Melbourne Marketing Brief / #025 / NEWS & ANALYSIS
A new Claude model and an Ads API deadline: check the work behind the tools
Published by Ajay Dabhi · · AI-assisted research and writing
8 October 2026 · NEWS & ANALYSIS · New model announcement and previously announced API deadline
Anthropic has introduced Claude Haiku 5.5 for narrow, high-volume AI work. Separately, Google’s scheduled retirement date for Ads API v22 has passed. For a service business, the useful questions are different: is a new model reliable enough for a specific task, and are the existing advertising connections still returning current data?
The model announcement is dated 7 October. Google’s reminder was published on 2 September and names 7 October as the sunset date. We are reporting that scheduled milestone, not presenting the older reminder as a new announcement. Both sources were checked on 8 October in Melbourne.
No customer account, API request or model performance was tested. The source claims below are separate from our analysis and hypothetical examples. An announcement from an AI provider does not establish availability in Growthcenter or any other business application.
TEST / INTERNAL WORKFLOWS
1. Claude Haiku 5.5 adds another option for narrow AI tasks
Announced 7 October 2026 · Vendor availability claim · Application access unverified
What changed
Anthropic positions Haiku 5.5 for high-volume tasks such as classification, summaries and database queries. It adds an adjustable effort setting and says the model is available through the Claude Platform and named cloud providers.
The same announcement reduces Sonnet 5.5 cache-read pricing from US$0.20 to US$0.10 per million tokens. Its computer-use and browser-use additions to the Python and TypeScript SDKs are labelled beta. Model availability and beta SDK support are different claims.
Anthropic’s benchmark results and average cost-saving estimates are vendor claims. They do not demonstrate the quality, total cost or Australian account availability of a particular business workflow.
Published 7 October 2026. Source checked 8 October 2026.
Who this affects
Australian businesses that already have someone maintaining a custom AI workflow, or are evaluating one. A business using a packaged receptionist should ask its supplier what models and settings are actually available. If your current system meets its needs and there is no specific problem to solve, there is no urgent reason to switch.
Our analysis: what it means for your business
Our recommendation is to test a small job whose answer a person can check. Sorting an enquiry into an approved service category is easier to evaluate than allowing an agent to interpret a complaint, promise a refund and alter a booking in one step. Keep the scope of the trial small enough to understand each mistake.
Hypothetical example: a cleaning business wants internal enquiry labels for regular cleaning, end-of-lease cleaning and work it does not offer. Build test examples that include vague wording, multiple requests and an unfamiliar suburb. Compare the proposed category with the team’s agreed answer, and keep uncertain cases for a person. This is a proposed evaluation, not a claim about Haiku’s measured accuracy.
A cheaper token rate is only one part of operating cost. For a useful comparison, record how often staff must correct an answer, how long review takes and whether the workflow has to retry. A small apparent saving can be unhelpful if it creates extra admin or directs an enquiry to the wrong person.
Do not confuse a model upgrade with permission to act. An internal draft or classification can remain subject to review. Access to calendars, refunds, customer messages or confidential records should be a separate decision with a defined purpose. A browser-use beta is not a reason to give an agent unrestricted access to business accounts.
What to do next
- Choose one internal task and write down what a correct result looks like before comparing models.
- Use synthetic or appropriately de-identified examples, including ambiguous requests and cases that should go to a person.
- Compare quality, delay, review effort and actual usage costs with the current workflow. Record the model and settings used.
- Confirm application access, data handling and permissions with the supplier. Keep customer-facing actions out of the initial trial.
CHECK NOW / DATA FRESHNESS
2. Check advertising connections after the scheduled v22 retirement
Reminder published 2 September 2026 · Scheduled sunset 7 October · No account failure independently tested
What changed
Google’s Ads API team said v22 requests would begin failing from 7 October 2026 and instructed developers to move to a newer version. That scheduled date has now passed.
Google’s general lifecycle page still lists October 2026 as tentative for v22; the specific dated reminder supplies 7 October. We have not tested a v22 request or verified a failure in any customer’s integration.
Google explains that recent method names in Cloud Console’s Google Ads API metrics identify the API version used. Its lifecycle guidance distinguishes a deprecated version, which can still work, from a sunset version, whose requests fail.
Published 2 September 2026. Source checked 8 October 2026.
Publication date not provided by the source. Source checked 8 October 2026.
Who this affects
Businesses using custom Google Ads reporting, conversion uploads or other software connected through the API, including connections maintained by an agency or software vendor. A business working only in the Google Ads interface has no custom API code to migrate itself; any connected third-party tools should be checked with their owner.
Our analysis: what it means for your business
The first check is whether the information is current. A dashboard can look normal while displaying the last successful import. A blank or unchanged report should prompt a connection check before someone concludes that demand has disappeared or changes campaign budgets. This is a possible failure pattern, not an observed outage.
Hypothetical example: a plumbing owner receives a daily advertising report that stops updating after the scheduled sunset. Ask the report owner for the latest successful refresh and the API version, then compare the same date range with the source account. Do not label the discrepancy as a campaign performance decline until the data connection is understood.
A conversion connection needs a different check from a read-only report. Ask whether records are waiting, failed or already accepted, and how the integration prevents duplicate uploads. Restoring a connection and blindly replaying everything are separate actions; the responsible developer should reconcile the queue before retrying records.
This deadline concerns an API version. It is not evidence that advertising campaigns themselves stopped serving, that every connected product is affected or that an account needs a different bidding strategy. If your provider confirms a supported version and recent successful operations, document that and move on.
What to do next
- List the external reports and conversion connections you rely on, with a named person or supplier responsible for each.
- Ask which API version each connection uses and request evidence of a recent successful operation after the scheduled sunset.
- If a connection fails, have its maintainer inspect logs and upgrade the integration using Google’s current migration guidance. Do not guess from a blank chart.
- Reconcile missed reporting periods or queued conversion records before replaying them. Preserve the audit trail and verify the next scheduled run.
ANALYSIS / BUSINESS CONTEXT
Where to spend your attention today
Give priority to an existing connection that may have stopped delivering useful data. A new AI model can wait for a controlled comparison; a stale report can mislead a decision today. That ordering is our operational recommendation, not an additional rule from either platform.
For both items, ask for evidence from the actual workflow: the last successful data refresh, the version in use, the test example and the person responsible for reviewing the result. Product announcements describe capabilities and dates; those checks establish whether a specific business can rely on them.
Practical checks
- Check the owner, API version and last successful refresh of important Google Ads connections.
- Keep AI trials narrow, internal and reviewable before enabling customer-facing actions.
- Judge the complete workflow using correct outcomes and staff effort, rather than treating a lower model price as a promised return.
These checks are recommendations, not additional requirements announced by the platform. The appropriate work depends on your actual configuration.
Publisher and editorial information
Ajay Dabhi publishes this report and also sells marketing and automation services. The commercial invitation above is separate from the sourced reporting. No client account test or direct interview is claimed in this report.
Research and writing use AI assistance. Original documents are linked so readers can check the facts. Read our ownership, sourcing and corrections policy or report a correction to Ajay.
All Melbourne Marketing Brief editions →