HOW WE WORK · THE QUANTANA PATHHow we work: no AI build until the use case is sized in dollars
Map, Solve, Prove, Grow. We find where AI cuts cost or grows revenue, work out the ROI of each use case, and prove it with real users before anything scales.
- 1MapFind where value is won or lostGate
- 2SolveExplore many answers before choosing oneGate
- 3ProvePut working prototypes in front of real usersGate
- 4GrowMeasure, improve and scale what worksGate
How does Quantana run an AI engagement?
Every engagement follows the Quantana Path: Map, Solve, Prove, Grow. We map where value is won or lost, explore many answers before choosing one, prove the best with real users, then measure and scale what works. Each stage ends in a decision gate, and every use case is sized in dollars before anything is built.
“Can AI help us?” is the wrong first question.
The better one: which process, which number, and how much better by the end of the quarter?
Everything below is built around answering that before anyone writes code.
What happens at each stage?
| Map | Solve | Prove | Grow | |
|---|---|---|---|---|
| Purpose | Find where value is won or lost | Explore many answers before choosing one | Put working prototypes in front of real users | Measure, improve and scale what works |
| You bring | Business goals; time with process owners; current cost, time and quality figures; a list of data, systems and AI tools already in use | Business owners, data lead, risk and compliance | The business owner, real users and a risk reviewer | The business owner, platform and support teams |
| You get | Process and workflow maps; data architecture readiness; an ROI feasibility matrix; a value range per use case; your AI roadmap | Options for each problem statement; where each model runs (hosted, private cloud or on-premises); data fixes; a risk screen; evaluation sets | A working pilot in the real workflow; evaluation results; measured change against the baseline; user feedback | Production release with monitoring, logging, a human override and a fallback; training; value tracking; the next wave |
| Timing | 14 days (AI Compass) | Agreed at the Map gate | Agreed at the Map gate | Ongoing, reviewed quarterly |
| Decision gate | Sponsor and budget confirmed; each chosen use case has a business owner, a baseline, a target and a written stop rule | Data passes quality checks; risk tier assigned; controls agreed with risk | Go to production only if the pilot meets its target and passes risk review; otherwise change scope or stop | Each quarter: keep, improve or retire each system on measured value and cost |
After Map, timing depends on your data readiness, scope and approvals, and is agreed at the Map gate. Regulated organisations should allow extra time for risk review. Our AI roadmap guide shows typical phase lengths.
Strategy, design, engineering and ISO-certified QA sit in one accountable delivery POD, so nothing is lost in a hand-off between consultants and builders.
Map: find where value is won or lost
Map is AI Compass.
We run user journey mapping and trace the critical paths through each process with the people who do the work, not just their managers.
We register every AI tool already in use, including AI inside software you already pay for, and check whether your data can support the use cases on the list. Most organisations overestimate their AI readiness, so this step often changes the order of the roadmap.
You leave with your AI roadmap in 14 days.
Solve: explore many answers before choosing one
S10 ideation workshops come before any build.
S10 is our ideation workshop: it generates several solutions for each problem statement before we commit to one.
For each data class we decide where the model should run: a hosted API, a private cloud tenancy, or on-premises hardware such as SecGPT Vault. We fix only the data the first use cases need, build an evaluation set from your real tasks, and agree controls with your risk team.
Prove: put working prototypes in front of real users
A pilot is an experiment with a result, not a demo.
We build quickly and iterate in the real workflow. Each pilot is measured against the baseline recorded in Map, in the same units: minutes per case, cost per transaction, error rate, conversion.
If it misses its target, we change the scope or stop. A stopped pilot is a good outcome when it saves a full build.
Grow: measure, improve and scale what works
Adoption decides whether sized value becomes real value.
Production systems get monitoring, logging, a human override point and a fallback process. AI Academy trains leaders and teams on their own work.
Each quarter we review every system, keep, improve or retire it, and pull the next wave from the ranked list.
How do you work out the ROI of an AI use case?
With five numbers. Each is written down with its source, so finance can challenge it.
- Baseline. What the process costs today per unit: minutes of handling time multiplied by a loaded hourly rate, plus rework and error costs. For growth cases, the baseline is current conversion, retention or revenue per customer.
- Addressable volume. How many units a year AI can realistically touch. Exclusions, such as complex cases that must stay manual, come off here.
- Expected lift. The improvement per unit: share of time saved, error reduction, or relative change in conversion. Lift comes from your pilot where there is one, from comparable published results otherwise, with a low, expected and high case.
- Confidence. A factor from 0 to 1 that discounts the estimate for evidence quality. As a guide: 0.9 measured in your own pilot; 0.7 strong comparable evidence and clean data; 0.5 plausible but untested; 0.3 speculative.
- Payback. One-off build cost divided by monthly net value (value minus running cost). Running cost includes model or hardware cost, support and monitoring.
