The Voice-Profile Playbook

A Sales Connector field manual for clients and white-label partners.

SECTION 01Foreword

Most outbound fails for one boring reason. It sounds like outbound. The prospect reads three words, the brain goes "this is a sales pitch from a stranger," and the message is gone. No reply. No callback. No second chance.

The Sales Connector system is built around a simple fix. Every message that goes out under your name should sound like you wrote it. Not your industry. Not your title. You. Specifically. The cadence you actually use, the words you actually pick, the small verbal habits that prove a human typed this.

That fix lives in a file we call your voice profile. It is the most important asset we maintain on your behalf, and it is the input you have the most leverage over. This playbook explains what it is, how we build it, what we need from you to make it sharp, and how to keep it sharp as your business changes.

It is not a marketing document. There are no aspirational sentences about authentic communication. This is a manual. Read it once during onboarding, then refer back to it every quarter when you refresh your inputs.

Act 1. Why Voice Match Decides Reply Rates

SECTION 021.1 The "every salesperson sounds the same" problem

Open your LinkedIn inbox right now and scan the cold messages from the last seven days. Without reading them in detail, just glance.

You will notice a few things. They all start the same way. "Hi {name}, hope you're doing well." Or "Quick question for you." Or "I noticed you're the {title} at {company}." They all close the same way. "Worth a quick chat?" or "Open to a 15-minute call?" The middle of the message is interchangeable. Swap the sender names and you cannot tell who wrote which.

That is the failure mode. Not bad spelling. Not bad targeting. Voice flatness. The messages have been laundered through so many sales training programs and SaaS templates that they have lost any signal that a specific person sat down and wrote them.

The prospect's brain is running a single binary classifier. Real human writing me, or sales template aimed at me. Once the classifier flips to template, the message is dead. It does not matter how good your offer is. It does not matter how relevant your timing is. The reader is not reading anymore.

This is what we are fighting. And we are fighting it at scale, because outbound only works at scale, which means the temptation to use generic copy is structural. Every vendor in this space leans on generic copy because generic copy is easy to produce and easy to scale. We do the opposite. We make voice match the foundation of the entire system.

SECTION 031.2 What "voice" actually is, technically

When we say voice, we do not mean tone in the loose sense. We mean a measurable bundle of writing habits. Specifically:

• Sentence length distribution. Some people write in short, declarative bursts. Others write longer compound sentences. Some mix both with a clear rhythm. The distribution is a fingerprint.
• Vocabulary preferences. Which words you actually pick when several are available. Do you say "help" or "support"? "Buy" or "pick up"? "Customer" or "client"? "Team" or "folks"? Tiny choices, repeated, become unmistakable.
• Punctuation habits. Whether you use semicolons. Whether you use exclamation points and how often. Whether you use parenthetical asides. Whether you use ellipses. Whether your questions tend to be direct or hedged.
• Opener and closer patterns. How you start a message to a stranger versus a warm contact. How you sign off. Whether you greet by first name, by full name, or skip the greeting entirely.
• Hedging level. How often you soften claims with "I think" or "might" or "could be." Some operators are blunt. Others are diplomatic. Both work, but only if the message matches the person.
• Idioms and small verbal tics. The phrases that recur in your real writing that nobody else would use the same way. "Honest take," "quick gut check," "stupid question but," "the boring answer is." These are the highest-signal tokens we look for.
• Topic-shape preferences. Whether you tend to lead with the ask, lead with context, or lead with a hook. Whether you tend to give numbers or stories or both.

A voice profile is a structured representation of all of this, stored in a way the AI can use when it drafts a message. It is not "a paragraph about how I like to sound." It is a working specification.

SECTION 041.3 The two failure modes we are designing against

There are really only two ways outbound fails, once targeting is correct.

Failure mode one. The message sounds like a vendor. Tone is generic, structure is template-shaped, vocabulary is industry-flat. Reply rate sits below one percent regardless of volume. Most outbound vendors live here. Their pitch to you involves words like "personalization tokens," which is jargon for putting the prospect's first name into a generic message.

