Ask a generic AI tool a question and if it doesn’t have quite enough information to answer, it will usually answer anyway, filling the gap with a plausible-sounding assumption rather than admitting the gap is there.

For an accountant relying on that answer, an invented assumption dressed up as fact is worse than no answer at all.

The Croner Intelligence virtual assistant is built to behave differently. Rather than guessing, it is instructed to notice when it doesn’t have enough information to go on, and to say so. Two situations trigger that.

When the question is the problem

Some questions are clearly too vague to search properly. Ask ‘what’s the deadline?’ without specifying what for, and obviously the system will not have the necessary information to respond; there is no ‘likeliest interpretation’. Instead, the Croner Intelligence virtual assistant asks one targeted clarifying question before any search takes place, rather than speculating about what was meant.

When the sources fall short

More often the gap appears mid-answer, once the sources have been retrieved and don't fully settle the point. Here the instruction is explicit: where crucial facts are missing, list what is needed before proceeding, and never hypothesise facts or fill gaps with assumptions.

A termination payment query demonstrates this in practice. Rather than taking a settlement agreement’s label at face value, the system identifies that the payment could fall under different statutory treatments, cites the authority for each, and specifies the facts still needed to decide between them.

Where the knowledge base holds nothing relevant, the response says so plainly rather than reaching for general knowledge instead. Where it holds partial or adjacent material, it states what isn’t covered before offering what is a stated gap, rather than an answer stretched beyond what the sources support.

Checked, not hoped for

None of this is left to chance. Every draft answer is reviewed against a list of known failure modes, including unsupported claims, before it is returned. A separate automated check then tests whether the finished answer is genuinely grounded in the material it cited.

Why it matters

A confident answer resting quietly on an assumption is the hardest kind to catch, because nothing on screen signals anything is missing. An answer that names what it doesn’t know hands that judgment back to the practitioner, where it belongs.

An admitted gap is more useful than a fluent guess, a small design choice with an outsized effect on whether an answer can be trusted. This is what makes Croner Intelligence a gamechanger for accountants.

*** Try out Croner Intelligence for yourself. Book a demo here ***