The first month is for establishing what normal looks like. Without a baseline, every later number is uninterpretable — you will not be able to tell an improvement from noise, or a degradation from a hard week.
Take the baseline before you start
The thing people skip and regret. Before the first batch, write down what the manual process costs today:
- How many people get contacted per week.
- How long a screening call takes, including notes.
- What share of dials reach a person.
- How long from application to first contact.
Rough figures are fine. Once automated calling is running, nobody remembers what the old numbers were, and every claim about improvement becomes an argument.
The five to track
1 · Connect rate
Your list and your timing, before the agent has said anything. Everything downstream multiplies by it.
2 · Average duration
Your direct cost lever. Rising duration with flat outcomes means the agent is talking too much — a prompt fix.
3 · Completion rate
What share of connected calls got through all your questions. Low means the prompt is losing people — too long, too abrupt, or asking the awkward question too early.
4 · Rating spread
Not the average. If everything lands on 3 and 4, the rubric is not discriminating and your shortlist is arbitrary.
5 · Cost per useful outcome
Total spend ÷ people who reached the next stage. The only number that answers "is this worth doing", and the one to take into a budget conversation. Usually has to be worked out by hand.
Read transcripts, weekly
Ten at random. Every call under sixty seconds. Every call that scored in the middle band.
An hour a week, and it will tell you things no dashboard shows — that the agent is mispronouncing your company name, that everyone asks the same question you never anticipated, that the third question confuses people.
Change one thing at a time
The temptation after week one is to rewrite the prompt, swap the voice, retune the timing and change the rubric at once. Then you cannot attribute anything.
One change per batch. It feels slow for three weeks and then you actually know what works.