Control AI cost

Who this is for: anyone sending thousands of emails who wants the AI bill to stay predictable.

AI cost comes from tokens: the text the AI reads (input) and writes (output). About 4 characters make one token. Fewer, shorter, better-aimed reads and writes cost less, with Swift Pilot credit and with your own key alike.

Where the tokens go

For each lead the AI reads your instructions, your global instructions, the lead’s row and, with research on, the website text. Then it writes a subject and an email. Website text is usually the biggest part of the input. The Swift Pilot usage table shows this: Website research calls read the most.

The levers, from biggest to smallest

1. Research mode (per campaign)

2. Pages and text per lead (campaign, Advanced: limit AI usage)

• Website pages read per lead: each extra page adds text to read. 1 or 2 pages is enough for most small businesses. Swift Pilot always reads 1.
• Website text read per lead: the cap on how much text is kept. Lowering it from 8,192 to 4,000 roughly halves the research input.
• AI requests per lead: the most requests one lead may use. The default of 5 is fine; lower it to 3 to put a hard ceiling on each lead.

3. Saved websites (Settings, Website research)

Remember a website for (days) reuses what was read before instead of reading it again. Raise it when you run several campaigns on the same leads within a few weeks. Do not set it to 0 unless you need fresh reads every time.

4. Email length (Settings, AI connections)

Maximum email length (tokens) caps each answer. Instructions like “under 90 words” keep emails short anyway; the cap is the safety net.

5. Clean lists

Every lead you remove before the campaign is an email you do not pay to write. Remove leads with no email address and Invalid addresses first. See Clean up a list.

6. Test on a few leads first

Start new campaigns with I approve each email first and check the first emails before writing thousands.

With Swift Pilot

• Watch Swift Pilot usage with the This month filter: the Cost tile is your spend so far.
• Your monthly credit is used first; bought credit only after. Bought credit never expires, so a larger top-up is not wasted.
• When the credit runs out Swift Pilot pauses, it never overspends.

With your own key

• Set a Monthly token budget on each connection. When it is used up, the next connection in your list writes.
• The row shows This month: N requests · tokens of budget. Check it weekly at first.
• Put a cheaper model first for big campaigns, and keep a stronger one second as a fallback.
• Set a spending limit in your provider’s own dashboard too, as a second safety net.

Tip: A rough budget: 2,000,000 tokens is about 2,500 emails. Compare that with your provider’s price per million tokens to estimate the cost of a campaign before you start.

Common mistakes

• Researching 10 pages per lead “just in case”. Most of the useful detail is on the first page or two.
• Setting “Remember a website” to 0 while re-running campaigns on the same list.
• Writing for leads that will be skipped. Verify and clean the list first.

Next step

Make every token count: Write prompts that convert.