What does each party usually do?
A managed service provider (MSP) typically looks after accounts, devices, networks, backups, security and software licences, and has privileged access to your systems to do it. An AI consultant works on the business side of the project: which processes to change, whether AI is the right tool, what success looks like, how risks are managed and what to buy or build. A builder or integrator connects systems and writes the automation. Some firms play more than one of these roles, so ask which role they are playing on your project. Your own team owns the process, the data and the decisions.
When should your IT provider lead?
Your IT provider is usually the right lead when AI arrives as a feature of software they already manage. The National AI Centre suggests most organisations start with AI built into existing software, then standalone tools, and only later move to integrations or custom solutions. The first two often sit within an MSP's existing work, with examples below. Even here, someone in the business should decide which staff get access and what information may be used, because those are business decisions.
- Enabling Copilot Chat or a paid Copilot licence and assigning it.
- Reviewing SharePoint, Teams and OneDrive permissions before rollout.
- Controlling which third-party apps staff can connect to company accounts.
- Applying security controls and logging to AI tools.
When does independent advice help?
Independent advice is most useful when the open question is what to do. That includes choosing between processes, comparing products from different vendors, testing whether a simpler fix would do, building a business case and setting governance. It also helps when your IT provider resells a particular product, because a recommendation from someone who earns a margin on the licence is harder to weigh. Ask any adviser, including us, whether they receive commissions or referral fees from the vendors they recommend, and get the answer in writing.
Who should be accountable for the outcome?
A person in your business should be. The National AI Centre's foundations guidance asks organisations to assign a senior leader as the overall AI governance owner and, as use matures, a specific accountable person for every AI system. It also asks organisations to clarify supply chain accountabilities when several parties are involved, such as vendors, developers and integrators, so you know who to talk to if something goes wrong. Suppliers can recommend and implement, but the business owns the decision to use AI in its processes and the results that follow.
How do you split the work without gaps?
Gaps appear at the handoffs: the consultant designs a workflow that assumes access the MSP has not granted, or the MSP switches on a feature nobody decided how to use. Write a one-page split before work starts, covering the points below, and have every party confirm it. Keep it with the AI register entry for the project. If a supplier changes during the project, update the split on the same day.
- Who decides which process and use case to pursue.
- Who sets up accounts, licences and permissions, and who approves them.
- Who builds integrations and who maintains them afterwards.
- Who monitors the system after launch and who is called when it fails.
- Who holds administrator credentials, and how they pass to the business at the end.
What should you ask each party before engaging them?
Ask your IT provider the security questions the Australian Signals Directorate publishes for managed service providers: whether they implement the Essential Eight, how they securely administer client systems, whether they monitor activity, whether they regularly assess their systems and whether they are prepared to respond to incidents. Then ask each party the AI-specific questions below. The answers show which work each can do well and where you need someone else.
- Which AI projects like ours have you delivered, and can we speak to that client?
- Do you resell, or receive referral fees for, the products you would recommend?
- Who on your team will do the work, and what is their AI experience?
- What will we own at the end: documents, configurations, code and credentials?