What does outbound AI cover, end to end?
Outbound has six stages, and a reply depends on all of them. Someone has to find a company worth contacting, decide whether it fits, reach the right person, say something relevant, get the email into the inbox, then deal with whatever comes back. Skip one stage and the others cannot make up for it.
Most tools sold as outbound AI automate one or two of those stages, usually writing and sending. The rest stays manual, spread across a lead database, a spreadsheet, an email finder and a shared inbox. Startories runs the whole chain in one place, which is also why it can show you, for any email, where the lead came from and why it was picked.
| Stage | The question it answers | What Startories does |
|---|---|---|
| 1. Signals | Who has a reason to buy now? | Reads Reddit, X, Product Hunt, directories and niche search results for fresh buying signals |
| 2. Qualification | Does this company fit? | Matches the signal to a real company, checks your ICP and scores it with written reasons |
| 3. Contact | Who decides, and can we reach them? | Finds the decision-maker and verifies the business email |
| 4. Copy | What do we say? | Writes an angle around the event that triggered the lead |
| 5. Sending | Will it reach the inbox? | Sends from warmed-up inboxes on separate domains, with daily caps and follow-ups |
| 6. Replies | What happens next? | Classifies each reply and drafts a suggested answer |
Outbound AI, AI SDR, AI sales agent: which is which?
The labels overlap, so look at the job instead. "AI sales agent" is used for website chatbots, voice agents that place calls, research assistants and outbound systems alike. "AI SDR" usually means software that fills the prospecting role (what an AI SDR is). Here, outbound AI means all six stages above in one system. Startories is that system for email. If most of your buyers arrive through your website, an inbound chat agent solves a different problem.
What should AI do in outbound, and what should it not?
The useful question is not whether AI can do a task, but whether it should do it unsupervised. A simple test: hand AI the work that is high-volume and easy to check, and keep the work that needs judgment about your business.
Good jobs for AI
- Reading thousands of posts, launches and listings a day and flagging the few that matter.
- Applying the same ICP criteria to every company, without getting tired or lenient late on a Friday.
- Pulling the specific detail (the tool someone complains about, the role a company is hiring for) into a first line.
- Sorting replies at 2 a.m., so that an interested answer does not sit until Monday.
- Stopping a sequence the moment someone replies, unsubscribes or bounces.
Jobs that should stay with people
- Choosing the market, the offer and the promises you are willing to make.
- Deciding what a good lead looks like, and correcting the AI when it gets that wrong.
- Handling a nuanced objection, negotiating and running the sales call.
- Deciding when a funnel has earned the right to run without review.
Things outbound AI should never do
- Invent a reason to write, such as a fake referral or a compliment it cannot back up.
- Guess an email address and send anyway.
- Keep following up after someone said no.
- Hide why a lead was chosen.
How much control do you keep over the AI?
In Startories, human-in-the-loop is a setting, not a promise. For each funnel you choose whether first emails and replies wait for your approval or go out on their own within fixed sending limits. Most teams start strict and loosen over time:
Review the leads
Read the first batch with the reasons shown next to each company. If the AI kept agencies when you sell to SaaS, or skipped a region you want, fix the ICP before a single email goes out.
Approve the first emails
Edit or approve each draft. This is where you catch a claim you cannot make or a tone that does not sound like you.
Approve the replies
Each reply arrives labeled (positive, interested, neutral, negative, out of office, unsubscribe) with a suggested answer. Send it, edit it or write your own.
Let proven funnels run
Once a funnel produces leads you agree with and emails you would have written yourself, let it run automatically within its caps, and keep reviewing new funnels the same way.
How much of your week does outbound AI take?
Take a six-person software development agency on the Growth plan, running two funnels: newly launched B2B SaaS products, and companies hiring their first data engineer. The pipeline runs every day; the table groups the work by day to show who does what during the first month, while approval is on.
| Day | What the pipeline does | What the agency owner does |
|---|---|---|
| Monday | Qualifies the launches and job posts found since Friday, then finds and verifies decision-makers | Skims the new leads and their reasons, and rejects the ones that miss |
| Tuesday | Drafts first emails around each launch or job post | Approves, edits or rejects each draft |
| Wednesday | Sends within each inbox's daily cap and labels replies as they arrive | Answers the replies marked positive or interested |
| Thursday | Sends follow-ups to people who have not answered, and stops any sequence that got a reply | Takes the first calls |
| Friday | Reports replies and meetings for each funnel | Compares the two funnels and notes what to change in targeting or copy |
After the first month
Once a funnel runs on its own, the owner's week shrinks to three jobs: answering replies, taking calls and the Friday comparison. The approval work moves to whichever funnel is newest. If even that is too much, the Growth plan includes ongoing optimization by the Startories team, and fully managed acquisition hands over the whole week.
