Issue No. 09 - AI in BDR

Most BDR Teams Don't Need More AI. They Need Better Internal Workflows.

The tools exist. The problem is nobody has translated the actual BDR motion into the workflows that use them. Here's how to do that - for reps and the managers trying to coach and inspect them.

7 min read June 2026

Here is the real problem with AI in BDR right now, and it has nothing to do with the tools.

Most teams have plenty of tools. They have enrichment platforms, signal aggregators, AI email writers, conversation intelligence, dialers. Most of those tools were bought to solve a problem someone diagnosed with a vendor's framework rather than their own. They do not talk to each other particularly well. And none of them were built for how this specific team works, what this specific ICP looks like, or what actually needs to happen differently on Monday morning.

The result is not an AI problem. It is an operating problem. Reps are moving fast with partial visibility. Managers are stretched across too many people to coach or inspect consistently. Pipeline data is uneven. Nobody quite knows which activity is real and which is the sophisticated appearance of productivity. And the AI tools being purchased are solving the vendor's version of those problems - not the actual version.

The version that works is internal. Built around the actual motion. Designed by someone who understands the BDR operating system deeply enough to translate it into a workflow, a prompt, an adoption plan, and a measurement loop. That does not require a product team. It requires knowing your team and your buyers well enough to build something specific to both.

"The prompt is the product.
The AI is the engine.
Someone still has to know what good looks like."

Start With the Manager Problem

Most AI BDR content leads with rep productivity because that is the visible problem - reps spending 40 minutes on research, reps sending generic emails, reps not personalizing. Those are real problems worth solving. But the manager problem is where the operating leverage actually is, and it is the one that gets closest to what hiring leaders are actually worried about.

The manager problem is this: I cannot see enough. I cannot coach consistently. I do not know which pipeline is real until it is too late to do anything about it. My Friday review covers three of fourteen reps if nothing else goes wrong. My coaching feedback is vague because I don't have time to listen to every call. And the activity metrics in my CRM tell me whether my reps are busy, not whether they are effective.

That is the problem AI internal workflows are uniquely good at solving - not because they replace manager judgment, but because they make the right work visible at scale.

01
Manager Tool - Call Quality Inspector
Every call reviewed against your criteria - not a vendor's generic rubric

Pull call transcripts from your dialer into a workflow that runs them against a prompt you wrote - one that reflects your actual ICP, your qualification criteria, your coaching standards, and your team's specific failure patterns. The AI flags the specific moments: the buying signal the rep heard and moved past, the closed-ended question that killed the conversation, the point where they stopped listening and started pitching. You get a scored summary of every call in minutes instead of hours. You review the flags, not the full recordings. You coach to the moment, not the outcome. The difference between "you need to create more urgency" - which means nothing and changes nothing - and "at the four-minute mark on Tuesday's call, she told you her team had been managing this manually for eighteen months and you moved to the next question" - which is specific, citable, and coachable.

Workflow architecture
Dialer transcript n8n/Make trigger ICP + coaching prompt scored call summary manager Slack flag manager reviews coaching delivered outcome logged prompt iterated
100% call coverage vs 20% Coaching time - 75% Specific feedback vs vague notes Consistent quality standard across team
Manager reviews every flag before rep receives feedback. No automated coaching delivered without human sign-off. Transcript data stays within approved internal systems. No personal prospect data in the prompt.
02
Manager Tool - Pipeline Quality Checker
Know which deals are real before the forecast call - not during it

Before every pipeline review, run your open opportunities through a prompt that checks for the signals that historically indicate whether a deal is real or exists because nobody has been willing to call it dead yet. Is there a next step booked with a specific date and a prospect-confirmed action? When did the buyer last do something - not the rep, the buyer? Has the close date moved, and who moved it? Is the documented pain specific or is it "interested in learning more"? Is there a champion or just an initial contact who took a call? The AI reads the CRM data that is already there and surfaces a pre-meeting summary in two minutes. You walk into the pipeline review knowing which deals need a conversation and which need to be walked back to an honest stage. The reps are not lying to you. They are often not being fully honest with themselves either. The data in the CRM is more accurate than the story being told about it. This tool reads the data.

Workflow architecture
CRM open opps quality criteria prompt deal health summary manager pre-meeting brief pipeline review stage adjustments logged accuracy tracked over time
Pipeline review time - 50% Forecast accuracy improved Earlier risk identification Cleaner stage progression
Works on CRM data the team already owns. No new data collection required. Pre-meeting step only - not used for rep performance evaluation without manager context.

