs at send time, driven by intent, region and what consent you hold — not to a procurement meeting eighteen months earlier.
10DLC registration without tears
US carriers block unregistered A2P traffic outright. The process is two layers, it takes longer than your launch date assumes, and most rejections are avoidable.
Getting messages into the Gulf: sender IDs, licensed aggregators and the AD- prefix
Regional A2P delivery is a licensing question before it is a technical one, and a global platform does not inherit local standing.
A template strategy that survives contact with production
Templates are approved in advance, billed by category and impossible to improvise. Design the set before you need it, not during an incident.
Measuring a conversational programme without fooling yourself
A method: choose one outcome, build a permanent holdout, instrument the funnel, and be honest about what the numbers cannot settle.
Containment is not resolution, and optimising the first destroys the second
A support agent can post excellent containment numbers while making customers angrier. The metric that separates the two is uncomfortable and worth adopting.
Telling people they are talking to a machine
From 2 August 2026 this is a legal requirement in the EU, with specific rules about how it must be presented. It is also, independently, the thing that makes automated support tolerable.
When to build messaging infrastructure, and when not to
A decision framework based on what actually breaks in production, rather than on a feature comparison.
Latency you can actually feel
The network is rarely the slow part. Template approval, queueing and carrier acceptance dominate wall-clock time, and only one of them is under your control.
Designing a journey that actually branches
Most 'journeys' are sequences with a delay between steps. Four decisions turn one into something that responds.
The support queue as a system, not an inbox
Routing rules, service-level targets and the distinction between first response and resolution — the machinery that separates a support function from a shared mailbox.
Arabic-first is not translation
Direction, numerals, name order, segment length and register. A localised product differs from a translated one in ways that show up in the first minute.
Adding RCS to an SMS programme without running two programmes
Every RCS deployment is two deployments: the rich one and the fallback. Designing the fallback first is what keeps it from being an afterthought your audience receives.
When video belongs in a support conversation, and when it is theatre
Video resolves a narrow class of problems dramatically faster and every other class slightly slower. Knowing which is which is the entire design.
A consent architecture that survives an audit and a vendor migration
A schema, a set of invariants, and the three queries every send path and every regulator eventually asks.
Federating systems without merging them
When several products serve genuinely different audiences, consolidation destroys the thing that made each one good. Published contracts do the same job without the loss.
The gap nobody fills: service-to-service identity
Every platform in this category issues credentials to humans and to applications. Almost none can express 'system A is acting on behalf of customer B, with these permissions, until this moment'.
The real cost of an unofficial transport
Unofficial protocol implementations are cheaper, faster to launch and genuinely useful. They also transfer a risk to your customers that you should be pricing.
What the analysts disagree about, and why it is the useful part
Three firms, three market sizes, one category. Reading the disagreement tells you more about your own decision than any single number does.
Article 50 for messaging teams: what the AI Act's transparency rules require
A reading of the transparency obligations that took effect on 2 August 2026, focused specifically on automated conversations rather than AI in general.
RCS after iOS: what changed in the numbers, and what to distrust
Adoption figures for RCS are widely quoted and rarely defined. What can be said confidently, and what is being counted loosely.
Evaluating a CPaaS contract: the clauses that decide what you actually bought
Unit price is the least consequential number in the agreement. Nine provisions that matter more, and what a good answer to each looks like.
Regional messaging in the Gulf: why global platforms underperform here
Four structural features of the market that a globally-designed product does not inherit, and that decide outcomes more than software quality does.
GSMA on RCS, encryption and the future of operator messaging
Collected primary material from the GSMA on where operator-run messaging is heading — the standards body's own account, rather than a vendor's summary of it.