What a 2–3 week AI Opportunity Sprint actually delivers

Most teams we speak to already believe AI could help somewhere in their business. What they don't have is a good answer to three questions: where exactly, how well would it work on our data, and what would it cost to run? Without those answers, AI projects either never start or start too big.

The AI Opportunity Sprint is how we answer them. It takes two to three weeks, has a fixed price between A$15k and A$30k depending on scope, and focuses on one workflow. At the end you have evidence, not a slide deck.

Who it's for

The Sprint suits a business that has a specific, repetitive, information-heavy workflow and wants to know whether AI can take real work out of it. Typical examples:

• A team that spends hours searching documents, specifications or past records for answers.
• Staff re-keying information from PDFs, emails or forms into another system.
• Support or operations people answering the same questions every day.
• A product team that wants to add an AI feature and needs to know it will hold up with real users.

It isn't the right starting point if the goal is "use AI somewhere" with no workflow in mind, or if the data it would need can't be shared with us at all. We'll tell you that on the first call.

What goes in

We ask for three things.

1. A workflow owner. Someone who does or runs the work day to day and can spend a few hours a week with us. They define what a good result looks like.
2. Real examples. A representative sample of the documents, tickets or records involved, shared under an NDA and processed where we've agreed they can be.
3. Access to the people and systems involved, so we can see how the work actually happens rather than how it's documented.

What happens across the two to three weeks

Week one: audit the workflow. We map how the work is done today: the steps, the hand-offs, where time goes and where mistakes happen. We look at the data: where it lives, how clean it is, and what we're allowed to do with it. We usually leave week one with a short list of places AI could help, and a few where it shouldn't be used.

Week two: prioritise and prototype. We rank the candidates by value against difficulty and risk, agree the top one with you, and build a working proof of concept on your real examples. Alongside it we write the first eval set. That's a scored set of test cases, so we can measure how well the prototype does against the success criteria your workflow owner set, instead of judging from a demo. (We've written separately about how we use evals.)

Week three, when the scope needs it: harden the answer. We extend the eval set to the awkward cases, estimate production running costs from real token and infrastructure use, and work out where people need to stay in the loop. Shorter Sprints fold this into week two.

What comes out

You get four things you keep whether or not you work with us afterwards:

• A prioritised opportunity list for the workflow, with the reasoning behind each ranking.
• A working proof of concept for the top opportunity, running on your data, in a repository you own.
• An eval scorecard showing how accurate the prototype is, where it fails, and what it costs and how fast it responds per request.
• A recommendation and a plan, including a production estimate, the integrations needed, where human review sits, and the risks. If the answer is "don't build this", that's what the recommendation says.

What it costs, and why it varies

Sprints are priced between A$15k and A$30k, fixed before we start. Where a Sprint lands in that range depends on:

• How many systems and data sources the workflow touches.
• How messy the data is. Scanned documents and inconsistent formats take more work than clean exports.
• How accurate the result needs to be. Workflows where an error has a cost need a larger, stricter eval set.
• Whether week three is needed to test the edge cases properly.

We agree which of these apply on the scoping call, and you get a fixed quote before anything starts.

What usually happens next

A Sprint ends with a decision, and there are three common outcomes.

1. Build it. The prototype meets the bar and the economics work. The next step is usually a Production Pilot: six to ten weeks, typically A$60k–150k at a fixed price, to ship that one capability to production with evals, monitoring and the integrations it needs.
2. Change the approach. The idea is sound but the first approach isn't good enough, or the data needs work first. The plan says what would need to change.
3. Stop. AI isn't the right tool for this workflow, or the running costs outweigh the time saved. That's a good result if it saves you a six-month build that wouldn't have paid back.

Teams that go on to build and keep improving several AI features often move to an Embedded Engineering Pod, a senior team of two to five people working through their roadmap month to month, from A$25k a month.

---

If there's a workflow you suspect AI could help with, the scoping call is where we find out whether a Sprint is the right next step, and what it would cost for your case.