We sell AI translation, so treat this as an interested party being careful. There are events where a human interpreter is the right answer and we will say so plainly below.
What human simultaneous interpretation costs
Simultaneous interpretation is priced per language pair, not per event, and it is almost never one person. Standard practice is a booth of two interpreters per pair, who swap roughly every 30 minutes, because the cognitive load makes solo work unreliable past that. In Europe you are typically looking at a day rate per interpreter in the hundreds of euros, doubled for the pair, plus booth hire, a technician, and receivers for the audience.
The number that surprises people is what happens when you add languages. Three languages is three booths, six interpreters, three times the receivers. Cost scales linearly with languages, and so does the floor space.
Where AI translation actually wins
Language count. The cost of the fourth language is the same as the first: nothing. That single property is why a mid-sized conference can offer 12 reading languages at once (from a catalogue of 58) when its budget covered two interpreters.
Reach. Interpretation reaches whoever holds a receiver. Captions on attendee phones reach whoever has a phone, and the recording, and the person watching remotely.
Lead time. Interpreters are booked weeks out. Software is not.
A written record. You end up with a transcript and, if you want one, minutes of the event. An interpreted event leaves nothing behind unless someone was also taking notes.
Where a human interpreter is still the right call
Be honest about this, because getting it wrong is expensive.
- Legal and medical settings. Where a mistranslation has consequences that a disclaimer does not cover, hire a certified interpreter. This is not a cost decision.
- Negotiation. An interpreter reads the room, catches hesitation, and can flag that a phrase does not carry. Software translates the sentence it was given.
- Highly idiomatic or ceremonial speech. Humour, poetry, liturgy and rhetorical set-pieces are exactly what machine translation flattens.
- Where accuracy must be attested. If someone has to sign that the interpretation was faithful, that someone has to be a person.
The pattern that works in practice
The events that go best do not treat this as an either/or. They put an interpreter in the booth for the one language pair that matters commercially, and run AI captions for up to eleven more languages (12 at once in all), for the delegates who would otherwise have had nothing at all.
Latency matters here. Translated captions have to land while the sentence is still relevant to what is on screen. SCAPTION delivers them about 1.7 seconds after the word is spoken (median), and within 2.5 seconds in all but the slowest one in ten, measured on our own service, not quoted from a vendor datasheet.
Try it against your own material
The trial is one free hour: no expiry, no card. Point it at a recording of last year’s conference and read the output in a language you actually speak. That is a far better evaluation than any comparison table, including this one.