Skip to main content

We use cookies to improve your experience and measure traffic. Decline to opt out of analytics and advertising cookies. Cookie preferences

For marketing agencies, managed IT providers and web studios

Your Clients Keep Asking for AI. Here Is What to Keep and What to Hand Over.

If you run an agency or an IT practice in Australia, the AI questions have stopped being occasional. Clients want a receptionist that answers the phone, a bot that knows their product range, a workflow that stops someone retyping invoices. Some of that you should absolutely do yourself. Some of it will eat a year and a good client.

This page is a straight sorting exercise: the jobs worth keeping in house because they sit close to what you already do, the jobs worth handing over because the failure modes are brutal and the support load never ends, and the honest test for telling them apart. It also explains why we run a referral programme rather than a white label one, since white label is usually what people are searching for when they land here.

Realistic ROI

20%
Of the setup fee, if you refer rather than build
Paid within fourteen days of your client’s setup invoice clearing, on the amount confirmed to you in writing before they signed
No white label
Not at any tier, not at any volume
The reason is on this page and it is about support and expectations, not about protecting a brand
No business hours
Is the support reality once a voice agent is live
A phone line that answers customers does not stop at five, and neither does the complaint when it misbehaves
The client stays yours
In writing, not as a handshake
No approaching your other clients, no outbound lists, no pitching around you to a business you introduced

Four Things Agencies and MSPs Get Wrong About AI Work

None of these are about capability. They are about which risks travel with the job once it is live.

The build is the cheap part, the support is the commitment

Standing up a demo that impresses a client in a meeting is genuinely easy now, and that is exactly what leads people astray. The cost lands later: the model behaves differently after a provider update, an integration authentication token expires at midnight, a customer says something the flow never anticipated, and the business rings you rather than the vendor. If you take on AI delivery you are taking on a support obligation with no natural end date, and it needs to be priced, staffed and rostered before the first client goes live, not after.

White label puts the risk in the wrong place

We are asked for white label regularly and we decline it every time, so it is fair to say why plainly. The partner sets expectations we were never part of, quotes a timeline we never saw, and then hands over the support load while keeping the relationship. When something breaks, the person the client rings cannot see what happened, and the person who can see it is not allowed to speak to the client. That is a bad architecture for a bad afternoon. It also means every scope misunderstanding surfaces as a fight between two suppliers rather than a conversation with the client.

Anything touching money, health or a phone line raises the stakes

A blog draft that reads oddly is embarrassing. An automation that pays a supplier twice, an intake flow that mishandles patient information, or a voice agent that tells a caller the wrong thing about a warranty is a different category of problem entirely. Those jobs need error handling, an audit trail, a tested fallback to a human, and someone who has seen them fail before. If your practice does not already carry that kind of risk, taking it on for an AI project is an odd place to start.

The honest test is who gets rung at 4pm on a Friday

For any AI request, ask one question: when this misbehaves in front of a customer, who does the client ring, and can that person actually fix it? If the answer is you and you genuinely can, keep the job. If the answer is you and you would then have to ring someone else, you have accidentally become a switchboard with liability attached. That single test sorts almost every AI request into keep or hand over, and it does it faster than a capability assessment.

The Six Requests You Are Getting, and What to Do With Each

Sorted by how much of the risk stays with you after go live rather than by how hard the build looks.

Keep it

Content and campaign work

Drafting, repurposing, ad variations, briefs, first pass research and summarising client material. This sits inside what you already sell, the failure mode is a bad draft that a human catches, and nobody rings you at midnight about it. Build the habits properly: a house style guide the tools actually follow, a review step nobody skips, and an honest position with clients about what was machine drafted. This is the highest margin AI work most agencies will ever do and it needs no outside help.

Keep it, or co deliver

Team enablement and training

Getting a client’s staff genuinely using the assistant tools they already pay for, building shared prompt libraries, running a workshop that turns curiosity into a routine. Very deliverable by a good agency or IT provider, and it builds trust that leads to bigger work. Where it gets away from people is when a client asks for the tools to be connected to their internal documents and systems, which is a different job with different security questions and belongs in the next category.

Keep the simple version

Website chat and lead capture

A chat widget that answers questions from published content on a site you built and hands off to a form is well within reach, especially if you already maintain the site. The line to watch is when it needs live data: stock levels, booking availability, account balances, order status. At that point it stops being a website feature and becomes an integration with an uptime expectation, and the person who owns the site is suddenly on the hook for a system they do not administer.

Hand it over

Voice agents on a real phone number

Telephony is unforgiving in a way web work is not. Number porting, carrier behaviour, latency, interruption handling, accents, background noise, call recording obligations that differ by state, and an escalation path that has to work when the agent gets out of its depth. It runs at three in the morning and misbehaviour is audible to a paying customer. This is the category we most often see agencies regret taking on, usually about six weeks after the demo went well.

Hand it over

System to system integration

Connecting an accounting system to a job management tool, a store to an ERP, a booking system to a practice management package. The build might take a fortnight. The maintenance is permanent, because vendors change interfaces, tokens expire, someone renames a field, and a silent failure that nobody notices for a week is worse than an outage. This needs monitoring, alerting and a person who will look at a failed queue on the day, which is an operations function rather than a project.

Hand it over

Custom assistants over private data

A client wanting an assistant that knows their contracts, their policies or their product documentation. Getting a passable answer from a pile of documents is a weekend. Getting an answer that is right, cites its source, refuses when it does not know, respects who is allowed to see which document, and stays current as the documents change is a proper engagement. The demo to production gap is wider here than anywhere else in AI, and clients rarely believe that until they are in it.

