Trusting AI answers

AI tools are very good at sounding sure of themselves. That’s fine when you’re drafting an email. It’s a problem when the answer is a margin figure going into a board paper, or a policy clause that decides whether a claim is paid.

In business, a wrong number delivered with confidence is worse than no number at all. So the useful question about

any AI tool isn’t “is it clever?” It’s “can I check it?” Here’s how GYCO is designed so that the answer is yes.

AI writes the words, not the numbers

The most important design choice in GYCO is about who does the counting.

When you ask about your business data, the AI’s job is to turn your question into a precise query: which figures, which filters, which time period. Your database then does the counting. The numbers in the table and chart come from running that query on your data, not from the AI’s memory or its sense of what sounds plausible.

AI Insights works the same way. Every figure on an insight card is calculated from your data: the value, the change,

and the split by region or product. The AI writes the recommendation in plain language, around numbers it didn’t

produce.

Answers from your documents are different, and we’re upfront about that. There, the words are the AI’s, drawn from the passages GYCO found. That’s exactly why every document answer carries citations.

Read-only, by design

GYCO can read your data. It can’t change it. Every query it runs is read-only, so it never adds, changes or deletes a

record. It also caps how many rows a single query can return.

It works on its own copy of your data, which our team loads during onboarding. Asking a question never touches your source systems.

See exactly what was counted

A real number can still be the wrong number if it counted the wrong thing. Did “last month” mean the calendar month or the last 30 days? Did “returns” include exchanges?

GYCO gives you two ways to check.

Clarifying questions

When a question could reasonably mean two things, GYCO can ask which you meant, sometimes with options to pick from, rather than guess.

Show the working

Two settings in your own preferences let you see the exact query behind an answer and the steps GYCO took to reach it. Both are off by default to keep answers clean, and anyone can switch them on. An analyst can read the query line by line. A manager can at least see which figures and filters were used.

Citations you can click

When GYCO answers from your documents (a policy, a standard operating procedure, a manual), the answer carries

numbered markers. Click one to see the passage it came from. In Data Retriever, you can open the original document and read the surrounding context for yourself.

GYCO is built to treat text inside your documents as information, not as instructions. A line in a document that says

“ignore your rules” is read as part of the document, not obeyed.

Answers that stay the same when you share them

When you download an answer table to CSV or Excel, the file comes from the stored result of the query you saw.

GYCO doesn’t run the query again, so the numbers in the spreadsheet match the numbers on screen. You can also

export a whole conversation, tables included, to PDF.

A record behind every answer

GYCO keeps a record of each answer and the steps behind it. If an answer mattered, it can be reviewed later: what

was asked, what was run and what came back. Changes to users, roles and settings go into a separate audit log that

administrators can search.

Where the work happens

Each customer gets a dedicated GYCO environment of its own, and you choose how the AI models behind it run.

• Self-hosted. The AI models run inside your dedicated environment, so your questions and your data are never sent to an outside AI service. It’s the choice for organisations with the strictest privacy and security requirements.
• A private managed service. The AI models run on a private, managed AI service from the cloud provider, in the same region as your environment. Running costs are lower, and your data is never sent to public AI chat services.

Five questions to ask any AI tool

Whether or not you choose GYCO, ask these before you rely on any AI tool for business decisions:

1. Where do the numbers come from? Are they calculated from my data, or generated by the AI?
2. Can it change my data, or only read it?
3. Can I see what it counted, and how it read my question?
4. Does every answer from a document point to its source?
5. Is there a record I can go back to?

If the answer to any of them is “no” or “not sure”, the tool is asking you to take its word for it. Good decisions deserve better than that.