GuestMaker's voice analytics shows sentiment for exactly the calls it has scored so far, and its leadership report grades every claimed booking by evidence.
A dashboard that reads "62% positive" is making a promise. It is telling an operations lead that most of this week's calls went well. What it rarely says is how many calls that percentage is actually built from, and that missing detail is exactly where trust in a metric quietly breaks.
GuestMaker's voice agent answers a hotel group's phone line the way a well-trained switchboard operator would: taking an availability question, sending a booking link (or, on plenty of calls, turning that call directly into a booked reservation), routing a complaint to the right desk, or logging a callback when nobody could pick up. It's the same agent covering every property and venue, so every one of those calls lands in the call center's analytics, and that screen was built around one rule that is easy to state and unusually hard to hold to. A number that is only partly measured has to say so, in the same breath it states the number.
Sentiment is not read from a live keyword scan mid-call. It comes from a full read of the finished conversation, run as calls end and, for anything the live pass missed, caught up again by a background sweep over the following hours. The practical effect is that at any given moment, some calls in a reporting window carry a sentiment reading and some do not yet.
Most tools would quietly average the ones that do and present that as the answer. GuestMaker's analytics attach the denominator to the figure itself. Picture a property whose line took several hundred calls this week, and the classifier has only caught up with a slice of them so far. The screen does not print "62% positive" and stop there. It shows 62% positive over that slice specifically, with a visible note that turns amber the moment coverage drops below complete. A reader who sees that note knows exactly how much of the picture they are looking at. A reader who does not see it knows the number is whole.
The same discipline runs underneath, not just on top. Where a metric genuinely has nothing behind it yet, the interface shows a dash, not a zero. A zero and "not measured yet" look identical on a chart and mean opposite things. One is a real result, and often a bad one. The other is an unfinished measurement. Collapsing the two into the same digit is how a quiet outage starts reading like a quiet success, and how a hotel manager ends up congratulating a line that was never actually scored.
Sentiment gets this treatment because it is exactly the kind of number people act on fast. A team that sees "positive" trending down on a Monday might change how weekend calls get handled by Wednesday. That reaction only makes sense if the number reflects the calls it claims to, not a partial sample dressed up as the whole week.
Sentiment discipline on the daily screen is one thing. It matters more in the report that leaves the dashboard and reaches leadership, sometimes forwarded to people who never log in at all.
GuestMaker's voice performance report is built for that audience specifically. It is reachable as a live link that rebuilds itself for whatever period and scope the reader chooses, rather than a document frozen the day it was generated and slowly going stale in someone's inbox. It draws from the same measurement the analytics screen uses, forced to refresh rather than served from a stale cache, so the figure a general manager reads on a Monday morning cannot quietly disagree with what a marketing manager sees on the dashboard that same afternoon. Where the underlying data has nothing to show for a given slice, a filter simply does not appear, rather than sitting on the page offering choices that would all return the identical, empty result.
A familiar move in phone analytics is to multiply call volume by an assumed booking value and print a clean revenue number at the top of the report. It looks satisfying. It is also not a real figure, because it is built on an assumption nobody checked against what actually happened afterward.
GuestMaker's report does not do this, on purpose, and it says so plainly rather than leaving a quiet gap where that number usually sits: no revenue is estimated from raw call volume, because multiplying requests by an assumed average booking value produces a figure no property's own booking system would ever agree with.
What replaces it is a ladder, and every rung is counted rather than guessed:
The middle rung is not treated as one uniform fact either. A booking traced back to a property's own confirmation page firing the exact marker the agent's link carried is the strongest kind of evidence available. A booking that followed the same guest opening a tracked link, on the same stay, is solid but one step removed. A booking that simply happened at the same hotel in the same window as a link going out is a reasonable inference, and the report labels it as an inference, not a proof. Those three kinds of evidence are never quietly merged into one confident total, because doing that would let the weakest kind borrow credibility from the strongest.
There is a fourth number worth mentioning precisely because it is kept separate. Voice agents sometimes end a call believing it resulted in a booking, and that belief is a real signal about how the conversation went. It is not the same kind of fact as a confirmed reservation, so the report keeps it, labeled clearly as the agent's own account, on its own line. It is never added into the counted total. Adding an AI's read of an outcome to hard evidence of that outcome does not make the number more accurate. It makes it look more confident while actually meaning less.
And when a rung on the ladder reads zero, the report does not leave that as an unexplained flat line. It names which of a few real possibilities is true: no trackable link was ever issued on that channel, links were issued and simply did not convert, or the property's own booking confirmation is not sending back the signal it should be. A rejected signal and one that was never switched on look identical from a bare zero. Only one of those needs a call to the property's web team, and the report tells the reader which.
It would be simple, and would look more impressive at first glance, to round every partial number up to full confidence and publish one clean revenue figure at the top of the page. It would also make the report wrong in a way an operations leader eventually catches. Once a report has been caught overstating one number, every other number in it becomes suspect too, including the ones that were correct all along.
Disclosing coverage instead of hiding it, and grading evidence instead of inventing certainty, is what lets a reader trust the number actually in front of them enough to act on it without a mental asterisk. That is a small thing to ask of a report and a genuinely hard thing to deliver consistently, which is exactly why it is worth building into the product rather than into a disclaimer at the bottom.
For more on how the agent behaves on the call itself, see one number, one brain. The same refusal to overclaim what the system can actually vouch for runs through how GuestMaker keeps AI safe with real guests as well, and the full picture of what the voice line can already answer, plus what it deliberately still leaves open, lives on the voice product page.
Thirty minutes, your real properties, no generic demo environment.
Most AI hotel phone lines end with a promised callback. GuestMaker's voice agent is built to create the real reservation live, during the call, and read back a genuine confirmation code.
Inside every GuestMaker AI conversation: spend limits, mandatory human escalation, manipulation defenses, automatic PII scrubbing, and house rules configured for your group.
Most hotel phone lines go unanswered, are picked up blind, or recite a script. GuestMaker's voice agent gives every property and venue a real memory instead.
