How to run a 30 day pilot of an AI SDR

Run a 30 day AI SDR pilot by deciding the goal, the baseline and the pass rule on day one, starting with one audience and approval mode, measuring replies, positive replies and booked meetings the same way each week, and reading the conversations. Sellinger offers 30 days free on messaging modules and bundles, no card. A pilot ends in a decision: scale, change or stop.

Most AI SDR pilots end in a shrug. The vendor says it went well, the buyer is not sure, and nobody wrote down in advance what well would mean. This guide gives a plan that avoids that: a goal and a pass rule before you start, a small first wave, a weekly rhythm and a decision at the end. It works for any AI SDR, ours included, and it ends with questions to ask the vendor.

Day 0: decide what you are testing

Write five lines before anything is connected.

1. The goal. "Book qualified first calls with heads of operations at logistics firms in Bulgaria."
2. The audience. One segment, with a sentence that defines it.
3. The baseline. What your current process produces: replies, calls booked, hours spent. If you have no baseline, run a small manual batch in week one.
4. The pass rule. The result that would make you scale. For example: "At least X positive replies per 100 contacts, and at least Y calls booked, with no complaints I cannot explain."
5. The stop rule. What would make you end early: complaints, restrictions on an account, a wrong claim sent.

Writing the pass rule before the data arrives is what keeps the pilot honest. After the data arrives, every result looks like a success to someone.

Week 1: set up and read

Connect one channel, not five. For a B2B pilot that is usually LinkedIn or email. Set up the agent: persona, rules, knowledge base, calendar link. See set up your first AI SDR in one afternoon. Sellinger is usually live within a day, and we have not timed a standard setup, so allow for your own.

Then turn on approval mode so every message waits for a person. Send the first 20 to 30 messages and read each one.

Check:

• Is every statement about the prospect true?
• Does it carry any claim you cannot support?
• Does it sound like you?
• Does it ask one question?

Fix the rules, not the individual messages. If the agent made the same mistake twice, the rule is missing.

Week 2: widen the sample

Release the rest of the first wave, keeping approval on for new message types. Aim for a few hundred contacts in one segment. Read a sample of threads every day, not just the replies. Look for:

• wrong person, wrong company, wrong language
• replies the agent mishandled
• replies that needed a person and were not flagged

Check the limits. Sellinger's LinkedIn defaults are 20 invitations a day and 150 a week, and email defaults to 50 sends a day per mailbox. A pilot that tries to hit a big number in week two is testing your account's tolerance, not the AI.

Week 3: follow ups and the first results

By week three the sequences are completing. This is where much of the value lies: Instantly's 2026 report puts 42% of cold email replies on follow ups, and our own LinkedIn data shows follow ups adding 45% more replies. A pilot that stops after the first touch will understate the result.

Count, for the segment so far:

Use the same definitions every week. Our post on how we define reply rate sets them out, and counting meetings honestly explains why booked and held differ.

Week 4: compare and decide

Put the numbers next to the baseline and the pass rule you wrote on day zero. Then read about 30 conversations in full, chosen at random, not the best ones.

Three outcomes.

Be careful with a surprising win. Our own data shows how results move: replies held at about 40% all year while meetings per person messaged fell from 2.4% to below 1%. One good month is not a trend. See the case study.

Who should be in the pilot?

Three roles, even in a small company.

• An owner who wrote the pass rule and makes the decision at the end.
• A reader who goes through the conversations every day in week one and every few days after. This is often the person who will work with the leads later.
• A closer who takes the booked meetings and says whether they were good. A meeting that the closer calls a waste is not a success, whatever the calendar says.

In a one person company, you are all three. Block the time in your calendar. A pilot without a reader is an experiment nobody watches, and its failures arrive as complaints from prospects.

What should you ask the vendor?

1. How does your tool connect to my channels, and what limits does it enforce?
2. Can I see every message before it is sent?
3. What happens when a person replies "stop"?
4. How do you count a reply and a meeting?
5. What does the agent do when it cannot answer?
6. What is the exit: can I export my data and cancel without lock in?

Our buyers checklist for AI SDRs has the long list, and whether AI SDRs actually work gives our honest answer.

What does a pilot cost?

Sellinger offers 30 days free on messaging modules and bundles, with no card. Usage modules such as enrichment start with a $10 credit. You can cancel, upgrade or change modules at any time, with no lock in contracts and no setup fees on self serve plans. See the pricing page for current prices. The real cost of a pilot is your time: setup, reading threads and the weekly review. Plan an hour a day for the first week and less after.

What are the common pilot mistakes?

• No pass rule. The result is interpreted afterwards.
• Too many changes at once. New list, new message and new channel in a week.
• Ending after the first touch. Follow ups are where replies come from.
• Counting booked as held. A calendar says completed when nobody came.
• Skipping the conversations. Numbers do not show a rude or wrong message.
• Piloting on your best account. If something goes wrong, you risk your best relationships. Pick a segment you can afford to be wrong about.

A 30 day checklist

• [ ] Day 0: goal, audience, baseline, pass rule, stop rule written.
• [ ] Week 1: one channel, approval on, first 20 to 30 messages read.
• [ ] Week 2: a few hundred contacts, threads read daily.
• [ ] Week 3: follow ups complete, definitions fixed.
• [ ] Week 4: compare, read 30 conversations, decide: scale, change or stop.