01
Grounding and citation
Whether answers come from your content with the source shown, and how that content is kept current when a policy changes.
SourcesFreshness
An Appnox product · Conversational support
A chatbot is only as good as what it is allowed to say.
Conversational support grounded in your own content, with sources shown, clear handover to a person, and explicit limits on the actions it will attempt on a customer record.
The problem
The same questions, endlessly, at the wrong time.
A predictable share of inbound contact is the same handful of questions whose answers already exist in your own documentation — just not in a form anyone can find quickly.
The same questions arrive repeatedly and answering them is nobody's measured job.
Answers live in documents customers cannot search and staff cannot find quickly.
Documentation and policy change, and the answers given drift out of date.
Customers ask outside working hours and wait.
An existing bot answers confidently and occasionally wrongly, which is worse than silence.
There is no clean path from the bot to a person who can actually help.
An Appnox product
A product, reported separately from consulting.
Developed and reported on separately from Appnox consulting engagements, with no shared performance claims in either direction.
Conversational interfaces are also the most over-sold category in applied AI. The useful version is narrow, grounded and honest about its limits; the version that answers everything confidently is the one that eventually generates a complaint you have to honour.
Before you buy
What to demonstrate before you buy.
The differences that matter in this category are not visible in a scripted demo. These are worth testing with your own content and your own awkward questions.
02
Behaviour at the edge
What it does with a question it cannot answer. Declining and handing over is correct; generating a plausible answer is the failure mode that costs you money.
RefusalHandover
03
Permissions
Whether it respects who is asking. An assistant that answers account questions without establishing identity is a data protection incident waiting to happen.
IdentityAccess
04
Action boundary
What it may actually do to a record as opposed to what it may say. These are different permissions and should be configured separately.
ReadWriteApproval
Honest limits
What we will tell you plainly.
A chatbot does not fix a process problem. If customers are asking because your journey is confusing, the answer is to fix the journey; a conversational layer over a confusing process produces confused conversations.
It also does not replace a support team. It absorbs a share of the repetitive volume and should route everything else onward faster than before. Products sold on headcount reduction tend to disappoint on exactly that measure.
And where a customer is already frustrated, an additional interface between them and a person makes it worse. Detecting that and getting out of the way is a feature worth more than any answer it could have given.
Availability, current feature set, setup work and licensing are confirmed in a proposal. Product capability is never presented as evidence of consulting delivery, or the reverse.
Before we talk
A few useful answers.
Grounding, permissions and boundaries. A general model with no access to your content will invent answers about your business, and it will do so fluently. What matters is what it can see, what it is allowed to say, what it refuses, and what it can do to a record — none of which a general model has configured for you.
That is a deliberate configuration decision, and the default should be no. Read and write are different permissions, and anything carrying commercial or duty-of-care exposure belongs behind a human approval. A product that enables write actions by default has made that decision for you.
Grounding in your own content with citation reduces it substantially; nothing eliminates it. The practical defences are constraining the scope of what it will attempt, showing sources so an answer can be checked, and making refusal an acceptable outcome rather than a failure state.
Yes, as a consulting engagement, when the requirement is genuinely specific to your operation. If a product covers it, the product is cheaper and faster, and we will say so.
Keep exploring
Related work
Start with a conversation
Product, or a build around what you already run?
Twenty minutes with a solutions architect is usually enough to establish which. We are comfortable telling you the product does not fit.
Free · 20 minutes with a solutions architect · No obligation to commission an audit · sales@appnox.ai