Then Build for the Reps

03
Rep Tool - 5-Minute Account Brief
Research that used to take 40 minutes, delivered before the rep opens a tab

Your reps are reinventing the same research process from scratch every day for every account with wildly inconsistent results. Some of that research requires human judgment and cannot be automated - the observation the rep makes, the thing they notice that no prompt would surface. But most of it is findable. Company context, recent news, tech stack signals, relevant triggers, CRM history. Build a workflow that pulls from public sources and your CRM, runs it through a prompt built on your ICP criteria, and delivers a five-bullet account brief before the rep opens the email or picks up the phone. Not a paragraph of AI copy. Five bullets: trigger, pain hypothesis, relevant context, suggested angle, one open question to lead with. The rep reads it in 90 seconds, edits what needs editing, and shows up with the context that was previously taking 40 minutes to build - if they built it at all.

Workflow architecture
CRM account record public enrichment layer ICP + persona prompt 5-bullet brief CRM task / Slack notification rep reviews + edits brief informs outreach outcome logged prompt improved
Research time - 80% Higher-quality first touches Better account prioritization Consistent ICP application
Rep reviews and edits every brief. No auto-send. Brief informs outreach - it does not generate it. Public data sources only. No personal prospect data processed.
04
Rep Tool - Signal-to-Draft Pipeline
AI watches for buying signals and writes the hypothesis. The rep decides whether it's worth sending.

Set up a workflow that monitors target accounts for the signals that matter to your motion - funding rounds, leadership changes, job postings for the role you sell into, competitor news, earnings commentary. When a signal fires, the AI drafts a hypothesis-based opener: one paragraph referencing the specific trigger and connecting it to the problem you solve for that persona. The rep gets a Slack notification with the draft and the account link. They review it, adjust for tone and accuracy, add the human layer the AI could not have known, and decide whether to send. What this replaces is the rep who would have missed the trigger entirely because they were working their list in order and had not gotten there yet. The AI never misses it. The rep still owns what happens next. This is categorically different from Clay doing the same thing - because this runs on your signal criteria, your messaging framework, your tone, and your ICP definition. Not a vendor's assumptions about all of those.

Workflow architecture
Signal source (public) account match in CRM persona + ICP prompt hypothesis draft rep Slack alert rep reviews + edits rep sends or discards outcome logged
Zero missed triggers Faster time-to-first-touch Higher relevance on first contact Less dead air between research and outreach
Human approval before anything sends. Public company-level signals only - no personal prospect activity. GDPR compliant when restricted to company-level triggers. No auto-send under any circumstances.
05
Rep Tool - AI Ramp Coach
New reps practice against an AI built on your actual buyer - before they ever call a real one

The fastest path to a ramped rep is repetition against realistic resistance. Most managers do not have time for daily role plays, and the ones that happen are often too comfortable because nobody wants to be the person who makes the new hire spiral in week two. Build a practice environment where new reps run cold calls and discovery conversations against an AI persona built on your specific ICP. Not a generic VP of Sales. Your VP of Sales archetype - their objections, their patience level for a weak opener, their specific language around the problems you solve. The AI gives structured feedback after each attempt based on your qualification criteria. The manager reviews flagged sessions rather than sitting in on every call. The new hire gets repetitions in before they ever call a real prospect. Ramp time drops because the reps have already made the obvious mistakes somewhere the cost was zero. The prospects who were previously receiving the "I'm still learning" experience are no longer doing that. Everyone wins.

Workflow architecture
Manager defines ICP persona + qualification criteria Lovable / custom build rep practice session AI feedback against rubric session logged manager reviews flags 1:1 coaching informed
Ramp time - 2-3 weeks Consistent qualification standard Manager coaching time focused on edge cases New hires arrive at real calls prepared
Built entirely on internal ICP knowledge and manager-defined archetypes. No real prospect data used. Practice sessions stay in approved internal systems only.

What Makes These Actually Work

The tool is not the hard part. n8n, Make, Zapier, Lovable - these are accessible, well-documented, and manageable without a product team. The hard part is the layer underneath: knowing the motion well enough to design something that fits into it without requiring the rep or manager to change how they work just to use the new thing.

Here is the operator checklist. Every workflow needs all of these before it goes live.

