Insights · AI Consulting

How to compare AI consulting proposals.

Ask three firms for a proposal and you get three documents that are hard to put side by side: one fixed price, one discovery phase, one rate card. How to read them against each other, eight things to check in each, the warning signs, a simple way to score them, and the Australian terms that belong in the contract.

Once you have sent a brief to two or three AI consulting firms, the responses come back in different shapes. One proposes a fixed price for a working system. Another offers a paid discovery phase and promises a price afterwards. A third sends a rate card and an estimate of days. Each is reasonable on its own terms, which is exactly what makes them hard to compare.

This piece sets out how to put AI consulting proposals on the same footing, eight things to check in each one, the warning signs that should slow you down, how to score them, and the Australian terms worth settling before you sign. If you have not written the brief yet, start with our guide on how to write an AI consulting brief, because a clear brief is what makes the responses comparable at all.

Put them on the same footing first.

Before you score anything, restate each proposal in the same four lines. It takes an hour and removes most of the confusion.

  • What will exist at the end. A working system in production, a pilot, a report or a recommendation. These are different products at different prices, and proposals often blur them.
  • Who will do the work. Names, seniority and the number of days each person spends on it, not only the people who came to the pitch.
  • What it costs in total. Fees, plus any software, model usage or cloud costs you will pay directly, plus the cost of running the system for its first year.
  • What you need to provide. Access to data, time from the people who know the process, and decisions from a named owner on your side.

Once every proposal fits those four lines, the differences that matter are usually obvious. A cheaper proposal that ends in a report is not cheaper than a dearer one that ends in a working system. Our piece on what AI consulting costs in Australia sets out the usual price ranges for each kind of work.

Eight things to check in every proposal.

  • The problem, in your words. Does the proposal restate your problem accurately, including the awkward parts? A response that could have been sent to anyone was probably sent to everyone.
  • A definition of done. A measurable outcome, such as the share of cases handled without rework or the hours returned to a team, and how it will be measured. Our guide to measuring an AI consulting engagement covers what good measures look like.
  • The data plan. Where the data comes from, who prepares it, and what happens if it turns out to be worse than expected. Data is where most AI work runs late, so a proposal that does not mention it has not priced it.
  • The named team. Who does the work each week and how much of their time you get. Ask whether any of it will be subcontracted, and to whom.
  • Where your data will be processed. Which models and services the firm intends to use, where they are hosted, and whether any personal information leaves Australia. Our piece on where your AI data can live covers the options.
  • How quality will be tested. An evaluation set of real cases with agreed correct answers, built early and run again on every change. Without one, nobody can tell you whether the system is working.
  • What you own at the end. Code, prompts, evaluation sets, configuration and documentation, and whether you can run and change the system without the firm.
  • What happens after launch. Who supports the system, at what cost, and how that cost should fall over time. Our piece on what AI consulting should look like after launch sets out what ongoing support should cover.

Warning signs.

None of these rules a firm out on its own. Each one is a reason to ask a harder question before you go further.

  • A price with no assumptions. A fixed price that does not say what it assumes about your data, your systems and your people will be revisited once the work starts.
  • A demonstration in place of a plan. A polished demonstration on sample data shows the firm can build a demonstration. It says little about your data or your processes.
  • No mention of risk. AI work has known failure points: poor data, unclear ownership, staff who do not trust the output. A proposal that names none of them has either not looked or is not saying.
  • Seniority that disappears. Senior people at the pitch, and a delivery plan staffed mostly by people you have not met.
  • Everything is proprietary. If the system only runs on the firm's own platform, ask what it costs to leave before you agree what it costs to start.

How to score them.

A simple weighted scorecard is enough for most organisations. Agree the weights before the proposals arrive, so the scoring is not shaped by whichever response you liked first. A reasonable starting point:

  • Understanding of the problem and the definition of done: 25 per cent.
  • The data plan and the approach to testing quality: 25 per cent.
  • The named team and their relevant experience: 20 per cent.
  • Total cost, including the first year of running the system: 20 per cent.
  • Ownership, exit terms and support after launch: 10 per cent.

Then speak to referees for the two firms that score highest, and ask each referee one question above all others: did the system keep working after the firm left?

Australian terms to settle before you sign.

  • Personal information. If the work touches personal information, the contract should say how the firm will handle it under the Privacy Act, including where it is stored and whether any of it is disclosed overseas, which Australian Privacy Principle 8 governs.
  • Automated decisions from 10 December 2026. Privacy policies must describe certain decisions made with the help of computer programs. If the system will help make decisions about people, agree who drafts that disclosure. Our piece on AI agents and the December privacy rule sets out what is covered.
  • CPS 230 for APRA regulated entities. Banks, insurers and super funds must manage their material service providers. If the system will support a critical operation, the contract may need to meet that standard.
  • Intellectual property. Agree in writing who owns what the firm builds for you, and what licence you hold over anything the firm already owned before the work began.

Why RUBIX.

RUBIX is an independent, Australian owned data and AI consultancy, founded in Melbourne in 2011. In 15 years we have delivered more than 450 projects for over 115 organisations, and answered a great many briefs along the way. We would rather be compared carefully than chosen quickly.

  • Proposals that name the work. Each one sets out what will exist at the end, the named people doing it, the assumptions behind the price, and what you will own.
  • Data first. We began as a data consultancy, so the data plan is where our proposals start, not a line near the end.
  • Built to be handed over. We build the evaluation set, the monitoring and the runbooks as part of the system, and our forward deployed engineers work inside your team, so the knowledge stays with you.

If you are not yet sure what kind of firm you need, our guide to which kind of AI consulting firm you need is a useful place to start. If you would like a proposal from RUBIX to compare against the others, talk to us.

Frequently asked questions.

How do you compare AI consulting proposals?

Restate each one in the same four lines: what will exist at the end, who will do the work, the total cost including the first year of running the system, and what you need to provide. Then score them against weights you agreed before the proposals arrived.

What should an AI consulting proposal include?

The problem in your words, a measurable definition of done, a data plan, a named team, where your data will be processed, how quality will be tested, what you will own at the end, and what happens after launch.

What are the warning signs in an AI consulting proposal?

A fixed price with no stated assumptions, a demonstration on sample data in place of a plan, no mention of risk, senior people at the pitch who are absent from the delivery plan, and a system that only runs on the firm's own platform.

Quick Summary: AI consulting proposals are hard to compare because they arrive in different shapes. Restate each in the same four lines: what will exist at the end, who will do the work, the total cost including the first year of running it, and what you must provide. Then check eight things: the problem, a definition of done, the data plan, the named team, where data is processed, how quality is tested, what you own and what happens after launch. Score against weights agreed in advance, and settle privacy, automated decision, CPS 230 and intellectual property terms before you sign.

General information only, not legal, financial or regulatory advice. Current as at November 2026.