Keep or Hand Over: The Quick Sort

TaskTraditionalOur honest verdictNotes
AI assisted content productionPrompting, style guide, human reviewKeep itClosest to what you already sell, and the failure mode is caught before a customer sees it.
Staff training on assistant toolsA workshop and a prompt libraryKeep itHigh trust, low risk, and it reliably leads to the client asking for the bigger work.
Chat widget over published website contentConfiguration plus content curationKeep itStays yours as long as it never needs live data from a system you do not administer.
Chat that quotes stock, pricing or availabilityLive integration plus caching plus error statesHand it overThe moment it reads live data it inherits that system’s uptime and its edge cases.
AI receptionist on the client’s main numberTelephony, porting, escalation, recording lawHand it overRuns overnight, fails audibly in front of customers, and support has no business hours.
Accounting or ERP integrationMapping, monitoring, permanent maintenanceHand it overA silent failure noticed a week late is worse than a visible outage, and it touches money.
Assistant over contracts, policies or product docsRetrieval, permissions, citation, freshnessHand it overThe gap between a convincing demo and something defensible is wider here than anywhere.
Anything in health, credit or safetyAudit trail, human fallback, professional riskHand it overCheck your professional indemnity cover before you even scope it, not after.

What Goes Wrong When an Agency Takes AI Work In House

The support tail is never priced in

Project pricing assumes the work ends. AI work does not: providers change model behaviour, integrations break when a vendor updates an interface, and clients keep asking for one more small thing. Agencies routinely win a project fee and then absorb months of unpaid attention because there was no support agreement and no monitoring, only goodwill. If you are keeping a job, sell a support arrangement alongside it from day one, even a small one, because it is far harder to introduce after the client has had six months of free help.

One person in the team becomes the single point of failure

AI capability inside a small agency usually lives in one enthusiastic person. When they are on leave, or they resign, the client is still live and nobody else can safely touch the configuration. Before you take on delivery, decide who the second person is and make them build something real, not just watch. If you cannot name a second person, you are not resourced to deliver it and referring the work is the responsible answer rather than the timid one.

The demo sets an expectation the production system cannot meet

A demo built on ten clean documents and a friendly script looks close to finished. Production brings ambiguous questions, out of date documents, permissions, edge cases and users who type in shorthand. If a client has seen the demo, the remaining work reads to them as delay rather than as the majority of the job. Either show something deliberately rough, or be very explicit at the demo about what is still ahead, because that expectation gap has ended more agency AI projects than any technical problem.

Professional indemnity cover may not follow you into this work

Your policy was written for the services you described when you took it out. Automated decisions affecting money, health information, credit or safety may sit outside what is covered, and an exclusion is not something you want to discover during a claim. Ring your broker with a concrete description of the job before you quote it. This is a five minute call that occasionally saves an entire business, and it is one of the strongest practical arguments for handing certain categories over.

White labelling other people’s work multiplies the problem

Taking a job you cannot support and quietly subcontracting it looks like the safe middle path and is usually the worst of both. You carry the client relationship and the contractual liability, you have no visibility into what was actually built, and the person who can fix it is not allowed to talk to the client. Every scope disagreement becomes a slow relay between three parties while the client waits. If the work is beyond what you can support directly, an open referral costs you far less than a badge deal.

Referring away the wrong job weakens your position too

This cuts both ways. If you hand over the content and enablement work because it feels technical, you have given away the part that was closest to your existing service, easiest to support and highest margin, and kept nothing. Hand over the categories where the risk lives after go live, and keep the ones where a mistake is caught in review. An honest partner should be telling you to keep those, and if a supplier is happy to take everything you offer, that is worth noticing.

What Working With Us Actually Looks Like

A straight keep or hand over answer

Send us the request your client made and we will tell you which side of the line it falls on, including when the answer is that you should keep it and we should stay out. We are not trying to hoover up every AI job in your client base, because the ones you can support yourself are better off with you.

Twenty percent of the setup fee if we deliver it

Confirmed to you in writing at proposal stage before your client signs, paid within fourteen days of their setup invoice clearing. Flat rate, no volume tiers, no status ladder, and the same rate on your first referral as your tenth.

We stay in the lane you set

No approaching your other clients, no adding the referred business to outbound lists, no pitching services around you. If your client asks us about something new, you hear it from us first, and if it sits inside what you do we send them back to you.

A firm no on white label, said early

If you need your own badge on the delivery, we are not your supplier and we would rather tell you that in the first twenty minutes than three months into a client engagement. Plenty of Australian firms do offer white label. We think it serves neither side well, so we do not.

From First Client Request to Delivered Work

Five steps. Most partners test us with a single client before deciding whether this is a habit.

Send us the request, not the client

Before any introduction happens, tell us what your client asked for. We give you a keep or hand over verdict and the reasoning, which is useful even if nothing else follows and costs you nothing.

Agree how you want to appear

Copied on everything, summaries only, or hands off after the introduction. We also agree what we say about where the referral came from, so your client hears the same story from both of us.

One page agreement

Referral rate, payment timing, the commitment not to sell around you, and how either side ends it. No exclusivity, no minimum volume, no term to serve out. Signed by email.

We scope and propose

A real conversation with your client about what they need, a fixed price proposal with defined scope, and your referral amount confirmed to you in writing before they sign anything.

Delivery, support and payment

We build it, we support it, and the client keeps dealing with you for everything that was already yours. Your fee is paid within fourteen days of their setup invoice clearing.

FAQ

Send Us One Request and We Will Tell You Which Side of the Line It Sits On

Book a short call. Bring the actual thing a client asked you for. If you should keep it, we will say so and there is nothing in it for us.

All discussions held in confidence. Australian-based consultants.