What happens when a prospect replies?
The sequence stops the moment a prospect answers. The reply is then labeled and a suggested answer is drafted, so the inbox is sorted before you open it. The replies that need speed are the positive and interested ones, and those are exactly the ones that get buried in an unsorted inbox. Here is what each label looks like in practice, with made-up replies:
| Label | Example reply | What happens next |
|---|---|---|
| Positive | "Sure, does Thursday at 2 work?" | The suggested answer confirms the time; you send it and take the call |
| Interested | "How is this priced for a team of 12?" | An answer is drafted; you check the details only you know, such as price or roadmap |
| Neutral | "Not a priority this quarter. Maybe in January." | A short acknowledgment is suggested; you decide whether to come back later |
| Negative | "We built this in-house, so no thanks." | A polite close is suggested, and the sequence is already stopped |
| Out of office | An automatic reply until the 14th | Labeled as an auto-reply, so you know no person has answered yet |
| Unsubscribe | "Please remove me from your list." | The contact is suppressed across every campaign, current and future |
Reach companies with a reason to buy this week
Startories finds the buying signal, verifies the decision-maker and runs the outreach until they book a call.
Why is deliverability part of the AI's job?
An AI that can write five thousand emails a day can also burn a domain in a week. Inbox providers judge each sender on authentication, bounces and spam complaints. Google, for example, asks senders to keep the spam rate reported in Postmaster Tools under 0.1% and to avoid ever reaching 0.3% (Google email sender guidelines). Microsoft followed in 2025: domains that send more than 5,000 emails a day to Outlook.com addresses must pass SPF, DKIM and DMARC (Outlook requirements for high-volume senders). Volume makes those rules harder to keep, not easier.
So in Startories the sending rules belong to the pipeline itself, not to a separate tool you configure later:
- Outreach leaves from separate domains and inboxes that were warmed up first, so the address your team uses every day stays out of the line of fire.
- Each inbox has a daily cap, and follow-ups end as soon as the prospect answers.
- Sending pauses on its own when bounces start to climb, instead of you finding out from a blocklist.
- An address that cannot be verified is simply not contacted.
- Every message carries an opt-out, and anyone who unsubscribes or bounces is suppressed in every campaign.
How is outbound AI different from a cold email tool with an AI writer?
Cold email platforms such as Instantly, Smartlead or lemlist are very good at what they were built for: sending from many inboxes, warming them up and tracking campaigns at scale. Many now include an AI writer that drafts emails from the columns of your CSV. If you already have a researched list and someone who owns targeting, that setup is flexible and often cheaper.
The difference is what happens before the first send. An AI writer can only personalize from the data you give it, so personalization often means a first name, a company name and a line lifted from a website. An outbound AI pipeline decides who to write to and why, and that "why" becomes the email.
| Cold email tool with an AI writer | Outbound AI pipeline (Startories) | |
|---|---|---|
| Where leads come from | A list you import | Buying signals it detects in public sources |
| Why this company | Usually not recorded | Written reasons and a link to the original signal |
| Decision-maker and email | Found and verified before import, often in other tools | Found and verified inside the pipeline |
| What the AI writes from | The columns of your spreadsheet | The event that triggered the lead |
| Replies | A shared inbox, with labels that vary by tool | Classified, with a suggested answer |
| Best when | You have a list and a team to work it | You need the list, the timing and the outreach handled together |
What a do-it-yourself stack is made of
Each part below can be the best in its category. The hidden cost is the glue between them: exports, deduplication, re-imports and the hours someone spends moving rows from one tool to the next. If that someone is a founder, count those hours at what a founder's time is worth.
- A lead database or a scraper for the list.
- An email finder, and often a separate verifier.
- A sending platform with warm-up and several inboxes.
- An AI writer or a prompt sheet for first lines.
- A spreadsheet that remembers why each company was picked.
- A shared inbox, and someone to sort it.
How do you know if outbound AI is working?
Judge it stage by stage, not on sends or opens. Open rates are unreliable anyway, because some mail apps load tracking pixels on their own. For each funnel, track five ratios. Each one points at a single stage, so a weak number tells you where to look.
