An AI collection call works by using what your books already know before it dials, and recording what the party says after it hangs up. Before each call, the AI agent pulls more than 20 inputs from Tally and your settings: your business name, the party's name and number, the amount due and the bills behind it, the language set for the company, the ageing slab, due dates, the assigned collector, earlier promises, the last reminder sent and the credit terms. On the call, it names your business, states what is due and asks when it will be paid, in the party's language. After the call, it returns more than 30 outputs. The most important is the outcome: a promise with a date, a dispute, a callback request or an escalation. Duration, connection status, any part-payment offer, the reason for delay and the payment mode mentioned are logged too, against the party's record. This post walks through each stage.
A collection call is only as good as two things: what the caller knows before dialling, and what gets written down after hanging up. Most office calls are weak on both. The caller opens a ledger five minutes before, and the outcome lives in their memory.
Takkada's AI agent is built around those two ends.
Before the call: 20+ inputs
Every call starts with more than 20 parameters pulled from your books and settings, so the agent never calls blind. They include:
- Your business name, so the party knows who is calling
- The party's name and phone number from the Tally ledger
- The amount due, and the bills behind it
- The language set for that company
- Ageing slab, due dates, collector assigned, previous promises, last reminder sent, and credit terms
Because the inputs come straight from Tally, the amount the agent quotes matches the ledger. Nobody reads out a number from last week's printout.
During the call
The call itself is short. The agent names the business, says what is due, and asks when it will be paid. It speaks in the party's language.
🎧 Listen to a demo AI call
After the call: 30+ outputs
This is where most calling falls apart, and where the AI agent does its real work. Each call returns more than 30 parameters. The most important is what the party committed to:
- A promise with a date, such as "Will pay by 22 Sep"
- A dispute, so someone can look at the bill instead of calling again
- A callback request, with when to call back
- An escalation, for parties who need the owner's attention
- The call duration, connection status, partial-payment offer, reason for delay, sentiment, and payment mode mentioned
All of it is logged as a follow-up against the party. With the Recovery dashboard on, it shows on their timeline and on the recovery board, the same place your team's own calls are logged.
Why the outputs matter more than the call
A call that ends with "theek hai, dekhte hain" and nothing written down is wasted. A call that ends with "promised ₹45,000 by Friday" on the party's record changes what happens next. The next reminder can reference the promise. The owner can see which promises are due this week. A dispute reaches the person who can fix it.
That is the difference between making calls and running collections.
What it costs
Charged per connected minute, with the cost shown before you confirm. Switched on per company on request, as an add-on on any plan, or as one of the 4 modules you pick in Enterprise.