The formulas
- Cost-saving value a year = baseline cost per unit × addressable volume × expected lift × confidence
- Growth value a year = addressable volume × baseline conversion × relative lift × margin per conversion × confidence
- Net value a year = value a year − running cost a year
- Payback in months = one-off cost ÷ (net value a year ÷ 12)
Worked example: a cost saving
Illustrative numbers, not a client result or a quote.
A shared-services team handles 60,000 inbound emails a year at 8 minutes each and A$45 an hour loaded, so A$6.00 per email. AI can address 70% (42,000 emails) and save half the handling time on each. Confidence is 0.7.
A$6.00 × 42,000 × 0.5 × 0.7 = A$88,200 a year. With running costs of A$12,000 a year, net value is A$76,200. A one-off build of A$60,000 pays back in about 9.5 months.
Worked example: growth
Illustrative numbers, not a client result or a quote.
A business receives 2,000 qualified leads a year and converts 10%. Faster, AI-assisted response lifts conversion by 15% in relative terms, and each win is worth A$4,000 in margin. Confidence is 0.5 because it is untested.
2,000 × 10% × 15% × A$4,000 × 0.5 = A$60,000 a year. With A$15,000 running cost and a A$50,000 build, payback is about 13 months.
On these numbers, this use case could return about A$88,000 a year, and it pays back in about 9 months.
On these numbers, this use case could return about A$88,000 a year, and it pays back in about 9 months.
How this works
- Cost case: units × minutes ÷ 60 × hourly rate gives today’s cost. Growth case: leads × conversion × margin.
- Value a year = that baseline × share AI can address × lift × confidence (0.9 measured in your pilot, 0.7 strong comparable, 0.5 plausible, 0.3 speculative).
- Low and high cases use 60 and 130 per cent of your lift.
- Net value = value − running cost. Payback in months = build cost ÷ (net value ÷ 12).
| Case | Value a year | Net a year | Payback |
|---|---|---|---|
| Low | A$53,000 | A$41,000 | 17.6 months |
| Expected | A$88,000 | A$76,000 | 9.4 months |
| High | A$115,000 | A$103,000 | 7.0 months |
We then rank sized use cases on the ROI feasibility matrix, weighing value against data readiness, feasibility, time to value and risk. Our AI roadmap guide uses the same criteria, our cost guide covers running costs, and the AI cost calculator estimates model spend.
Why do we work this way?
- Design first. We start with users, journeys and critical paths, then show the AI. Design thinking, UI, UX and AX set us apart.
- People approve. The AI reads, drafts, matches and flags. People decide, and own the relationships.
- One fixed fee, never per user. AI Compass is From A$15,000 + GST. Builds are tiered by scope, not seats.
- Stop rules in writing. Every use case has one before work starts, so a weak pilot ends early and cheaply.
- Certified delivery. ISO 9001 and ISO/IEC 27001, 250+ projects in 25+ countries, from Brisbane, San Ramon and Hyderabad.
We've been doing AI before it was called AI.
Ramesh EmandiCo-Founder and CTOSee it applied: AI consulting in Australia, our Brisbane team, or the stage-by-stage guide on how to implement AI in your business.
How we work, answered.
What is the Quantana Path?
It is how every Quantana engagement runs: Map, Solve, Prove, Grow. Map finds where value is won or lost, Solve explores many answers before choosing one, Prove puts working prototypes in front of real users, and Grow measures, improves and scales what works. Each stage ends in a decision gate.
How do you calculate the ROI of an AI use case?
Multiply the baseline cost per unit by the addressable volume, the expected lift and a confidence factor for how good the evidence is. Growth cases use conversion and margin instead of cost. Subtract running cost to get net value, then divide the one-off build cost by monthly net value to get the payback period.
How long does the Map stage take?
Map is delivered through AI Compass and gives you your AI roadmap in 14 days, for one fixed fee from A$15,000 + GST, never per user. It covers process and workflow mapping, data architecture readiness and an ROI feasibility matrix, so you know where AI will pay off before you build.
What is a decision gate?
It is a checkpoint at the end of a stage where the sponsor decides to continue, change scope or stop. For example, a pilot goes to production only if it meets its target and passes risk review. Every use case has a written stop rule before work starts.
What do you need from us?
An executive sponsor, time with the people who run each process, current cost, time and quality figures, and access to data and system owners. Regulated organisations should involve risk and compliance early, because their sign-off is a gate.
Is the value sizer a quote or a guarantee?
No. It is an indicative estimate based on the numbers you enter. AI Compass replaces assumptions with measured baselines from your own processes and gives each use case a value range, a confidence level and a payback estimate before you commit to a build.
Get your AI roadmap
Your AI roadmap in 14 days, with every use case sized in dollars. Get your AI roadmap, see our case studies, or talk to us.
One question for your next leadership meeting: which AI idea on your list has a baseline written down?