Services · AI strategy consulting
Find out where AI pays in your operation. And where it doesn’t.
A paid diagnostic for ecommerce businesses doing $1M–$50M. Two to three weeks, one written document, a baseline you can measure against afterwards. Roughly half of what it contains will be things we think you should not automate.
A$5,000–A$8,000. Ace takes the first call. Bring your numbers.
The problem
You are pitched AI weekly and none of it is falsifiable.
Every app in your stack has added a feature with a sparkle icon on it. Every agency has a new practice area. The claims are large, the case studies are thin, and none of them come with a number you could check afterwards even if you wanted to.
Meanwhile the actual cost in your business is invisible. It’s not a line item. It is a capable person spending Thursday morning reconciling three systems that disagree, a support queue full of questions the order data could answer, and a production or fulfilment decision made on a stale number because the fresh one takes an hour to assemble.
The failure mode is well documented. Pilots do not fail because the technology cannot do the task. They fail because nobody measured anything before starting, so when the pilot ends there’s no way to tell whether it worked, and the safe organisational decision is to quietly stop. That is a measurement problem wearing a technology problem’s clothes.
So the first job is not choosing a model or a vendor. It is writing down what the operation costs you today, in hours and in errors, with enough precision that a change would be visible. Everything else follows from that, including the decision not to build.
95%
The method
Six steps, in this order, every time.
It is deliberately unglamorous. The value is in doing the counting before the choosing, which is the step everyone skips because it is the only part that isn’t fun.
1. Map the operation end to end
Order in to cash collected, including the returns loop and the exceptions. We sit with the people doing the work rather than the people describing it, because the description and the work are rarely the same thing. This is where most of the two to three weeks goes, and it is the part vendors skip.
2. Count the repetitive work
A task inventory: what happens, how many times a week, how long each instance takes, how much it varies, and who does it. Anything that happens two hundred times a week and always the same way is a candidate. Anything that happens twice a year is not, no matter how annoying it is.
3. Test each candidate against four questions
Is the volume high enough that saving minutes matters? Can the rules be written down by a human before a model is involved? Is a wrong answer recoverable, or does it go straight to a customer? Is the underlying data clean enough that a system could read it today? A candidate that fails any of the four goes on the no list with the reason written next to it.
4. Set the baseline
The numbers as they stand, written down and dated: hours per week by task, error and rework rates, where the systems disagree and by how much. Nothing gets built until this exists. It is the single reason most AI pilots cannot prove anything — they started without one.
5. Rank the work and price it
A written order of work: what to do first, second and third, what each one is likely to cost, and what it should return. Every item is marked build, buy or leave alone. Buying an existing product is often the right answer and we will say so.
6. Hand it over
The document is yours. If you take it to your own developers, or to a cheaper firm, or to nobody at all, the engagement has still done its job. That is what charging for it buys — a document that does not have to end in a proposal.
Qualifying
Who this is for, and who it isn’t.
The second list is longer than most firms would print. It saves everyone a fortnight.
This is for you if
- You are doing between $1M and $50M and the operation has outgrown the tools that got you here.
- There is at least one process someone performs a hundred or more times a week, by hand, the same way each time.
- Your systems disagree with each other and somebody rebuilds the truth in a spreadsheet.
- You can give us read access to real data — orders, inventory, fulfilment, support tickets.
- You want to know what not to do as much as what to do.
This is not for you if
- You are pre-revenue or pre-product. There is nothing to measure yet, and a baseline of zero is not a baseline.
- You need a strategy deck for a board or an investor. We write working documents, and they are not attractive.
- The decision is already made and you want it validated. We will probably disagree in writing, which is an expensive way to buy agreement.
- You want a storefront chatbot. Gorgias, Intercom and similar do that well and cost a fraction of this.
- You want an organisation-wide programme with a change-management workstream. That is a different kind of firm and they exist.
Scope and price
A number before we start.
The range is where these engagements land, not an opening bid. The variable is how many systems are in play and how bad the data is. We agree the figure in writing before the kick-off call.
| Item | Detail |
|---|---|
| Duration | 2–3 weeks, from kick-off to the handover call |
| Price | A$5,000 – A$8,000, quoted before we start |
| Your time | Roughly 6–10 hours across the engagement, mostly the ops lead |
| Access needed | Read access to Shopify, your 3PL or WMS, support desk, and finance exports |
| Deliverable | One written document: reconciliation map, hours bill, ranked order of work, baseline |
| What happens next | Nothing automatically. A build is quoted separately, or not at all |
Questions
What people ask before booking.
It is working out where machine automation pays in a specific business, and where it doesn't, before anyone spends money building it. In practice that means mapping the operation, counting the repetitive work, testing each candidate against whether it is actually automatable, and writing down the numbers as they stand today so any later change can be measured. The output is a ranked list with prices against it, not a capability tour.
A free audit is a sales document. It was always going to conclude that you need the thing the firm sells, which is exactly why nobody believes the findings. Charging for it removes that. We are paid for the analysis, so the analysis can say 'buy this off the shelf' or 'do nothing here' without costing us anything.
Two differences that matter. First, we set a measured baseline — hours, error rates, the size of each discrepancy between systems — so the work can be judged later against something other than opinion. Second, the people doing it are running an ecommerce operation this week, so the judgement about which exception matters comes from having handled it rather than having read about it.
Then that is what the document says, and you keep the document. It is a common outcome. Off-the-shelf software has got considerably better, and a well-configured standard stack beats a mediocre custom build every time. Finding that out for the price of a diagnostic rather than the price of a build is the point.
Yes, read access at minimum — orders, inventory, fulfilment events, support tickets and finance exports. Without real data we would be guessing, and a strategy built on guesses is what produced the pilots that never went anywhere. We work under an NDA and we do not need write access to anything.
Roughly six to ten hours in total, most of it from whoever runs operations, spread across the two to three weeks. We would rather spend an hour standing next to the person doing a task than three hours in a workshop about it.
Yes. Most of it is asynchronous — data review, mapping, writing — with scheduled calls in overlapping hours for the US and Europe. If a piece of the operation genuinely needs someone physically present, we will tell you that during scoping rather than after you have paid.
Nothing, unless you want it to. If you want us to build the first item on the list, it is quoted separately as fixed-scope work with a number agreed before anything starts. Plenty of clients take the document to their own team instead, and that is a legitimate outcome we priced for.
Next step
Start with the diagnostic.
Two to three weeks, A$5,000 to A$8,000, and a document you keep regardless of what happens next.
Ace takes these calls. Not a salesperson, because there isn’t one.