PKCO, LLC d/b/a Caliber Integrations installs AI systems inside small businesses. This page explains how we approach that work: what these systems are typically built to do, who checks them, where they fail, and what we will not build. It is not legal advice, it is not an offer or a promise of any particular system or result, and it does not form part of any agreement. See also the website terms and the privacy policy.
This page describes how we generally work, not what your build includes. What Caliber delivers to a specific client, how it is configured, and what either of us is responsible for are set only by that client's signed Client Services Agreement and statement of work. Where this page and a signed agreement differ, the signed agreement controls.
What the AI actually does
Across the work we do, AI drafts and prepares. Depending on what a scope calls for, that can mean writing replies, quotes, summaries, posts, or paperwork; sorting and routing incoming work; or assembling a recurring report from what already happened. These are examples of what such systems can be built to do — not a list of what any particular engagement includes. Your signed scope decides which, if any, of them get built.
What does not change is the role. In the systems we build, AI produces a first draft and does clerical work. It is not the decision-maker.
A person approves before anything leaves your business
We build to a default of human approval: as we configure and hand a system over, it drafts, and one of your people reads what was drafted and sends it. Where the tools allow it, we set that boundary in the configuration rather than leaving it as a written rule, and we test the off switch with you during the installation. If a scope ever calls for something to send without review, that is a separate, written decision agreed in advance — not something we would do quietly.
After handoff, the system is yours to run and yours to change. You own the accounts, so you or anyone you give access to can widen permissions, connect new tools, alter instructions, or tell a system to act without review. If that happens, the system stops matching what is described here, and what it then does is outside our control and outside what we configured. If you are considering a change like that, ask us first — that is what the support window and any ongoing agreement are for.
The same rule applies before we deliver anything to you. When we use an AI assistant to prepare for a call, a person reviews and edits it first. No AI system makes a decision about you.
What these systems get wrong
- They state wrong things confidently. A draft can invent a detail, a date, a price, or a policy that does not exist, and it will read as well as a correct one.
- They are not consistent. The same input can produce a different draft on a different day.
- They miss context they were never given, including anything decided in a hallway, a phone call, or someone's head.
- They change underneath you. The providers we build on — currently Anthropic and OpenAI — update, deprecate, and retune their models on their own schedule, and a system that behaved one way in March can behave differently in June.
- They depend on services we do not control. Availability, latency, and pricing belong to the provider.
This is the reason for the approval step. We design around the assumption that a draft is sometimes wrong, rather than promising it will not be.
Decisions that need a qualified human
Do not let an installed system be the last word on a legal, tax, accounting, medical, financial, employment, insurance, safety, or regulatory question, on hiring or firing, on extending credit, or on anything else that materially affects a person's rights or money. Those need the judgment of someone qualified to give it. Our site content and pre-engagement discussions are general business information, not regulated professional advice.
What we need from you
- Only give a system data you are authorized to share, and keep sensitive or regulated data out of anything that does not require it.
- Own the accounts. We build in your own accounts rather than ours wherever the tools allow, so the access, the data, and the ability to switch anything off stay with you.
- Check the source material. A system built on an out-of-date price list will confidently quote out-of-date prices.
- Name the approver. A workflow built to draft-and-approve only works if that role belongs to a real person who actually reads it.
- Tell us when it drifts. If a system starts producing work you would not have sent, that is worth an email, not a workaround.
Your data and your name
Your business data stays yours. We build in your own accounts rather than ours wherever the tools allow, and anything you share with us before an engagement is handled under the privacy policy. For paid work, your signed scope names the exact accounts, approved vendors, and permitted data. The third-party services involved process information under their own terms.
We do not publish your work, your data, your screenshots, your results, or your name without your written permission. Anything we demonstrate publicly is one of Caliber's own live systems, synthetic sample data, or client work we were given permission to show — and we say which it is. Any client case study we publish will carry written permission and state how the result was measured.
Work we will not build
We will decline or pause work where a system would deceive the person receiving its output, impersonate a real person without disclosure, make consequential decisions about people without human review, discriminate against a protected class, produce unlawful output, or operate on data the client is not authorized to use. If we find one of these mid-engagement, we stop that workflow and tell you why.
What this page does not claim
Caliber holds no AI-specific certification, audit report, or accreditation, and this page is not one. It is a description of how we work, reviewed on the date above. We do not claim that an installed system is accurate, secure, unbiased, or compliant in the abstract. What we do is build so that a person is in the loop before output counts, and tell you plainly where these systems fail.
Where the binding terms are
This page describes practice. The enforceable commitments for paid work — which models and providers are used, what data they may touch, how long anything is kept, and who is responsible for what — live in the signed Client Services Agreement and statement of work for that engagement. Where this page and a signed agreement differ, the signed agreement controls.
Questions
If something an installed system produced concerns you, or you want to know how a specific workflow was built, email contact@caliberintegrations.com with the workflow and what happened. We answer these before feature requests.