A useful proposal should make the work easier to understand. These questions help you compare suppliers and see what you would actually be buying.
Before you choose a supplier
Choose an AI implementation partner by how well they explain your workflow, the limits of their proposal and how you will accept the work. Ask for a scoped process, relevant evidence, failure handling, ongoing costs and a usable handover. A polished demo is the start of that assessment.
Give each supplier the same brief
Comparing proposals is difficult when one supplier prices a chatbot, another prices an integration and a third assumes a full system replacement. Describe the current task, the people involved, your tools and the point where work stalls before asking for a solution.
A useful brief might say: “Our office copies website enquiries into the job system. We want complete requests assigned once, with missing details visible and a way to recover failed transfers.” That gives suppliers something concrete to respond to.
Include the rough frequency of the task, the outcome you want to improve and any constraints. Use the readiness checklist and downloadable worksheet to prepare a common brief.
Seven questions that make proposals easier to compare
| Ask the supplier | A useful answer includes |
|---|---|
| What problem are you solving first? | A specific start and finish, plus explicit exclusions |
| Why does this need AI or custom work? | What existing features and simpler rules were considered, and where they fall short |
| Can you show relevant work? | The supplier’s actual role, what was delivered and which results are measured versus illustrative |
| What happens when it is wrong or unavailable? | Human review, alerts, recovery and a way to pause the workflow |
| How do we accept the build? | Written examples of expected behaviour, including exceptions, tested in your agreed environment |
| What will we pay after launch? | Licences, usage, maintenance, support boundaries and what triggers additional work |
| What do we control if we leave? | Agreed access to accounts, data exports, documentation and any custom code or configuration |
Write “unanswered” beside a missing detail. Do not fill the gap with an assumption from the sales conversation. Ask the supplier to put material commitments in the scope or agreement.
Ask the demo to handle an exception
A happy-path demonstration shows what can happen when the inputs are complete and the tools are available. Ask to see the same workflow with a missing field, an ambiguous request or a temporarily unavailable destination.
For an enquiry handover, check whether the system preserves the customer’s wording, distinguishes a requested appointment from a confirmed booking and leaves a clear action for a person. You can try those distinctions in our illustrative plumbing example.
If a supplier presents a case study, ask which parts they built and how the result was measured. A project from another industry can show a relevant capability; it does not establish the same commercial outcome for your business.
Understand where your information goes
Ask the supplier to identify the tools that will receive information, the access each needs, retention settings and who can change or remove that access. Use de-identified examples for early demonstrations and arrange an appropriate channel before sharing live customer records.
The OAIC’s guidance on commercially available AI products recommends assessing suitability and privacy risks before adoption. Use it to frame questions about data handling and oversight; the obligations relevant to your business need their own assessment.
“Secure” is not a complete description of the arrangement. Ask what happens in this particular workflow: which fields leave your system, whether a supplier can use them for other purposes, and how an incorrect or exposed record would be handled.
Compare ongoing costs and the handover
Request a list of recurring accounts and who owns and pays for them. Ask what happens when usage grows, an API changes, the supplier is unavailable or you decide to stop. Distinguish fixing a defect from changing the agreed scope.
A workable handover should explain how your team operates the system, reviews exceptions, changes approved settings and gets help. Clarify ownership and licence terms in the agreement; custom work does not automatically mean every component transfers to you.
business.gov.au’s digital tools guidance recommends considering the total cost of ownership. Our implementation budgeting guide provides a structure for comparing setup and ongoing costs.
Make the decision against the first workflow
Compare scope, evidence, acceptance criteria, data handling and support side by side. Treat unresolved requirements as follow-up questions, rather than inventing a numerical score that makes unlike proposals look equivalent.
Before committing to a wider rollout, agree what a first implementation must demonstrate and what happens if it does not. That may mean refining the approach or deciding a simpler change is enough.
AppliedAgents builds tailored AI systems, automations and business software. Our workflow automation approach is to understand the bottleneck and existing tools before recommending a build. These are the same questions you should ask us.
Practical thinking on AI, automation and the work behind a business.
