Tallyby Outpost

Theme

Start for free

AI products

For AI agents: which demo or listing brought the team now paying per task.

An AI agent gets bought after someone watches it finish a real task, checks what it can touch and runs it on their own work; billing then follows the tasks it does. Tally links the first payment to the visit behind it and keeps every metered cycle with that source, so you can see where your paying teams found you.

Start for free

14 days free, no card needed.

01How they sell

  • UsageCharges per task, run or minute of work, metered against a monthly cycle, so a customer's bill grows as the agent takes on more of the job.
  • SubscriptionA base plan with some runs included and overage on top; teams tend to switch to yearly once the agent has earned its keep.
  • One-offA paid pilot or setup fee charged once before a team commits, which can be larger than months of usage and deserves reading apart.

02Payments it reads

  • StripeMetered prices on subscriptions plus invoices for pilots; a paid invoice is read like any checkout payment.
  • PolarAgents sold to developers bill metered usage through Polar; every cycle with its usage charge is read, and lost disputes count as chargebacks.
  • PaddleTeams buying from many countries pay through Paddle as merchant of record; transactions and renewals are read, refunds as adjustments.

What Tally can't read

  • What the agent did on each run, its steps, tokens or failures, isn't read; only the charges your provider bills for that work.
  • A pilot paid against a purchase order by bank transfer, outside a connected provider, isn't read.
  • A company buying through a cloud provider's marketplace pays that marketplace rather than your checkout, so that revenue isn't read.

03Where buyers come from

  • Recorded runsA screen recording of the agent finishing a real task sends viewers who come to check price and permissions; path to purchase shows how many visits follow.
  • Agent directoriesAgent and AI tool directories send comparers in bulk; revenue by source ranks each listing by what its visitors went on to pay.
  • MCP registriesDeveloper-facing agents listed in MCP registries draw people who read the docs before trying anything, and the registry shows as their referrer.
  • Hacker NewsA launch thread brings skeptical developers who probe the docs and security page; revenue per visitor says whether the thread sold anything.

04What Tally answers

  1. Which demo brought the team that now pays the most?

    Open that customer's first sale. Its receipt lists the visits before the payment, the credited one, the link method and its confidence, and every metered cycle since carries the same link.

    Receipts

  2. How long does a team evaluate before the first charge?

    Path to purchase gives the median days and visits between first visit and first payment, plus the pages buyers saw, such as permissions or security, on the way.

    Path to purchase

  3. Where do teams drop out between the demo and connecting a tool?

    Send tally('connected', { tool }) when an account is linked and build a funnel from demo page to sign-up to connection; each step's drop shows where trust runs out.

    Goals and funnels

  4. Do teams from the MCP registry spend more over time?

    Lifetime value sorts customers by first source with lifetime revenue, average, renewals and repeat rate. With per-task billing, a rising average means those teams handed the agent more work.

    Lifetime value

  5. Can our own agent pull these numbers?

    Yes. Tally has a REST API, a command line and an MCP server, so an assistant or an internal agent can ask for revenue by source without anyone opening the dashboard.

    API, CLI and MCP

05Numbers worth watching

Median time to buy
Days from first visit to first charge; a pilot stretches it, so read it with the pages buyers saw before paying.
Lifetime value by first source
Average lifetime revenue by first source, which under per-task billing reflects how much work each channel's customers trust the agent with.
Coverage
Invoices your team raises by hand carry no visitor id; coverage shows how much revenue that leaves unlinked and why.

06An example

Example

Path to purchase

Last 30 days · 105 of 124 new sales, the ones linked to a visitor

Median time to buy

3days

from their first visit

Median visits

3visits

the one they paid in included

Bought on the first visit

12%

13 of 105

Pages before buying

11

median pages viewed

Where buyers came from, and who closed the sale

105 new sales: where each buyer first came from, and the source credited with the sale.

50% of new sales were credited to the source that first brought the buyer in. Hover or focus a flow for its numbers.

  1. Googlecredited toGoogle26
  2. x.comcredited tox.com10
  3. Other (3 sources)credited toOther (3 sources)10
  4. Directcredited toOther (3 sources)9
  5. ChatGPTcredited toChatGPT9
  6. Other (3 sources)credited toNewsletter9
  7. Redditcredited toReddit8
  8. Googlecredited toNewsletter6

6 smaller flows in the table below.

Show the flows as a table
105 new sales: where each buyer first came from, and the source credited with the sale.
Brought them inCredited with the saleNew salesShare
GoogleGoogle2625%
x.comx.com1010%
Other (3 sources)Other (3 sources)1010%
DirectOther (3 sources)99%
ChatGPTChatGPT99%
Other (3 sources)Newsletter99%
RedditReddit88%
GoogleNewsletter66%
DirectGoogle55%
x.comGoogle55%
RedditGoogle33%
DirectReddit22%
ChatGPTGoogle22%
Other (3 sources)ChatGPT1<1%

Time from first visit to purchase

  1. 13Same visit: 13 sales (12%), bought in their first visit
  2. 13Same day: 13 sales (12%), a later visit, under 24 hours after the first
  3. 301–3 days: 30 sales (29%), 1 to 3 days after the first visit
  4. 194–7 days: 19 sales (18%), 4 to 7 days after the first visit
  5. 151–2 weeks: 15 sales (14%), 8 to 14 days after the first visit
  6. 102–4 weeks: 10 sales (10%), 15 to 28 days after the first visit
  7. 51–3 months: 5 sales (5%), 29 to 90 days after the first visit
  8. 03+ months: 0 sales (0%), 91 days or more after the first visit

Pages buyers saw before paying

Share of the 105 buyers who viewed each page before paying

  1. /pricing96%, 101 of 105 buyers
  2. /71%, 75 of 105 buyers
  3. /signup63%, 67 of 105 buyers
  4. /docs38%, 40 of 105 buyers
  5. /blog/stripe-fees26%, 28 of 105 buyers
  6. /compare21%, 23 of 105 buyers
  7. /changelog8%, 9 of 105 buyers
  8. /blog/launch-week5%, 6 of 105 buyers

Only new sales count here. Renewals and upgrades belong to the visit that won the subscription, so counting them would count that journey again.

Fig. 1Path to purchase on the example site. For an agent, look at the pages buyers saw before paying: when permissions and pricing pages lead, trust is the deciding step.

07Setting up

  1. PolarOpen Polar checkout with tally_visitor_id in its metadata; every metered cycle after the first payment keeps that link.
  2. StripeFor pilots invoiced in Stripe, call tally('identify', { email }) when the evaluating user signs in, so the invoice can link by the buyer's email.
  3. Put the tag on the docs and demo pages as well as the homepage; evaluators spend most of their visits there.

08Questions

  • Our agent bills per task. Is each charge a separate sale?

    Each payment your provider records is read. A metered cycle on a subscription comes in as a renewal and keeps the first payment's source; a one-off charge is a sale of its own.

  • Can Tally tell us which customers ran the agent most?

    No. It sees visits and payments, not runs. When usage is billed, the bill reaches Tally as revenue; the runs stay in your own logs.

  • Does a pilot we invoice by hand count?

    If it's a Stripe invoice, yes: it's read as a payment. It links by email when the buyer was identified on your site; otherwise it's counted as unlinked, never guessed.

More in ai productsEvery business