- Qualification rate = leads kept ÷ signals detected. Shows whether your sources and your ICP describe the same market.
- Agreement rate = leads you would have picked yourself ÷ leads you reviewed. The best early check on targeting, and only measurable when the reasons are shown.
- Verified contact rate = leads with a verified decision-maker email ÷ leads kept.
- Positive reply share = replies labeled positive or interested ÷ prospects emailed.
- Meeting conversion = meetings booked ÷ positive and interested replies. A low number here points at the offer or the reply handling, not the targeting.
| What you see | Stage to look at | What to change |
|---|---|---|
| Very few leads | Signals | Add a source or a signal type, or widen the ICP slightly |
| Many leads you disagree with | Qualification | Tighten the ICP using the exact reason you rejected them |
| Leads kept, but few contacts found | Contact | Check that the role you target exists at companies that size |
| Emails sent, almost no replies | Copy or offer | Make the ask smaller and tie the first line closer to the event |
| Replies, but mostly negative | Timing or offer | Check the signal still means what you think by the time you write |
| Interested replies, few meetings | Replies | Answer faster, and answer the question asked before proposing a call |
A worked example (invented numbers)
Two funnels, one month. Funnel A detects 300 signals, keeps 90 and finds a verified decision-maker for 81 of them. Funnel B detects 120, keeps 70 and finds a verified contact for only 21. Both look healthy on leads, but B converts 30% of its leads into contacts against 90% for A. The likely cause is the role, not the data: B targets a VP of engineering at companies of five people, where the founder is the buyer. Change the role and the ratio moves. None of these numbers is a benchmark; they show how to read your own.
Is outbound AI a fit for your team?
It fits B2B companies that sell to a definable buyer and close deals worth at least several hundred dollars: founder-led B2B startups, AI and automation agencies, IT service firms and consultants. The tighter your ideal customer profile, the better the AI qualifies, so it pays to fill in our ideal customer profile template before you launch.
It is not a fit for consumer outreach, for products so cheap that one email cannot pay for itself, or for a sales motion built on phone calls or LinkedIn steps: Startories runs email outreach and automates neither. For the strategy around the tooling, read our outbound sales strategy guide.
How do you start with outbound AI?
Pick one funnel where the signal is obvious, for example companies complaining about a competitor you replace, and run it in approval mode for the first weeks. Starter ($99 a month, $1 for a 3-day full-access trial on your first project) covers one funnel, one ICP and one intent source. Growth ($499) adds funnels, intent sources, campaigns and ongoing optimization by the team; Scale ($999) adds volume, sender accounts, advanced research and CRM integrations. Compare them on the pricing page.
If you want help getting the first funnels right, the done-for-you setup ($1,500 to $2,500, one time) covers ICP research, intent strategy, copywriting, sending domains and inboxes, CRM setup and launch. If you want the outcome without running the pipeline at all, the same system is available as done-for-you lead generation, from $1,999 a month. If you mostly want an AI that prospects and writes like a rep, start with the AI SDR page, or see why timing matters in signal-based outbound.
Frequently asked questions
What is outbound AI?
Outbound AI is software that runs the steps of outbound prospecting: finding companies with a reason to buy, checking their fit, finding and verifying the right contact, writing and sending emails, following up and sorting replies. A good system also shows why each lead was chosen.
Will outbound AI replace my sales team?
No. It replaces the research, list building and inbox work that eats most of a rep's week. People still choose the market and the offer, answer difficult replies and run the calls. Small teams use it so that a founder or account executive only spends time on real conversations.
Can I approve emails before they are sent?
Yes. Each funnel can require your approval for first emails and for replies. Once a funnel produces leads and emails you agree with, you can let it run on its own within its daily sending limits.
Does AI-written cold email hurt deliverability?
The writing is rarely the problem; volume, bad addresses and complaints are. Startories sends from separate, warmed-up domains, caps daily volume per inbox, contacts verified emails only, stops on reply and pauses when bounces rise.
Does Startories automate LinkedIn or phone calls?
No. Startories runs email outreach only and does not automate LinkedIn or phone. If those channels matter in your sales process, keep running them yourself or with a human SDR, alongside the email pipeline.
What do I need besides Startories?
An ideal customer profile, an offer worth replying to and someone to take the calls. Signal sources, company matching, contact search, email verification, sending inboxes and reply handling are included, so you do not need a separate lead database or cold email tool.