Agent Optimization Guide
A simple way to improve an agent: measure where calls drop off, change one thing, and compare. This guide links to the articles that explain each number and setting.
1. Measure
Start with the numbers, not the script.
- Open the agent and look at Performance (last 7 days): Total dials, Answered, Transfers, AI conversion and Cost per transfer. See AI Agents Overview.
- For a campaign view, open the campaign's Reporting tab. See Campaign Report.
- Read 5 to 10 calls in Dashboard › Call log, good and bad. See Call Log.
Find the biggest drop:
| Where calls drop | Look at | Usually fixed by |
|---|---|---|
| People don't pick up | Connection Rate, Human Answer Rate | Campaign settings: caller ID, Ring Timeout, call times, data quality |
| People hang up early | Avg. answered duration, call transcripts | The opening line, Who Speaks First?, Response Speed |
| Conversations don't convert | AI Conversion Rate | Objectives, script, objection handling, knowledge |
| Transfers don't stick | Transfer Quality | Transfer type, hold prompt, your team's pickup |
| Each transfer costs too much | Cost Per Transfer | Everything above, plus retry settings |
Every tile and its benchmark is in the Metrics Glossary.
2. Change one thing
Pick the one change most likely to fix the biggest drop. Change only that, so you know what made the difference.
| To change | Where | Article |
|---|---|---|
| Opening, flow, objections | Objectives and script | Objectives and Scripts |
| Answers to common questions | Knowledge and skills | Agent Knowledge |
| Voice and tone | Voice and behavior › Voice, Voice Tuning | Voice and Persona |
| Turn-taking and pacing | Who Speaks First?, Response Speed, Interruption, Silence Timeout | Agent Behaviors |
| Hand-off to your team | Call transfer, Hold conversation | Transfers and Call Flow |
| Voicemail | Voicemail and messaging | Voicemail |
| Ring time, retries, call times | Campaign settings | Retry and Stop Rules, Dialer Concurrency |
Before the change goes live, run a Roleplay and try the case you're fixing. See Test Your Agent.
Tip from the Human Answer Rate tooltip: if the rate is low, check the ring-out time first. Short settings (16–20 s) cut calls before they connect. The campaign's Ring Timeout tooltip recommends 20–25 s.
3. Compare
You can compare two ways:
- Side by side (better): duplicate the agent, make the change on the copy, put both in one group and pick the group in the campaign. Compare them in the A/B group panel and each agent's Performance. See Agent Groups.
- Before and after: change the agent, then compare this week's numbers with last week's. Other things change too (lists, days, times), so treat the result with care. If it gets worse, put it back with Restore to here. See Agent Versions.
Give each version enough calls on the same kind of contacts before you decide. A single good day isn't a result.
When one version wins, keep it, archive the other, and start the next test. See Lock, Duplicate, Archive and Delete Agents.
Going further
- Economics: with the Economic Analysis add-on, see value and ROI by campaign. See Campaign Economics.
- Post-call analysis: with the Post-Call Analysis add-on, see why calls ended the way they did. See Post-Call Analysis.
What Taalk can't do for you (yet)
- No traffic split setting for A/B tests.
- No per-version performance report. Use two agents in a group to compare versions.
- No automatic suggestions on the agent page (Update with AI is Coming soon).