Anno One
A review of your AI initiative: workflow, data, cost and control
We go through the AI initiative from the workflow it should change to what it costs when everyone uses it. You get a written recommendation on what to start with.
The workflow first
Which task should take less time or give a better result, who does it today and how you will know it works. An AI project without a measurable workflow tends to stay a pilot.
Where the data is
Documents, the CRM, the ERP system, shared drives and email. We look at what the model needs to see, what it must not see, and whether the data is clean enough to answer from. Personal data and trade secrets decide which providers and regions are possible.
Cost at real volume
A pilot with ten users says little about the bill when the whole company uses it. We estimate usage from the number of requests, the length of the texts and the model, and look at where a smaller model or caching gives the same result for less. The guide on your own API keys and spending caps describes the controls.
Keys, logs and providers
Whose contract the usage runs on, where the API keys are stored, who can see the logs and what happens when a provider changes its terms. An AI gateway puts that control in one place; ModelHarbour is our own.
Rules that apply
GDPR for personal data, and the EU AI Act's transparency requirements, for example telling users when they talk to an AI. We point out what applies to your use and where you need legal advice.
What you get
A written recommendation: the workflow to start with, the data it needs, an estimate of the running cost and the controls to have in place before launch. AI and automation covers the build.
Questions we get
We already have a ChatGPT or Copilot licence. Is an AI review still useful?
Yes. The licence covers the tool; the review covers which workflows it should change, what data it may use and what a wider rollout costs.
Do you recommend a specific AI provider?
We recommend what fits the workflow, the data and the rules that apply. Often that is more than one model.