Failure mode two. The message sounds like a stranger trying to sound like you. This is what happens when an AI is given some samples but not enough, or the wrong samples, and produces output that is uncanny. The cadence is almost right but the word choices are wrong. A close contact who actually knows you would notice immediately. A stranger might not consciously notice but their classifier still pings "off."

The job of your voice profile is to keep us out of both modes. Mode one is solved by giving the AI any real samples at all. Mode two is solved by giving us enough of the right samples, structured so the model can pull from them under different conditions, and refreshing them as your voice evolves.

SECTION 051.4 The reply-rate math, briefly

We track reply rate by voice-profile completeness. The data is internal but the shape is consistent.

A campaign run with no voice profile, just a default tone shaped by your industry, lands somewhere between 0.6 and 1.4 percent reply rate. This is the floor. It is what most generic outbound platforms produce.

The same campaign, same audience, same offer, run with a complete voice profile lands between 3.8 and 7.2 percent reply rate. We have seen specific clients sustain above ten percent for narrow audiences they know well.

That is roughly a four to seven times improvement, holding everything else constant. The only thing that changed is whether the message sounds like the sender or sounds like a sales platform. Voice match is not a polish item. It is a multiplier on every other thing you spend money on.

SECTION 061.5 What you control, what we control

You control the inputs. We control the structure, the model behavior, the testing, and the iteration.

This division matters because the most common reason a voice profile underperforms is not that we built it wrong. It is that the inputs were thin or unrepresentative. If you send us three short emails and a forwarded press release, we cannot conjure your voice from that. We will produce something that drifts toward generic, because that is the safest fallback when signal is missing.

The good news is that the inputs are bounded. We do not need a hundred samples. We need the right five to ten samples, and a fifteen-minute voice memo. We will explain exactly how to assemble them in Act 2.

Do this now

1. Open your LinkedIn inbox and read your last ten received cold messages. Notice how many you could swap senders on without anyone catching it.
2. Open your sent folder, find five emails to people who are not your team or your family, and read them out loud. Listen for the patterns that are unmistakably yours.
3. Block thirty minutes on your calendar this week titled "voice profile inputs." We will use that block in Act 3.

Act 2. How the System Works

SECTION 072.1 The five-element input checklist

A complete voice profile is built from five categories of input. Each category contributes a different signal. Missing categories mean missing signal. We can compensate for one weak category if the others are strong, but missing two or more starts to push your profile toward generic territory.

The five elements are:

1. A structured voice memo. Ten to fifteen minutes, recorded by you, following our prompt list.
2. Five past emails. Specifically chosen for emotional and stylistic range, not random.
3. Three to five social posts. LinkedIn posts, tweets, or other public writing under your name.
4. One sales call recording or transcript. A real call where you were selling, qualifying, or closing.
5. A brand guide if you have one. Optional but useful when present.

We will go through each one in detail. The goal is for you to leave this section knowing exactly what to send and why.

SECTION 082.2 Element one: the structured voice memo

This is the highest-signal input. A fifteen-minute recording of you talking, when you know we are going to transcribe it and train on it, captures more of your voice than fifty written emails.

Why? Because writing is filtered. You think about each sentence. You edit. Speech under low pressure is the closest we get to your unfiltered cadence, vocabulary, and sentence structure. The model can pick up rhythms from speech that simply do not appear in even your best emails.

But the voice memo only works if it is structured. A free-form ramble produces a transcript that is hard to use, because the topics are scattered and the energy is uneven. We give you a prompt list and ask you to talk through it.

The prompt list

Record yourself answering these in order, without notes. Talk for one to two minutes per prompt. If you stumble, keep going. Do not edit. Send the raw audio.

