The half-cent answer

Every pricing model in customer support is a bet on one number: what it costs to answer a question. For most of the industry's history that number was a wage — the cost of a human-handled ticket is commonly cited at several dollars — and every seat license, every metered plan, every usage tier was priced downstream of that assumption. The assumption just quietly stopped being true.

The math, in the open

On Bubblio, a support answer costs about $0.006 in model tokens. Six tenths of a cent — less than a hundredth of a postage stamp at today's rates. The title rounds that in our favor; the arithmetic below does not.

• Chat runs on a small, fast model. At today's published API prices, that class of model costs about $1 per million input tokens and $5 per million output.
• Every reply is hard-capped at 512 output tokens — about $0.0026 of output. Add the question, the conversation so far, and the knowledge the model reads before answering, and a typical answer lands around $0.006.
• Conversations are capped at 40 turns. A conversation that maxes out every cap costs at most about twenty cents in tokens. A typical one costs two to five cents.

Set that against the commonly cited several dollars for a human-handled ticket — a gap of a few hundred times. The point is not that the model replaces the human; ours emails a human the moment one is needed. The point is narrower: the marginal cost of a good first answer has collapsed by orders of magnitude, and it happened within the last couple of years.

The strange part is the pricing

Here is what we find strange. Support software is still mostly priced as if answers were expensive: per seat, per resolution, per metered unit of help. Those models made sense when every answer consumed a slice of someone's workday. They persist because meters are good business — not because the costs underneath them still exist.

When an answer costs half a cent, metering answers is a business model, not a necessity.

A meter also does quiet damage on the customer's side of the counter. When every conversation is a small expense, the software drifts toward deflecting conversations instead of inviting them. We wanted the opposite incentive: a bubble that is genuinely glad to see your customers.

A utility bill you never have to check

So Bubblio is flat: $29 a month in beta. Not $29 plus usage, not $29 per seat. The token math above is what makes it possible — at typical conversation costs, five hundred conversations run somewhere between ten and twenty-five dollars of tokens, and the caps put a hard ceiling on the rest. The only prepaid thing in the product is the optional avatar face, because realtime video genuinely costs more; when a video balance runs out, the bubble degrades to text chat. It never cuts a customer off.

Flat pricing needs a fair-use line, so here is ours, plainly: $29 covers roughly 500 conversations a month — one visitor or agent, one thread, however many messages it takes. Past that, nothing shuts off and no meter starts — a human from Bubblio emails you and you work out together what your volume needs. The bill does not grow on its own. We treat "a surprise bill is structurally impossible" as a design constraint, not a slogan: there is no path through the product that ends in an invoice you did not already know the size of.

It answers for us at 3 a.m.

The pricing is also a bet we place on ourselves. We run Bubblio on our own site. It answers visitors at three in the morning, and it emails the founder only when a customer actually needs a human. Most mornings that means conversations that needed nobody, and occasionally one email that needed him. That is what support priced like a utility feels like from the inside: not a cost you manage, a fixture you installed.

When answers cost half a cent, support can finally be priced like the boring bills — a flat number you set once and stop checking. The install is about five lines, and the beta is $29 a month, flat. If that reads like your kind of bill, request access.