How it runs

What a readiness review actually involves

Five stages, five working days, a fixed fee. At the end you get a go or no-go on each task you were considering, and if it is a no, the specific list of what would change that.

The service page explains why this exists. This one is the method: what happens on which day, what we ask you for, and what lands on your desk at the end. It is here because "we will assess your readiness" is the kind of sentence that could mean anything, and you should be able to see the shape of the work before you pay for it.

Why it usually comes first

Most failed AI projects did not fail at the model. They failed because the data the model needed was not reachable, nobody could tell whether the output was right, and there was no record of what it did. Those are answerable questions, and answering them costs a fraction of discovering them halfway through a build.

It is not a prerequisite for everything. Connecting two systems that already hold clean data does not need one, and neither does a report that runs off a database you already trust. It earns its place when a language model is going to make a judgement that someone will act on.

The five stages

Day 1Name the tasks. Each candidate automation gets written down as one sentence with an input, an output and a frequency. "Handle our email" cannot be assessed. "Read incoming supplier invoices, extract eight fields, post them to the ledger, about ninety a week" can. This stage alone often removes a task from the list, because writing it down exposes that it is really four tasks.

Day 2Trace the data. For every field the task needs, we follow it back to the system that owns it and establish three things: whether it can be fetched programmatically, how current it is, and what happens when it is missing. Data that exists but takes a person to locate is not available to an automation, and this is where most no-go answers come from.

Day 3Build the evaluation set. Twenty to fifty real historical examples with known-correct answers, chosen to include the awkward ones rather than the tidy ones. If we cannot assemble this from your existing records, that is itself a finding: it means you would have no way to tell whether the automation was working.

Day 4Test the accountability trail. We take a handful of past decisions and try to reconstruct why they came out the way they did, using only what your systems retain. If we cannot do it for a human decision made last month, an automated one will be no easier, and that matters the first time a customer disputes an outcome.

Day 5Cost it and decide. Build estimate and running cost per task, then a go or no-go on each. Running cost is the number most proposals leave out, and for anything calling a model at volume it is frequently the number that decides whether the automation is worth having at all.

Stages two and three are where the real work sits. One and five are short, and four is usually half a day unless the audit trail turns out to be the problem, in which case it becomes the main finding.

What we need from you

Roughly four hours of your people's time, across the week

An hour at the start with whoever actually does the task now. Not their manager. The person doing it knows where the exceptions are.
Read access to the systems involved, or a screen share with someone who has it. Read only, and time-boxed to the review.
Sample records, which can be structure-only or with names replaced. We ask for real data only where the assessment genuinely depends on real values. That rule is set out in how we handle client data.
Half an hour at the end to walk through the findings, so a no-go is a conversation rather than a document you receive and disagree with.

What you get

A verdictGo or no-go on each task, separately. Not a readiness score out of ten, which sounds rigorous and tells you nothing you can act on.

The gapsFor each no, what specifically would need to change, and a rough sense of the effort. Some of it you can usually do yourself.

The evaluation setYours to keep either way. If you build with someone else, it still works. If you build with us, it becomes the baseline for ongoing monitoring.

CostsBuild and running cost per task, so the decision is a number rather than a feeling.

The alternativeWhere a rules engine, a scheduled report or a fixed script would be cheaper and more predictable than a model, we say so. That happens more often than the market implies.

Cost and duration

Fee$1,500 fixed, covering up to three candidate tasks. Additional tasks are $400 each, because the data tracing is per task rather than shared.

DurationFive working days for up to three tasks. Six to eight if you are assessing more, or if the systems involved need access arranged through someone else.

What moves itThe count of systems, not the size of the business. Four tasks living in one database is faster than one task spanning four systems.

Prices are indicative for the scope described and current at the date of publication. The figure you are charged is the one in a written quote for your work. See terms.

Same fee as the Automation Audit, deliberately. They answer the two halves of the same question: the audit asks which task is worth automating, this asks whether your data can support it. If you already know which task, start here.

When you do not need one

Worth saying, because we would be selling more of these if we left it out.

The task has no judgement in it. Moving a record between two systems on a fixed rule does not need a model, so it does not need a readiness review either.
Everything lives in one system you already trust. If the data is in one place, clean, and you can already run reports on it, most of stage two answers itself.
The output is low stakes and reversible. Drafting text a person edits before sending can be tried directly, cheaply, without a formal assessment first.
You are below the size where it pays. If the task happens twice a week, the honest answer is often that a person should keep doing it.

If any of those describe you, say so when you get in touch and we will tell you to skip it.

Thinking about building something with AI? Tell us the task you have in mind and where the data lives. If the honest answer is that your data is not ready, you will get that, along with what would need to change.

Get your free review

Home AI readiness review Worked example Terms otr-world@pm.me