Trust Error — Not Verifying Facts and Numbers
Why you shouldn't blindly trust an agent's facts and figures, where it slips most often, and the verification habits that save you from costly mistakes.
You set up your agent, everything works — and suddenly credits run out sooner than expected. Or you look at your usage and think "where did it all go?" ASCN has two types of credits, and how fast they're consumed depends on how you use your agent.
The good news: most cases of overspending are easy to prevent.
What consumes credits
Credits are spent every time the agent processes a request: executing a scheduled task, responding in chat, or reaching out to integrations for data. The more complex the task and the more frequently it runs — the higher the consumption.
Pattern #1: Schedules that run too often
Mistake: A task set to "check for new emails every 15 minutes"
That's 96 runs per day. Over a month — nearly 3,000 just for email monitoring alone.
Solution: Ask yourself: do you really need notifications every 15 minutes? Or is every hour perfectly fine? For most monitoring tasks, the difference between "every 15 minutes" and "every hour" is barely noticeable in practice, yet credit consumption differs by 4x.
Rule: Set the minimum necessary frequency, not the maximum possible one.
| Task type | Excessive | Optimal |
|---|---|---|
| Email monitoring | every 15 min | 1–2 times per hour |
| Lead checking | every hour | every 2–3 hours |
| Morning digest | 3 times a day | once a day |
| Weekly report | every day | once a week |
| Deadlines | every hour | once a day |
Pattern #2: Chatting for the sake of chatting
Mistake: Using the agent as a chatbot for entertainment or unnecessary testing.
"Hey, how's it going?" "Tell me a joke" "Explain quantum physics" — each of these requests spends credits. If you're testing how the agent responds to various questions purely out of curiosity — that's valid, but it's still consumption.
Solution: Use the agent for real work. Testing prompts is a justified expense. Casual conversation is not.
Pattern #3: Vague tasks that trigger multi-step clarifications
Mistake: A vague instruction causes the agent to ask many clarifying questions or cycle through options.
Example: "Do something useful with the data from the spreadsheet"
The agent reads the spreadsheet, doesn't understand what's needed, asks for clarification, you respond, the agent reads the spreadsheet again, asks again. That's 5–6 exchanges instead of one execution.
Solution: A clear instruction from the start = one request = minimal consumption. (See the lesson on Mistake #1.)
Pattern #4: Duplicate tasks
Mistake: Multiple tasks that do the same thing or fetch the same data.
For example: a morning digest reads Gmail, and a separate task also reads Gmail every hour. Both tasks do something similar — the agent hits the same source twice.
Solution: Consolidate similar checks into a single task. One "morning digest" task that covers everything — email, calendar, spreadsheets — is better than three separate tasks for each source.
Pattern #5: Tasks that never find anything
Mistake: A task runs on a schedule, checks a condition, finds nothing, but credits are still spent.
For example: "Every hour, check whether an urgent order over 500K has come in." Such orders come in once a month. 720 runs per month, 1 useful one.
Solution: For rare events, use triggers instead of schedules (if the platform supports it), or reduce frequency to something reasonable. Keep hourly checks only for things that actually change frequently.
Credit efficiency checklist
Once a month, review your tasks:
- ☐ All scheduled tasks are set to the minimum necessary frequency
- ☐ No duplicate tasks working from the same source
- ☐ Similar checks are consolidated into a single task
- ☐ No tasks that almost never find anything (or their frequency has been reduced)
- ☐ Not using the agent's chat for off-topic requests
How to consolidate multiple checks into one task
Instead of three tasks:
- Task 1 (every hour): check Gmail for new submissions
- Task 2 (every hour): check for new rows in Sheets
- Task 3 (every hour): check deadlines in Calendar
One task (every 2 hours):
Check: 1) Gmail — new submissions in the last 2 hours, 2) Google Sheets tab "Orders" — new rows in the last 2 hours, 3) Calendar — today's deadlines that haven't been marked yet. If anything is found — send a notification to Telegram. If nothing — don't write anything.
Result: 3 tasks → 1 task, consumption reduced by 3x, frequency cut in half.