1. Tell me about your business. What do you sell, who buys it, and why does it matter?
2. Describe your ideal customer. Not the demographic. The specific human. What do they care about? What are they tired of?
3. Walk me through your last great sales conversation. What did you say first? What did they say back? Where did it click?
4. What is something most people in your industry get wrong, and what is the truth?
5. Tell me a story about a time you helped a customer solve a real problem. Names anonymized, no script.
6. If a stranger landed on your LinkedIn profile right now, what should they immediately understand about you?
7. What is a sales tactic you refuse to use, and why?
8. Tell me about a deal you lost, and what you learned from it.
9. What do you do for fun outside of work, and how does it influence how you operate?
10. If we wrote a message in your voice that landed in a prospect's inbox, what would make you say "yes, that sounds like me"? Be specific.

That is fifteen to twenty minutes of speech. The transcript will be roughly 2,500 to 3,500 words, which is enough material to seed every meaningful pattern in your writing.

How to record one that actually helps

Notes from reviewing hundreds of these.

• Use your phone's voice recorder or your laptop's built-in recorder. Do not over-engineer the audio. We just need clear voice.
• Sit somewhere quiet enough that we can transcribe cleanly. A car in a parking lot works. A coffee shop usually does not.
• Talk to a real person mentally. Imagine a specific friend or colleague asking you these questions. Do not imagine an audience or a microphone.
• Do not script. The point is to capture how you actually speak. Reading prepared answers defeats the exercise.
• Stumbles are fine. "Hmm, let me think about that." That is how humans talk and the model should learn to write that way too, when appropriate.
• Energy matters. Record when you have had coffee and slept, not at the end of a long day. You can hear the difference, and so will the model.

If you want to go above and beyond, record twice, once in the morning and once after a sales call. The contrast between cold and warm energy is itself useful signal.

SECTION 092.3 Element two: the five past emails

Five emails sounds small. It is. The reason it is small is that the goal is not volume, it is range.

We want five emails that, taken together, span the emotional and tactical range you actually use. If you send us five short scheduling emails, we have learned almost nothing. If you send us five emails of the same type to similar contacts, we have learned one mode of you.

The filter we recommend is what we call the "I'm proud I sent this" filter. Open your sent folder. Scroll back two months. Pick five emails that, at the moment you sent them, you felt good about. Not great, not perfect, just good. Emails where you felt the writing did the job and the writing felt like you.

Inside that filter, try to span this range:

• One cold or near-cold outreach. A first message to someone you did not yet know.
• One follow-up after silence. Where you re-engaged a lapsed conversation.
• One difficult message. Pushing back on a price negotiation, declining a request, addressing a concern. Conflict tone is rare in samples but high signal.
• One warm relationship message. To a long-time client or partner. Friendly, casual, low-stakes.
• One closing or commitment-asking message. Where you asked for a meeting, a signature, a yes.

If you cannot find one in each category, send what you have. Three good emails with range beat seven mediocre emails of the same type.

What "good inputs" look like

A good email sample is at least five sentences long. It contains some specifics, not just pleasantries. It demonstrates your voice handling something that has stakes, even small ones.

A weak input is a one-line "thanks, will do." That is not a sample, that is acknowledgment. Strip those. We are looking for emails where you actually communicated.

What we strip before training

Before any sample touches the model, we strip:

• Recipient names and personal identifiers
• Company names if you ask us to
• Any sensitive numbers, contracts, or NDA-bound content
• Email signatures (we will rebuild yours separately)
• Forwarded chains. We only want what you wrote, not what was below your reply.

If you have a privacy concern about a specific email, just leave it out. We can build a great profile without your most sensitive correspondence.

SECTION 102.4 Element three: the three to five social posts

Social posts capture a different mode of your voice than emails. Emails are private and one-to-one. Social posts are public and one-to-many. The cadence differs. Sentence breaks are usually shorter. Hooks are sharper. There is more deliberate punctuation.

