CampaignSMS

Gemini decides, SMS delivers: closing the last mile of workflow automation in 2026 – iTWire

|
GUEST OPINION: A workflow in which AI-assisted office tools hand a trigger to a phone notification, the last step of an automated business process.
By late 2026, most Australian IT teams can get AI to draft, summarise and prioritise. Ask any of them whether the customer actually hears about the result, and the answers get quieter. In our conversations with channel partners and MSPs, the same pattern keeps surfacing: the clever part of the workflow works, and the final step quietly fails.
That final step is where automation either earns its keep or doesn’t. An AI model can decide that a payment has failed or an appointment needs moving. Someone still has to tell the customer, in a way that arrives in seconds and can be trusted. This piece joins two topics we usually cover separately: AI inside the office suite, and the messaging layer that carries its decisions out the door.
Most workflows end in an inbox. A flagged invoice gets an email, a delayed order gets an email, a security alert gets an email. Then nobody reads it for six hours.
Time-critical messages such as one-time passcodes, order updates, appointment reminders and payment alerts need a different path. Teams that wire these alerts straight into a transactional SMS API get a delivery route built for urgency rather than one borrowed from marketing email. The message is short, it’s tied to a single event, and it lands on a device people carry everywhere.
There’s evidence that a well-timed text can do more than simply reach a handset. An Australian Government study into SMS reminders found that sending a reminder increased on-time reporting by 13 to 14 percentage points compared with sending no message. That doesn’t mean every text gets read or acted on, but it makes a stronger case for SMS in time-sensitive workflows: the value lies in prompting a measurable response, not just recording a successful delivery.
Reach isn’t the whole story either. SMS is plain text with no rich formatting, and long, complicated instructions belong elsewhere. Rich Communication Services (RCS) offers a more interactive alternative, supporting branded messages, images and actionable buttons. But not every device or network supports RCS, which is why SMS remains an important fallback for businesses that need to deliver time-sensitive notifications reliably.
Sender identity is now an infrastructure concern. Since 1 July 2026, only sender IDs registered on the ACMA SMS Sender ID Register display as the brand name on Australian handsets. Unregistered IDs fall into a single “Unverified” thread, alongside whatever else the network couldn’t vouch for.
The mechanics are simple enough. Sender IDs run from 2 to 11 characters, registration goes through an approved telco or messaging provider, and the check draws on the Australian Business Register. The Spam Act hasn’t gone anywhere, either. Consent and opt-out obligations still apply to every message you send.
For IT operations, the takeaway is practical. A customer who sees an “Unverified” label on a payment alert will, quite reasonably, assume it’s a scam. Registering your sender ID belongs on the same checklist as DNS records and certificates.
The risk extends beyond customer confusion. As we reported ahead of the July deadline, businesses that failed to register their branded sender IDs risked having legitimate appointment reminders, billing notices and urgent updates grouped with suspicious messages. For automated workflows, that creates a problem no amount of clever AI can solve: a notification may be delivered successfully but ignored because the recipient doesn’t trust its source.
The other half of the story is the decision layer. Gemini now works inline across Gmail, Drive, Docs, Sheets, Chat and Calendar, and Gemini for Google Workspace has moved well past the drafting-assistant stage. Google launched Workspace Intelligence in April 2026, and Workspace Studio lets teams build Gemini-powered automations in plain language, with custom Apps Script steps for anything the templates don’t cover.
Adoption is real. Google Cloud said in its Gemini at Work 2026 post, published in October, that around 80% of its cloud customers now use its AI products. It also published customer results, among them Nokia cutting issue resolution time by up to 80% and Tata Steel halving complaint turnaround. These are vendor-selected case studies, so read them as best cases rather than averages.
Independent evidence is thinner. A Forrester Total Economic Impact study from February 2026, commissioned by Google, found interviewees saving three to five hours per user per week. Commissioned studies tend to flatter their sponsors, but the direction matches what we’ve seen in our own coverage. We also looked at the agentic shift in Gemini traffic earlier this year, and the trend line points the same way.
Note what’s missing from all of this: none of it describes how a decision reaches a customer outside your organisation. Chat and email cover internal staff. Customers need something else.
What follows is an architecture suggestion, not a documented joint feature from either vendor. We haven’t found a source describing an official integration, and you shouldn’t assume one exists. The pieces do fit together, though.
Trigger. A form submission, a new row in a Sheet or an incoming Gmail message starts the flow.
Decision. Gemini classifies urgency, checks the record against your rules and drafts the message text.
Call. An Apps Script step sends the text to a messaging API using UrlFetchApp, which needs the external request scope Google documents for scripts.
Confirmation. The delivery receipt gets written back to the Sheet, so a failed send is visible rather than silent.
Step four is the one teams skip, and it’s the one that matters. Without a receipt, you’ve automated the sending but not the knowing.
Agents that can text customers need tighter rules than agents that summarise meetings. Before anything goes live, we’d want these in place:
Consent records. Know who opted in, when, and how they opt out.
Registered sender ID. Confirm it’s live on the ACMA register before launch.
Per-agent permissions and audit trails. Gemini at Work 2026 describes agents with their own identities, audit trails and spend caps. Use all three.
Human approval for sensitive messages. Anything involving money or security gets a person in the loop.
Fallback channels. If a message fails, decide in advance whether email, RCS or a call takes over.
Pilots tend to stall exactly here. In our view, AI spending keeps outrunning the governance needed to scale it, and a flow nobody has signed off on rarely survives a risk review.
The gap between AI adoption and operational oversight is already showing up in the numbers. Research we previously covered found that 56% of surveyed IT and governance professionals were unsure how long it would take to shut down an AI system following a security incident. For businesses connecting AI agents to customer-facing messaging, that uncertainty matters. An automated system needs more than permission to send a message; it needs clear limits, monitoring and a reliable way to stop when something goes wrong.
Pick one high-value flow, such as an appointment change or a failed-payment alert, and build it end to end. Don’t start with ten workflows and a steering committee.
Measure two things over 90 days: delivery rate and time from trigger to customer action. If both improve, expand. If they don’t, you’ve learned that cheaply, and the problem probably sits in the decision logic or the consent data rather than the channel.
AI will keep getting better at deciding what should happen. The teams that benefit will be the ones whose customers reliably find out.

14 Nov 2026
14 Nov 2026
14 Nov 2026
14 Nov 2026

source

Leave a Reply

Your email address will not be published. Required fields are marked *