Key takeaways
- Output quality depends on the inputs more than on the model.
- A specific ICP with disqualifiers beats a broad description of your market.
- Proof must be real and approved, or the agent will make claims you cannot back.
- Exclusions protect customers, open deals and partners from cold outreach.
- An AI SDR reads your context while it works; it does not need your data to train a model.
Which AI SDR inputs decide the quality of its outreach
Six inputs decide the quality of an AI SDR's outreach, and each one missing produces a recognisable weakness.
- Ideal customer profile: without it, the agent targets everyone and convinces no one.
- Offer: without a plain description of what you sell and the problem it solves, messages go vague.
- Proof: without approved examples and facts, the agent either says nothing concrete or invents it.
- Brand voice: without it, every email sounds like every other AI email.
- Exclusions: without them, customers and open deals get cold pitches.
- Conversation history: without it, follow-ups repeat what the buyer already heard.
How do you set up an ICP for an AI SDR
Set up an ICP for an AI SDR by describing who buys, why now and who never buys, in that order. Industry, company size and region narrow the market; the roles involved and the triggers that create urgency, such as a new hire or an expansion, tell the agent what to write about.
Disqualifiers matter as much as the profile. An illustrative B2B software firm might exclude companies under a certain size, public sector bodies and anyone already using a competitor on a long contract.
Does an AI SDR need training data
An AI SDR does not need training data in the sense of retraining a model on your files. It needs context that it reads at the moment it works: your company description, your offer, your voice and the history of each contact.
That distinction matters for security reviews. In Pebbles Ai, no customer's data trains a model; the assistants learn your company as you work, from Company Intel and Agent memory.
How do you give an AI SDR your brand voice and proof
Give an AI SDR your brand voice with a short written description and a few examples of messages that worked, plus a list of words and claims to avoid. Examples teach tone faster than adjectives do.
Keep proof as a list the agent may quote: named customers you have permission to mention, product facts and the outcomes you can stand behind. Anything not on the list stays out of the email.
How do you set up these inputs in Pebbles Ai
In Pebbles Ai, the inputs live in Company Intel, the CRM (labelled Beta) and Agent memory, and the assistants that research and write outreach read all three. Company Intel holds Business knowledge (Company, Approach, Offerings) and Brand voice (Conviction, Audience, Message, Expression), built from your website, company information, files and pasted text.
Ideal customer profiles, notes and contact history sit in the CRM. Short facts you state or confirm in chat are kept in Your memory, Project memory or Company memory, so the next conversation starts with them.
Where Pebbles Ai fits
- FoundationCompany IntelCompany Intel builds Business knowledge and Brand voice from your website, company information and files, and every assistant writes from it.
- FoundationAgent memoryAgent memory keeps short facts you share, at Your memory, Project memory and Company memory level, and you can correct them in plain words.
- CapabilityProspecting & customer recordsThe CRM (in Beta) holds ideal customer profiles, notes and activity, so drafts can reflect what was already said to each contact.
Common questions
Is my company data used to train the AI model?
Not in Pebbles Ai. Your company information is never used to train an AI model; the assistants read it as context while they work.
How often should I update an AI SDR's inputs?
Whenever the offer, the target market or the proof changes, and at least once a quarter as a check.
Can an AI SDR work without CRM history?
Yes, for first-touch outreach. Follow-ups and replies get noticeably worse without it, because the agent cannot see what was already said.
Which input do teams most often forget?
Exclusions. Teams describe who to target but forget current customers, open deals, partners and competitors.
About this page. Written by the Pebbles Ai team for people who run sales, marketing and growth. Last updated .