This matters because some of the messages we send for you will mimic the energy of public posting more than the energy of a private email. A great LinkedIn comment, for example, is closer to a social post than a cold email in feel. We want both registers represented in your profile.

Pick three to five posts that you wrote yourself. Not reposts. Not articles you commissioned. Posts where you sat down and wrote the words.

If you do not post on social media, you can skip this category. We will weight other inputs higher and produce a slightly less varied profile. The system still works, you just lose one register.

If you post regularly, send your three to five most recent strong posts. Recency matters because public voice tends to evolve faster than private voice as people experiment with what works on the platform.

SECTION 112.5 Element four: the sales call recording

A sales call recording is the most direct view of you in actual sales mode. Email and social are written modes. A sales call is improvised and reactive. The phrases you use under live pressure, when a prospect is pushing back or asking a hard question, are extremely useful for training reply behavior.

The reason replies need separate training data from cold messages is that reply quality is what determines whether a conversation continues. A great cold message that gets a reply, followed by a generic reply from your end, is worse than a mediocre cold message that you would have ignored anyway. The drop-off is brutal.

A thirty-minute sales call gives us:

• How you actually answer hard questions
• How you reframe objections
• How you ask for the next step
• The phrases you use to build rapport in real time
• The level of casualness or formality you default to under pressure

You can send the audio file or a transcript. Most modern call platforms produce decent transcripts automatically. If you do not have a recording, send a Loom or a fresh role-play recording where you talk through your usual sales conversation as if a prospect were on the other end.

If you absolutely cannot share a sales call, the structured voice memo from element one becomes the primary source for conversational voice. The profile will still work, just with slightly less calibration on reply behavior.

SECTION 122.6 Element five: the brand guide

Optional. If your company has a brand guide that includes voice and tone guidelines, send it. We will read it and align where we can without overriding what your real samples show.

A common case is a founder whose personal voice is sharper and more direct than the formal brand voice their marketing team has documented. We will default to the personal voice for outbound, because outbound is one human writing to another, and the formal brand voice is wrong for that channel. We will let you decide if there are specific words or claims that must or must not appear regardless of voice.

If you do not have a brand guide, that is fine. Most clients do not. We will infer constraints from what we see in the samples.

SECTION 132.7 What happens after you send the inputs

Here is what your account manager does once your inputs are received.

Day one. All raw materials are uploaded to a private workspace tied to your account. The voice memo is transcribed. The emails and posts are de-identified and de-noised. The sales call is transcribed if it was sent as audio.

Day two. Your account manager builds a first draft of the voice profile. This is a structured file, several thousand words long, that includes:

• A short prose description of your voice for human reference
• A list of vocabulary preferences with examples ("uses 'team' rather than 'staff,' uses 'genuinely' rather than 'truly'")
• A list of opener patterns and closer patterns drawn from your real samples
• A list of phrases that are unmistakably yours, flagged as high-signal
• A list of phrases or patterns to avoid because they sound generic or are not consistent with your samples
• Tone calibration on six axes: directness, formality, warmth, hedging, humor, technicality
• Reference samples for the model to draw from at message-generation time
• Any constraints from the brand guide

Day three. Your account manager runs a calibration test. We generate ten test messages across different scenarios using the new profile and compare them, side by side, against samples from your real writing. We score the matches and identify any gaps.

Day four. Your account manager sends you the draft profile and three sample messages generated from it. You review. You flag anything that does not sound like you. You can mark specific phrases as "never use" or "use more often."

Day five. Profile is finalized and goes live. From this point on, every message generated for your account is built using this profile.

The whole process takes four to seven business days from receipt of complete inputs. The most common reason it takes longer is that inputs were incomplete and we had to ask for more.

SECTION 142.8 How the AI actually uses your profile

This is the section where we explain what is happening under the hood. You do not strictly need to know this to operate the system, but understanding it helps you give us better inputs.

