The control layer for AI SDRs: seven rules to write before the agent sends anything

An AI SDR can research an account, write a first email, and send it, hundreds of times a day. Most of the setup work goes into the writing: tone, templates, what to mention. Much less goes into the questions a sales manager would ask a new hire on day one. Who are you allowed to contact? Who do you start with? When do you come and get me?

Those answers are the control layer. They are a short set of rules, written by a person, that sit between your CRM and the agent. This playbook gives you seven of them, each tied to a score so the agent has something exact to check.

What you need before you start

Three things have to exist on the record before any rule can work:

• A score. A number out of 10 on each contact or account.
• The reasons. The factors behind the score, on the same record.
• The time it was calculated. So the agent can tell a fresh score from a stale one.

If the agent has to work these out for itself, you do not have a control layer. You have an agent grading its own homework.

The seven rules

Rule 1. Set who it may contact

Pick one threshold. Below it, the agent sends nothing.

• The rule: contact only records that score 5.0 or higher.
• Why: an agent has no sense of when a record is not worth the email. Volume is free to it. Your sender reputation is not.
• Write it as: “If the score is below 5.0, do not contact. Record that you skipped it and why.”

Rule 2. Set who it must never contact

Some records are off limits whatever they score.

• The rule: no outreach to current customers, competitors, partners, or anyone on an open deal that a rep owns.
• Where it lives: as disqualifying factors in the scoring model, so they show on the record, and as a second check in the agent’s instructions.
• Write it as: “If any disqualifying factor is present on the record, stop. Do not draft. Log the factor.”

This rule is deliberately doubled up. A disqualifying factor lowers the score, which usually takes the record under the threshold on its own. The explicit check catches the case where it does not.

Rule 3. Set the order, and the daily limit

• The rule: work records from the top score down. Stop at the daily limit.
• Why: without an order, the agent picks one. Without a limit, it works its way down to records nobody would call.
• Write it as: “Sort by score, highest first. Send to at most 37 new contacts a day. If you run out of records above the threshold, stop. Do not go below it to fill the day.”

The number 37 is an example. Set yours from what your reps can follow up on, not from what the agent can send.

Rule 4. Set what it says

• The rule: the first line of the email comes from the top factor on the record.
• Why: this is what makes the message specific without the agent inventing a reason to write.
• Write it as: “Read the factor with the largest contribution. Open with it in plain words. Never mention the score, the model, or that the person was ranked.”

That last sentence matters. A buyer should never read “you scored 8.3”.

Rule 5. Set when it hands to a person

• The rule: from 8.0 and up, the agent drafts and a rep sends. Any reply goes to a person.
• Why: your best records deserve a human, and they are the ones where a clumsy email costs the most.
• Write it as: “If the score is 8.0 or higher, draft the email and assign it to the account owner for review. Do not send. If a contact replies, stop the sequence and notify the owner.”

Rule 6. Set what it does when data is missing

• The rule: if a factor contributes 0 because the field is empty, the record goes to enrichment, not to outreach.
• Why: a middling score caused by a blank field is a different thing from a middling score caused by a poor match. Only the reasons tell you which.
• Write it as: “If a factor is 0 because its field is empty, do not contact. Add the record to the enrichment list and name the empty field.”

Rule 7. Set what it writes down

• The rule: every action and every skip leaves a line.
• Why: when someone asks why a record was or was not contacted, the answer should take a minute to find.
• Write it as: “For each record, log the score, the top factor, the action you took, and the time.”

The whole thing on one page

These are example ranges. A band is a range of scores, and each band gets one play.

Band	What the agent does	What a person does
8.0 and up	Drafts the email from the top factor. Does not send	Reviews, sends, and owns the follow-up
6.5 to 7.9	Sends the first email and runs the full sequence	Takes over on any reply
5.0 to 6.4	Sends one short email. No sequence	Nothing, unless there is a reply
Below 5.0	No contact. Logs the reason	Reads the skip log each week
Any band, disqualifying factor present	Stops. Logs the factor	Nothing
Any band, a factor at 0 from an empty field	Adds the record to the enrichment list	Fills the field, or asks why it is empty

The rules, run on one record

Demo contact 6.9/10

• Account fit (from the company record) 35% 7.4
• Role and seniority 30% 7.1
• Department 20% 6.8
• Region 10% 9.1
• Personal email address 5% -2.6

The math: 7.4 at 35% is 2.59, 7.1 at 30% is 2.13, 6.8 at 20% is 1.36, and 9.1 at 10% is 0.91. That adds to 6.99. The disqualifying factor subtracts 2.6 at 5%, or 0.13, for 6.86, shown as 6.9.

Walk the rules. The score is above 5.0, so the contact is allowed. No factor is empty. It is below 8.0, so the agent may send. The largest contribution is the account’s fit, so the email would open with why the company is a match, not with the person’s title.

Then rule 2. A disqualifying factor is present. It is a small one, a personal email address, and under the rule as written the agent stops.

That is the playbook working. If you would rather the agent sent in this case, change rule 2 to name the factors that block outreach, and leave this one off the list. The point is that a person makes that call once, in writing, and not the agent, record by record.

Who owns the rules

Give the control layer one named owner, usually in revenue operations or sales leadership. Three habits keep it honest:

1. Read the skip log weekly. If good records are being skipped, a threshold or a factor is wrong.
2. Read a sample of sent emails weekly. Check that the first line matches the top factor.
3. Change one rule at a time. Move a threshold, wait a week, then look.

Where BoostIdeal fits

BoostIdeal supplies the first half: the score, the reasons, and the time of the last calculation, written to HubSpot properties. It can also share those reasons wherever your rep works, such as in Slack or by email. You build the model and set the weights, with AI helping you analyze your data and draft. It scores leads, contacts, accounts, and custom objects out of 10, and can score a contact using the fit of its account, as in the example. Scores recalculate hourly by default, and in near real time with webhook sync.

It does not supply the second half. BoostIdeal does not run an AI SDR, send email, or enforce the rules above. Those live in your agent and your workflows. It also does not change the weights on its own, so the priorities the agent follows stay the ones you wrote.

If the model itself is not written yet, begin with your ideal customer profile and the five-step framework for building an ICP scoring model.

Frequently asked questions

What is a control layer for an AI SDR?

It is the set of rules that decide who the agent may contact, in what order, what it says, and when it hands to a person. The rules read a score and its reasons on each record.

Should an AI SDR contact every lead?

No. Set a threshold and send nothing below it. An agent can send to everyone at no extra effort, which is the reason to stop it.

When should an AI SDR hand off to a human?

On any reply, and for your top band before the first email goes out. In the example here, records at 8.0 and up are drafted by the agent and sent by a rep.

Should the email mention the lead score?

No. Use the top factor to decide what to say. Never tell a buyer they were scored or ranked.

Does BoostIdeal include an AI SDR?

No. It provides the score and the reasons on the HubSpot record. Your AI SDR and your workflows act on them.

The takeaway

An AI SDR needs the same seven answers a new rep would: who is allowed, who is off limits, where to start, what to say, when to get a person, what to do with gaps, and what to write down. Put each answer in writing, tie it to a score and its reasons, and give the rules an owner.