The hidden costs of 'cheap' support software

By the Onsites AI team · Last updated August 28, 2026 · 5-minute read

The most expensive support software a team ever buys is the one that advertised itself as cheap. Not because its seat rate lies — the seat rate is usually honest — but because the seat rate is one line of a bill whose other lines arrive in months two through twelve: the AI add-on, the per-resolution pack, the second product the desk actually needed, the integration that should have been native, the migration fee, the renewal repriced, the export that never quite exports. This guide is the line-by-line audit of that invoice — the invoice behind the invoice. It exists because procurement compares logos and feature grids while meters compound quietly underneath; the discipline it teaches is to price lines, not logos. And it closes with the alternative bill shape: Onsites' published meters — seats $15 beyond three free, storage $1/GB, AI credits at about a cent — designed so that everything on this audit list is either a published line or does not exist.

Line one: the seat, and what rides on it

The advertised seat price is real and rarely the whole seat. The tier ride: the feature the team needs this quarter sets the tier for every seat — a $49 seat becomes $79 because routing or SLAs or surveys live upstairs; the tier-tax guides document the pattern desk by desk. The seat definition: do admins count? viewers? the finance lead who touches ten tickets a month? A "5-seat" desk that licenses honestly is often an 8-seat bill. The annual gate: the honest monthly rate is frequently only available billed annually — monthly billing prices 20–40% higher. The participant problem: occasional collaborators price like cores, so desks ration access to the tool meant to be shared. None of these makes a vendor dishonest; together they make the advertised line the cheapest line on the invoice, which is exactly how the audit must treat it.

Line two: the AI that lives above the seats

The modern surcharge arrives wearing intelligence. The assistant subscription: drafts, replies and summaries behind an add-on priced per seat or per month — the desk pays for the AI whether a slow month used it or not; the team quietly abandons the feature to stop paying for it. The resolution pack: per-resolution meters (≈$0.99 each) that look harmless at average volume and double at peak — the resolution guide works the full math. The hidden blend: "500 resolutions included" that nobody divides by the real AI-adoption rate. The gate itself: the desk's quality tooling — translation, summaries, drafts — held a tier above the seats, so a team that cannot budget the add-on runs without assists and wonders why rivals reply faster. The audit question for any AI line: what is the unit cost at my adoption rate and my peak month — and are the credits prepaid-and-published (cents, expiring never) or subscribed-and-bundled (a line that never sleeps)?

Lines three through five: the tools the desk actually needed

The advertised product is never the whole stack. The CRM gap: contact records included; accounts, pipeline and conversation-linking are the other subscription — the CRM guide shows the forked-memory cost. The documents gap: quotes, orders, invoices and contracts live in a second product (or the third), so "send the updated proforma" is a copy-paste tax paid weekly. The channel gap: WhatsApp, WeChat, LINE and Messenger arrive via paid connectors or marketplace partners, each its own line item. The knowledge gap: the help center product is separate, metered by article count or branding. The report gap: analytics behind the analytics plan. Count them at a team's real usage and the "cheap" desk's stack frequently double-counts its seat bill — three products to answer one customer. The audit lines these up in year one, when each is still a choice, rather than in year three's renewal spreadsheet, when each is a dependency.

Lines six through eight: the taxes nobody models

Three taxes arrive after the invoice and stay for years. Setup and migration: professional onboarding fees, data-import services, or "partners" charging a day's rate to move a mailbox — the honest vendor's total includes the week your team spends re-learning. Training and change cost: a UX designed around tiers (views locked, macros limited) re-trains the team repeatedly as tiers change. Lock-in and the exit tax: proprietary formats, throttled exports, APIs that price leaving — and the renewal repriced upward because loyalty is priced last. Add renewal's sibling, the price-increase email in year two, and the bill's tail grows past every comparison made at signup. The migration guide exists because the exit tax is real everywhere; the honest meter is the one whose exit costs an export and an afternoon.

The honest bill: lines that publish themselves

The alternative shape makes the audit list collapse by refusing the disguises. Seats: $15/mo beyond three free — admins, viewers and finance leads included at the same chair price, prorated daily, no tier ride. AI: no subscription, no pack walls — drafts (3), replies (6), translations (3), summaries (6) at prepaid credits (≈1¢) that never expire; a slow month's line is near zero without vanishing anything. Storage: 100 MB included, $1/GB beyond, published. The stack: CRM with auto-linked conversations, quotes, orders, invoices, contracts, help center, visitor tracking, team chat — one product, no connector lines. The exit: full export from settings, unthrottled — no hostage clause. The first payment: removes the watermark permanently. That is the whole price page. A team can audit it in one minute — which is the actual test of an honest bill: not that it is cheap (cheap is a marketing word), but that its lines fit on a business card and none of them hide.

A real desk's audit, worked in full

Run the audit once on paper so it can be run forever in minutes. The desk: eight humans (six agents, one lead, one finance lead who touches billing threads weekly), 2,800 conversations monthly peaking at 4,200 in Q4, AI at a 25% target share, attachments moderate. Line 1 — seats: advertised $49 × 5 "seats" for the agents; honestly licensed, the lead and finance lead are seats too — 8 × the tier the feature list forces (routing needed) = 8 × $79 = $632/mo. Line 2 — AI: assistant add-on $30/seat or packs: at 25% of 2,800 = 700 resolutions × $0.99 ≈ $693/mo average, ≈$1,040 peak. Line 3 — the gap tools: CRM with pipeline for the founder's side (from another vendor, $60), invoicing (that tool's seat, $30), WhatsApp connector ($25). Line 4 — the tail: setup week (a person-week, once), renewal repriced +12% in year two, exits throttled to CSV. The real bill: ≈$1,450/mo average, ≈$1,800 peak, plus the tail — for a desk whose advertised line read "$49/seat". The flat meter's version of the same desk: five paid seats = $60, credits for peak AI usage ≈$90, storage inside the band, no connector lines: ≈$150/mo, every line published. Same desk, same work — different decade. (The models guide generalizes the method.)