When the system needs to generate a message for you, here is the sequence:

1. Context gathering. The system pulls the prospect's profile, recent activity, any prior interactions, and the campaign goal.
2. Profile loading. Your voice profile is loaded into the model's context window. Not just the prose summary. The actual reference samples, the vocabulary preferences, the opener and closer patterns.
3. Draft generation. The model produces a candidate message under a prompt that requires it to match your voice profile and address the campaign goal.
4. Voice match scoring. The candidate is scored against your profile on the six tone axes plus a phrase-level similarity check against your reference samples. Anything below threshold is regenerated, not sent.
5. Topical and factual checks. Separate checks for accuracy, claim safety, and compliance with any brand-guide constraints.
6. Final pass. A short pass to clean up obvious template-isms and replace any phrases that show up in our generic-language blocklist.

The voice profile is referenced in steps two, three, and four. It is not a one-time setup that biases an output. It is an active filter on every message.

This is why the profile quality compounds. A weak profile leaks into every step. A sharp profile keeps the output inside your real voice across thousands of messages without any per-message babysitting from you.

SECTION 152.9 How replies are tested against your voice

Replies are different from cold messages in two ways.

First, replies need to maintain conversation continuity. The prospect just said something. The reply needs to acknowledge it accurately and move the conversation forward, all in your voice.

Second, replies are higher stakes per message. A bad cold message gets ignored. A bad reply ends a live conversation that you were already winning. So the bar is higher.

Our reply system runs your profile against an additional layer trained on your sales call recording. The conversational voice from the call is what calibrates the rhythm of replies. Email samples calibrate the structure. Voice memo calibrates the underlying personality.

Each reply gets the same scoring as a cold message, plus a conversation-coherence check that confirms the reply is actually responsive to what the prospect said. That second check is platform-wide and not voice-specific, but it interacts with the voice profile because some operators reply with questions and others reply with statements, and the system needs to match your style.

This is also where the absence of a sales call recording hurts most. Without it, the reply system relies entirely on email samples for conversational rhythm. Email rhythm is slower than spoken rhythm. Replies trained only on emails tend to feel slightly stiff. Not bad, just a little formal. With even one decent sales call recording in the inputs, this disappears.

SECTION 162.10 How the briefing file evolves over time

Your profile is not static. We update it on three triggers.

Trigger one. Reply data. Every reply gives us calibration data. We see which messages got replies and which got ignored, which phrases drew engagement and which got ghosted. Every quarter we run an analysis pass and update the high-signal phrases list and the opener and closer patterns based on what is actually working.

Trigger two. Your business changes. If you launch a new product, reposition, or shift markets, your voice drifts in ways the old profile cannot anticipate. We watch for this and flag drift to your account manager. You can also flag it directly.

Trigger three. Quarterly refresh. Even with no specific changes, voices evolve. New phrases enter your vocabulary, old ones drop out. Once a quarter your account manager will ask for a small refresh. Details in Act 3.

A profile that started as a rough sketch in week one becomes a precise, lived-in artifact by month six. Reply rates climb across this curve, not because the model is getting smarter, but because the profile is getting more specific.

SECTION 172.11 What if you do not share enough

The failure modes we see when inputs are thin.

No voice memo. Profile defaults to email-only voice. Messages are structurally fine but slightly stiff. Reply rate sits in the two to three percent range instead of four to seven. Recoverable as soon as you send a memo.

Only short emails as samples. Nothing to learn vocabulary range from. Profile hits the right tone at low complexity but falls apart on anything needing more than three sentences. Recoverable with one fresh batch of substantive emails.

No sales call. Conversational voice underdeveloped. Replies trend formal. Cold messages still strong. Recoverable with one recorded sales call or substitute role-play.

Only formal corporate writing as samples. Worst case, because the inputs are misleading rather than just thin. The profile learns the version of you that writes for a board deck. Cold messages in that voice fail. Recoverable but requires fresh, casual samples to override the formal training.

Inputs on Monday, then nothing. Profile goes stale within six months. Reply rate slowly drops. Recoverable with a quarterly refresh.

