Questions we get asked, answered as we would on a call.
Cost, data, ownership and what happens after handover: the questions every operator asks before the first call.
- Ownership
- Do we own what you build?
How long does it take to get a working AI agent running?
The first call is 30 minutes. Exploring takes one or two working sessions with the people who run the process, ideally on site, and shaping the solution a few weeks after that. The proposal is one meeting. The build itself takes roughly 8 to 12 weeks, step by step, with each step signed off before the next one starts. When connecting Claude to your ERP is part of the build, it's often the first thing we do: the one we've measured went live within a week of the first working session, and the team started asking it questions straight away.
What if we don't have clean data?
You don't need a clean dataset to start. If it is spread across several services, bringing it together can be part of the work. If it was never written down, the agent captures what it needs while it works: it runs alongside the process you already run, the work leaves structured data behind, and once it has handled enough real cases it can take on more of the process, with a person able to override it.
What happens to the answers once Claude has them?
The connector hands each answer to Claude so it can write the reply, so the answer becomes part of that Claude conversation, held by Anthropic under the terms of the Claude plan your team uses. On Anthropic's business plans (Team and Enterprise) and its API, Anthropic says it does not use your inputs or outputs to train its models unless your organization opts in, for example by sending feedback on a chat or joining its Development Partner Program. On Team and Enterprise, the plan's owners can turn feedback off. On individual plans, training is a setting each person controls. Which plan fits, and what your IT team needs to approve, is part of the first call.
Microsoft makes its own Business Central connector for Claude. Why not just use that?
Often you can, and we'll say so on the call. Business Central online now includes an MCP server that Claude can connect to: read-only by default, and limited to what each person is already allowed to see. If what your team asks about sits in Business Central's standard API pages, it's a sensible place to start, and setting it up can be all the connecting a build needs. It stops short in four places, and that's when we build one instead. It works with Business Central online only, not on-premises. It sees only what Business Central's API pages expose, so the custom fields and tables many businesses add aren't there until someone writes API pages for them. It hands Claude standard tools, one per kind of record, and leaves Claude to work out which to call; a connector built for your team has tools shaped like the questions people actually ask, with rules about what each one returns, so sales never sees cost fields, for example. And it reaches one system, when many of the questions worth answering cross two: the order in the ERP and the delivery date promised in an email. Either way, the connector is the part that's becoming standard. The work worth doing is finding the process where an answer saves real time, and building the agent that takes that work on.
Do we own what you build?
Yes. The system, the data, and the documentation are yours, and the people who own the process are trained to operate it. There is no license and no lock-in. We stay on afterward to monitor the agents, handle edge cases, and take the next process once value is proven — but that is a partnership around something you own outright, not a dependency you cannot leave.
What does an engagement cost?
Everything before the proposal is free. Every build is scoped and priced in the proposal, against a payback we agree up front. Each step is checked against the agreed gate and signed off by you before the next one starts. We don't publish a price list, because each build is scoped to your process. You get the scope and the number in writing before anything is built.
Why do most AI projects fail?
Four reasons, and none of them is the technology. Poor targeting, meaning work handed to AI that should have stayed human or a process too small to matter. Data that was never captured, is too messy to use, or sits somewhere the system cannot reach. No change management, where the tool ships but the work keeps flowing around it. And impermanent solutions that cannot react when the process changes. In MIT NANDA's 2025 study of AI in business, drawn from interviews, surveys and reviews of more than 300 initiatives, only about 5% of custom enterprise AI tools reached production.
Do we have to replace our existing systems?
No. Agents are wired into the systems your teams already use. The approach is to rebuild the workflow around the process you already run, not to tear out software that works and install something novel in its place.
What industries do you work in?
Chemical and industrial manufacturing, distribution and logistics, hospitality and real estate, and professional services — which includes marketing agencies, consultancies, independent consultants and freelancers, and fractional C-suite roles. The method is the same in any business with repetitive, rules-based work; only the processes differ.
What is the difference between agentic AI and traditional automation?
Traditional automation follows if-then rules defined in advance and breaks on anything it has not seen before. An agent works toward an outcome, handles variation, and can be corrected, which is what lets it take on a run of steps rather than one rule, with a person signing off what it can't decide.
Who runs AIdaea?
Robert Dopart founded AIdaea and leads it, and he is the point of contact on every engagement. He builds with his team, and brings in specialist partners when an engagement needs them.
Where do we start?
With a 30-minute call about how your business runs and where the work gets stuck. If we're a fit, we explore the process with the people who run it, shape a solution with your team, and agree what “working” means. Then you get one proposal: one scope, one number. Nothing is built before you've said yes to it.
Have a question we didn't answer? Ask it directly.
You'd be talking to Robert Dopart directly: the person accountable for the build, not a salesperson.