Before anything goes live - the operator standard
01
Tied to a specific failure point in the motion. Not "AI could make this better." Which specific, measurable thing is currently broken - and how does this workflow close that gap? No answer to this question means no build.
02
Input data is clean enough to trust. AI amplifies the quality of what you put in. Garbage CRM data produces garbage briefs. Before building on top of your CRM, audit what is actually in it. Incomplete records produce incomplete outputs that erode rep trust in the tool inside two weeks.
03
The prompt reflects your actual ICP, qualification criteria, and coaching standards. A generic prompt produces generic output. Write it like you are onboarding a new team member - specific, opinionated, with examples of what good looks like and what bad looks like at your company.
04
The output fits into the rep or manager's existing workflow. If using the tool requires a new tab, a new login, or a new habit, it will not be used. Deliver outputs where the work already happens - Slack, CRM tasks, email. Friction kills adoption faster than bad output does.
05
The human review step is explicit and non-negotiable. AI drafts. AI flags. AI summarizes. Humans decide, edit, send, and coach. This is not a preference - it is the line between a useful internal workflow and a compliance problem waiting to happen.
06
The behavior change is measured. Not "did we use the tool." Did research time drop. Did meeting quality scores improve. Did ramp time decrease. Did pipeline forecast accuracy get better. Tie the tool to a number before you build it. That is the only way to iterate it and the only way to justify the time.

The Governance Layer That Actually Matters

This is the section that separates "someone who got excited after a Lovable weekend" from someone a VP can trust with their team's data, their reps' development, and their company's compliance posture. Here is what internal AI governance actually looks like in practice.

Governance standard - non-negotiable

Approved inputs only. No personal prospect data in prompts. Company-level signals, job titles, CRM deal data, and call transcripts within your own system are fine. Personal contact information, scraped data from non-consented sources, or data from outside your approved stack is not.

Approved outputs only. AI output goes to a human reviewer before it reaches a rep or a prospect. Define in writing what the AI is allowed to produce, what requires manager sign-off, and what never gets automated. Document it.

Prompt versioning with an audit trail. Every prompt is dated, named, and logged with what changed and why. A prompt that worked in Q1 needs review when your ICP shifts, your product adds a use case, or your legal team updates the messaging guidelines. Treat prompts like policy documents.

No auto-send. Ever. Under any circumstances. For any reason. This is the one that protects you from the scenario where something goes wrong at 11pm on a Tuesday and you find out about it Wednesday morning via a very uncomfortable email from a prospect's legal team.

Data stays in approved systems. If your company uses Salesforce, outputs live in Salesforce. If your security team has not approved a tool, the workflow does not use it. This is not optional - it is the condition under which every other part of this works.

What a Good Prompt Actually Looks Like

Because "write a good prompt" is not an instruction - it is a homework assignment with no rubric. Here is the difference between a prompt that produces usable output and one that produces AI noise.

Example - Call Quality Inspector prompt (abbreviated)
Review this call transcript for three things only: 1. Whether the rep identified a business pain in the prospect's own words - not paraphrased, not assumed. Exact quote or flag as not found. 2. Whether the rep connected that pain to a specific business outcome - revenue, time, headcount, or risk. If they connected it to a product feature instead, flag it. 3. Whether they secured a next step with a specific date confirmed by the prospect. "I'll follow up next week" does not count. Flag only moments with direct transcript evidence. Do not give generic coaching advice. Do not summarize the call. Output three findings with the timestamp and exact quote for each. If something is not present in the transcript, say "not found" - do not infer.

Notice what it does not say: "give feedback," "summarize the call," "assess overall performance." Those are instructions that produce paragraphs of AI text nobody reads. Specific criteria. Specific evidence requirement. Specific output format. The narrower the prompt, the more useful the output - and the more it sounds like something a manager would actually say in a coaching session.

The Part That Requires You to Actually Know Your Team

Daniela built the call quality inspector first. Not because it was the most exciting - it was not. Because Friday afternoons were costing her four hours and producing feedback that was too vague to change anyone's behavior.

The prompt took longer than the workflow because she kept making it better. She kept adding her team's specific patterns - the qualification question they were asking too early, the persona objection they were mishandling, the moment they consistently lost the prospect's attention. That process of building the prompt was the coaching work. It forced her to articulate exactly what good looked like, in language specific enough that an AI could recognize it.

That is not something a vendor can do for you. A vendor builds for the average team. You need to build for yours.

The workflows in this post are not revolutionary. They are practical. They are buildable this week by someone who understands BDR motion, knows what their team's failure patterns are, and is willing to write a very specific prompt about both.

That is the work. Everything else is just the engine.

This Week's Play

Pick the one problem that is costing your team the most pipeline or the most manager time right now. Write the prompt that would catch it - not a generic one, a specific one. Run it manually on five real examples this week.

If it catches something useful, you have a workflow worth building. If the prompt is not specific enough yet, you now know exactly what to fix. Either way you are doing the work that most teams are waiting for a vendor to do for them.

Outbound, But Smarter
The Modern BDR Operator