What an AI assessment should deliver

The assessment should help you decide what to do

A formal assessment is useful when a business has several ideas but cannot yet choose a first project, identify the systems involved or explain the likely investment.

The consultant needs to understand the work. Where are people entering information twice? Which handovers involve waiting? Where do complaints or corrections keep occurring? Staff demonstrations and real forms, documents and system screens are useful evidence.

The review then checks whether records are complete, systems can be connected and someone can confirm how the process should work. If information is missing, the assessment should identify the preparation needed and its effect on cost and timing.

It should also establish who could be affected by an error or an action. A draft for a staff member to check needs different controls from a message sent automatically or a recommendation affecting someone’s employment or health. Record the intended use, the actions outside scope and the person responsible for reviewing important outputs.

Compare the recommendation with other approaches

Options can include existing software features, a process change, ordinary automation and AI assistance. Each should explain the work it covers, the information it needs and what staff will still do.

For example, slow enquiry handling may come from hard-to-find information or an unclear handover. The first problem may justify retrieval support. The second needs someone to take responsibility for the request. The same complaint about slow replies can lead to different recommendations.

An estimate of time saved should include current volume, time per task and the review and exception work expected after implementation. That gives the business a basis for comparing benefits with the project fee.

What you should receive

A useful assessment sets out:

• The priority problem and the scope of the first release.
• Required information, system access and preparation by the client.
• Information flows, permission to use the records, provider reuse, processing locations and retention arrangements.
• The recommended approach, alternatives and reasons for choosing it.
• Implementation budget, ongoing costs and staff involvement.
• Delivery sequence, responsibilities and acceptance criteria.
• Representative tests, human review and a review point after launch.

The depth of review follows the likely business impact, rather than the size of the software purchase. Where personal information is involved, define the privacy work early. The proposal should identify the required review, its owner and any specialist input, not leave those decisions until launch.

J2K’s indicative assessment fee is $3,500–5,000 NZD, excluding GST, for the agreed review and implementation plan. A business with a clear requirement and known access arrangements can also discuss implementation directly.

By the end of the assessment, you should be able to decide whether to proceed and know what must be ready before work begins.

References: Responsible AI Guidance for Businesses · MBIE, Privacy impact assessment guidance · Office of the Privacy Commissioner.