How to Understand Your AI Credits & Cost Page

Your AI Credits & Cost page is your one window into what your AI assistant is actually costing you, broken down by category, by day, and by conversation. Both WhatsApp and Instagram traffic show up in the same view.

Open it from the sidebar — Dashboard → AI Credits & Cost. The page has five sections, top to bottom: KPI cards (today, this month, projection, avg/msg), the 30-day timeline chart, the per-category balance bars, the 3-agent pipeline breakdown, and your top conversations by cost. There's also a "Request top-up" card that appears only when you're approaching a cap.

Every AI reply consumes credit from one or more of four buckets:

• Text AI — the brains. Includes the classifier (a fast model that decides how much thinking to do), the responder (writes the actual reply), and the verifier (double-checks the response). This is usually 70-80% of your bill.
• Vision — image understanding. Fires when a customer sends a product photo or you respond to an Instagram story. Pro+ plans only.
• Audio — voice-note transcription. Fires when a customer sends a voice note instead of typing.
• Documents — PDF and file OCR. Fires when customers send invoices, receipts, or forms.

Each bucket has its own balance bar, its own daily cap, and its own monthly cap. They're independent — running out of audio credit doesn't pause your text replies.

The 30-day chart shows your daily spend, stacked by category. Tall green days are conversation-heavy text days — that's normal traffic. Tall amber days are media-heavy days: image upload campaigns, lots of voice notes, or document drops.

Click any day-bar to jump to the conversations from that day. This is your fastest way to answer "why was Tuesday so expensive?" — usually it's one or two customers who sent a flurry of voice notes.

Botline's AI runs in three stages, not one. That's the secret to keeping costs down without giving up reply quality:

1. Classifier (Nova Micro) — a fast cheap model that looks at the incoming message and decides how much thinking the AI needs to do. A simple "what's the price?" gets a small budget; a complex multi-product order gets a big one.
2. Responder (your chosen model — Sonnet 4.6, Nova Pro, etc) — actually writes the reply, with the budget the classifier set.
3. Verifier (Claude Haiku) — reads the responder's draft and flags hallucinations, fabricated phone numbers, or off-script replies before they go out.

The cost-page row shows you each stage's contribution. Most of your text-AI cost is the responder; classifier and verifier together usually add ~10-15%. If verifier costs are creeping up, that's a signal worth investigating — usually means the responder is going off-script and getting regenerated.

The bottom table shows your most expensive conversations of the period, sorted by total cost. Audio-heavy and image-heavy chats consistently cost 3-5× more than text-only ones.

Use this view to spot patterns. If one customer accounts for 20% of your monthly AI spend, and they always send voice notes when text would be faster, that's an opportunity to politely ask them to type. Or to set a per-conversation rate limit.

The page automatically calls out conversations costing more than 2× the average per message — those are the ones worth looking at first.

Each category has both a daily cap and a monthly cap. The bars show used vs. allowed. Color-coded:

• Below 70% — green, you're fine.
• 70-95% — amber, watch closely.
• 95-100% — red, you're about to hit the cap.

Daily caps reset at midnight in your timezone (the same timezone Botline uses for your AI Reply Schedule), not server UTC. Monthly caps reset on the calendar month boundary.

Caps are set by your admin from the per-tenant page in admin tools. For now (Phase 1), the bars are display-only — they don't actually pause AI when you hit them. That's coming in Phase 2.

If you're approaching a cap before the period ends — say it's the 25th of the month and you've used 90% of your vision budget — click Request top-up on the bottom-left card. Pick a category (or "All categories"), choose daily or monthly scope, enter the dollar amount you need, and add a short reason.

Admin gets an email immediately. They can approve or deny from the admin AI Credits page. Approved top-ups show up on your bar within a minute and expire 30 days later (or whenever admin sets).

For 2026, top-ups are settled manually via your existing payment method (bank transfer / e-wallet / however you currently pay). Stripe usage-based metering is targeted for 2027.

When Phase 2 ships and a category cap is reached, Botline silently skips processing for that one media type. Your customer never sees an error — they just don't get an AI reply for that one image, voice note, or PDF.

You see a clear amber row in the conversation feed: "Image processing skipped — daily AI credit for vision exhausted. Top up →". Other categories keep working normally; if vision hits the cap but text is fine, your AI keeps replying to text questions.

Today (Phase 1), the bars only display — the pause logic comes in Phase 2. But the rails are already in the data so when Phase 2 ships you don't have to reconfigure anything.