The fine print
Everything I commit to, in writing. If it leaves your question out, ask me directly: info@ardinsights.com.
What I build
Always inside your own systems. What gets built depends on the problem; these are common examples. Pipelines that catch the paperwork coming in by email, text, and portal and file it where it belongs, flagging what's late. Drafting agents that prepare what goes out (quotes, invoices, renewals, follow-ups) from your own records, each draft waiting for a person to check and send. And the plumbing underneath: pulling the spreadsheets and systems that don't talk to each other into one place you can trust.
Ten commitments, every engagement
These hold in writing on every engagement.
Your systems and your data
Inside your environment.
Everything runs in your own cloud and systems. Your data stays there and isn't used to train any model.
Least privilege.
Each component gets only the access it needs, scoped to the specific mailboxes and systems it works in. A drafting agent can read and draft; it can't send.
Model-agnostic.
Where I choose the model, it sits behind an adapter, so the provider can change later.
No lock-in.
What's deployed in your environment stays yours to run if the engagement ends, with an operating guide, runbooks, and knowledge-transfer sessions. Components I bring from earlier work stay licensed to you for continued use.
Who decides
People approve.
Agents draft and flag. Nothing goes out or gets saved until you or someone you name approves it.
Everything logged.
Prompts, sources, model version, output, and the approval are recorded, and you can pull that record any time someone needs to see what happened and why.
Improves with use.
When your team corrects a draft or decides on a flag, that decision feeds back in. It stays your data.
Limits.
Automation stays inside each provider's terms of use. Where the rules don't allow it, I build a version someone on your side runs. I recommend; you decide.
What you commit to
Proven before handover.
Before you accept the work, I show it working in your business on real cases, then run it supervised for a set period. Where it's relevant, you see its output side by side with the old way.
Warranty, then month to month.
A warranty applies after acceptance. Past it, keeping the system running and improving is a monthly engagement: monitoring, repairs, model upgrades, security updates, and your team's corrections fed back in. You can end it at any time and keep what's deployed.
What leaving looks like
Commitment 2.4, drawn.
The work never leaves your systems, and leaving never costs you the work.
Questions come straight to me.
info@ardinsights.com · +1 404 532 9356