A general-purpose AI will answer almost anything you ask it, instantly and with total confidence. That is exactly the problem. When the question is how much notice a resident owes, or what your PTO carryover policy is, a confident answer that did not actually come from your documents is not helpful. It is a liability with good manners.
The cost is not the wrong answer. It is the trust.
A team that gets a wrong answer once and catches it will stop trusting the tool, and you have wasted the purchase. A team that gets wrong answers and does not catch them will act on them. They will quote it to a resident, put it in an email, build a decision on top of it. That second case is the expensive one, because the cost shows up later, somewhere you are not looking, with your name on it.
Where it bites
- A leasing question answered from the open web instead of your lease, so the notice period is wrong
- An HR question answered from a generic handbook instead of yours, so the policy is wrong
- A compliance question answered from last year's rule, because the model does not know it changed
- A number invented on the spot, because the model would rather answer than admit it does not know
Grounded is not a buzzword, it is the whole job
An answer is grounded when it comes from your actual documents, your lease, your handbook, your SOPs, and shows you exactly where. That last part matters as much as the first. A grounded answer you can trace is one your team can act on without re-checking. An ungrounded one has to be verified, which means it saved nobody any time. It just moved the work.
The honest version says it does not know
The tell of a grounded system is not that it always has an answer. It is that when the answer is not in your files, it says so, instead of filling the gap with something plausible. "I do not have a document for that" is a feature, not a failure. It is also a map. The questions that come back with no document behind them are exactly the documents you should write next.
What to ask a vendor
Three questions cut through the demo. Where does the answer come from. Can I see the source on every reply. What happens when the answer is not in my files. If the honest answer to any of them is "the model just knows," you are not buying grounding. You are buying confidence, and confidence is the thing that gets you in trouble.
Answers from your documents, with the source
The Access answers from your own documents and shows the source on every reply, and when the answer is not in your files, it tells you instead of guessing. That is the difference between a tool your team trusts and one they quietly stop using.