Melbourne Marketing Brief / #016 / NEWS & ANALYSIS
New invoice layouts, Claude Sonnet and a closer look at website speed
Published by Ajay Dabhi · · AI-assisted research and writing
29 September 2026 · NEWS & ANALYSIS · Labs and account limits identified
Three new releases offer different kinds of help: clearer invoices, another model for everyday AI work, and better context for website performance. Start with the problem your business actually has before changing a working system.
This edition reports announcements dated 28 September and checked on 29 September, Melbourne time. The recommendations and hypothetical examples below are original analysis, separate from the vendors’ factual announcements.
Growthcenter’s invoice item comes from HighLevel, its underlying platform. Access in a particular Growthcenter account has not been verified. No customer workflow, model or website was tested for this report.
CHECK / INVOICING
1. Growthcenter’s underlying platform adds invoice layouts in Labs
Confirmed announcement · Labs feature · Account access unverified
What changed
HighLevel announced an AI invoice layout builder on 28 September. Users can describe a design, adjust it manually and select a published layout for an invoice. Its instructions require activation under Settings → Labs → Sub-Account.
The layout locks after an invoice is sent or partially paid. Deleting a layout leaves previously sent invoices on their original design, while affected draft invoices switch to the default layout. These are layout behaviours, not a claim that invoice amounts or payment arrangements change.
Published 28 September 2026. Source checked 29 September 2026.
Who this affects
Australian service businesses already using Growthcenter invoicing and the person responsible for their billing process. If invoices are issued entirely through another accounting system, there is no reason to change that process for this release.
Our analysis: what it means for your business
A clearer invoice can make the agreed job, amount and next step easier to understand. That is a design objective, not evidence that this feature makes customers pay sooner or produces compliant invoices automatically.
Hypothetical example: a cleaning business wants the service address and agreed scope to remain easy to find on a phone. A sample layout could be compared with its current invoice before any customer sees it. Check a long description as well as a short one; an attractive empty template tells you little about a real bill.
My recommendation is to ask the billing owner to check the finished output, including totals, business identity and payment instructions. Keep the current layout available. Because this is in Labs, limit any approved trial to one account instead of enabling it across every business.
What to do next
- Ask the provider whether this Labs feature is available in the relevant Growthcenter account and whether any account conditions apply. Availability has not been checked here.
- Use clearly labelled sample details, preview the customer view and download, and confirm the important information remains readable. Do not send the sample to a customer.
- If the billing owner approves a trial, record the selected layout and check affected drafts after any later layout deletion. Keep changes to design separate from changes to prices or payment terms.
TEST / ONE DEFINED TASK
2. Claude Sonnet 5.5 offers another option for routine AI work
Confirmed release · Vendor performance claims · Integration access varies
What changed
Anthropic released Claude Sonnet 5.5 on 28 September. It describes the model as available across its platforms and major cloud providers, with claude-sonnet-5-5 as the Claude Platform identifier. That announcement does not confirm access inside Growthcenter or another third-party product.
Anthropic says output is faster and tasks can cost less than with Sonnet 5, although the per-token prices are unchanged. Those are vendor findings, not measured savings for an Australian business. The release also says integrations using thinking off must move to the between_tools setting before switching.
Published 28 September 2026. Source checked 29 September 2026.
Who this affects
Business owners or maintainers already using Claude for document work or a custom AI workflow. Businesses with a managed receptionist should ask their provider about supported models; this release alone is not a receptionist upgrade or a reason to replace a functioning system.
Our analysis: what it means for your business
The useful comparison is whether a completed task needs less correction, not whether the first response arrives sooner. An answer that omits a service exclusion can be fast and still create extra work. API usage is only one part of the cost when a person must review the result.
Hypothetical example: an electrical business prepares a draft reply from an approved service sheet. Give the existing model and the proposed replacement the same fictional enquiry, including an out-of-area address. Check whether each keeps the service boundary and leaves unknown availability unconfirmed. This is a proposed evaluation, not a customer result.
I would start with internal drafts that a person reviews. Compare factual errors, correction time, response delay and total usage for the same task. Retain the working version until the replacement meets the business’s own requirements. A coding benchmark does not establish appointment accuracy or lead quality.
What to do next
- Confirm the model is supported by the actual provider, account and region. Have the maintainer review the documented migration requirement before changing a custom integration.
- Use a small, repeatable set of fictional or approved non-sensitive examples. Include incomplete details and requests outside the business’s scope.
- Keep sending, bookings and account changes under the existing approval rules. Expand only after reviewing complete task outcomes; no improvement is guaranteed.
INVESTIGATE / MOBILE EXPERIENCE
3. Cloudflare opens a public view of real website performance
Confirmed dataset release · Broad context, not a site-specific audit
What changed
Cloudflare published BEACON on 28 September: anonymised real-user performance data from 10,000 large websites on its network, updated daily in Google BigQuery. It includes loading, visual stability and interaction measures, with country and browser dimensions.
Cloudflare removes website domain names and URL paths from the public data. The dataset therefore does not provide an identifiable performance report for your own site. Additional performance breakdowns in its customer RUM dashboard are described as coming in the following weeks, not as already available there.
Published 28 September 2026. Source checked 29 September 2026.
Who this affects
Website maintainers investigating mobile performance and business owners discussing a slow enquiry or booking journey. Owners do not need to learn BigQuery to act. If there is no performance concern, this announcement creates no urgent requirement to rebuild the site.
Our analysis: what it means for your business
The data offers context for a technical investigation; the business still needs evidence from its own pages and visitors. A sample of large websites is not a representative audit of Melbourne clinics, trades or professional practices, and an industry pattern cannot establish why one form feels slow.
Hypothetical example: a homeowner opens a plumbing landing page on a modest phone and taps the call or quote action. The practical question is whether the page appears, stays stable and responds. A fast result on the owner’s office laptop does not answer that question.
My recommendation is to ask the maintainer for a short explanation of the observed bottleneck before approving a fix. Separate a slow first load from an unresponsive button. Use the new dataset only where it helps that investigation; it is not proof of more bookings, better rankings or a need to change hosting.
What to do next
- Try the main enquiry journey on a phone and an ordinary connection, using a labelled test only if a submission is necessary and authorised. Record the page, device and symptom.
- Ask the maintainer to compare those observations with the site’s own performance evidence. Select a specific problem to investigate rather than pursuing a score without context.
- If the maintainer queries BEACON, confirm query costs and sample limitations first. Public dataset access does not mean every cloud query is free.
ANALYSIS / BUSINESS CONTEXT
Where to spend your attention today
For most owners, the order is simple: check a tool you already use, test one clear problem, then decide whether a change is worth the effort. Invoice presentation, AI task quality and mobile response are separate decisions; none needs a wholesale system replacement.
Keep the person responsible for billing, customer replies or the website involved. That person can define what a correct result looks like and what needs human review. The proposed checks here have not been run on client accounts.
Practical checks
- If Growthcenter issues your invoices, check Labs availability and the sample output with the billing owner.
- If Claude already supports a business task, compare one controlled draft before considering a model change.
- If visitors report a slow website, collect a reproducible mobile example and ask for a diagnosis before buying a fix.
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 →