In every case, the fix is "send better inputs." We do not need a lot. We need representative.

SECTION 182.12 A direct comparison: generic versus voice-matched

Here is the same outbound goal, the same target audience, the same offer, written two ways. The first is the kind of message that comes out of a generic outbound platform. The second is the same message rewritten through a real voice profile, anonymized for privacy. The client is a fractional CFO who works with early-stage SaaS companies.

Generic version

Hi {first name},

>

Hope you're doing well. I noticed you're the CEO at {company} and wanted to reach out.

>

I help SaaS companies optimize their financial operations and scale efficiently. Many of our clients have seen significant improvements in their cash flow management and burn rate optimization.

>

Would you be open to a 15-minute call next week to explore if there's a fit?

>

Best,

{sender name}

This message is fine. It is also dead on arrival. Every word in it could have been written by any one of two thousand fractional CFOs. The prospect's classifier flips to template inside the first six words.

Voice-matched version

Quick one for you. I poked around your stack page and the seed round announcement from March, and you guys are running pretty lean for a Series A run rate.

>

The boring answer to "when does a SaaS founder need a real CFO" is usually around the time the cash flow forecast starts being something you check on Sunday nights. If that's already happening I'd skip this email and call your accountant.

>

If it isn't yet, I do fractional CFO work for about a dozen seed and Series A SaaS teams. Most of them brought me in three months before they thought they needed to. Worth a 20-minute call?

>

Honest, no pressure either way.

>

{sender name}

Read those two side by side and notice what changed.

The voice-matched version uses "Quick one for you" instead of "Hope you're doing well." That phrase appeared three times in this client's voice memo and twice in his sample emails. It is unmistakably his.

It uses "the boring answer" as a setup. That phrase is in his social posts. It is not a phrase a generic AI would default to.

It does the unusual move of telling the prospect to call their accountant if a condition is met. That is from his sales call, where he said almost the same line. It signals he is not desperate for the meeting, which is itself the most attractive thing a fractional CFO can signal.

It closes with "Honest, no pressure either way." That is his exact email closer in three of the five samples he sent.

Reply rates. The generic version, sent at scale to similar prospects, ran 0.9 percent reply rate. The voice-matched version on the same audience ran 6.4 percent. Same offer. Same product. Same audience. The only thing that changed is the voice profile actually matched the sender.

Do this now

1. Block fifteen minutes today and record the structured voice memo using the prompts in section 2.2. Send the file to your account manager when done.
2. Open your sent folder and pick five emails using the "I'm proud I sent this" filter from section 2.3. Forward them to your account manager with the subject line "Voice profile samples."
3. Find one sales call recording from the last two months and send it. If you have nothing, schedule a thirty-minute role-play with a colleague this week and record it.

Act 3. Apply

SECTION 193.1 The voice profile worksheet

This worksheet is the single artifact you fill out and send to your account manager. When this is complete and submitted, your voice profile build begins.

Treat this as an actual document. Open a new note, a new doc, whatever you use. Fill out each section. Do not skip sections. If a section does not apply, write "n/a" with a one-sentence reason. Skipping silently produces a worse profile.

Section A. Identity basics

• Your full name as you want to appear on outbound:
• Title you currently use on LinkedIn:
• Company you represent:
• Your role in one sentence (not a job description, just what you actually do):
• One sentence describing your ideal customer:

Section B. Voice memo

• Voice memo recorded? (yes/no)
• If yes, file name and location (we will give you an upload link, this is for your tracking):
• If no, when will it be done? (give a real date, not "soon")

Section C. Five email samples

For each of the five, paste the email body with names removed. Label each one with which category it covers from section 2.3.

• Email 1. Category:
• Email 2. Category:
• Email 3. Category:
• Email 4. Category:
• Email 5. Category:

Section D. Social posts

• Post 1 (paste body):
• Post 2:
• Post 3:
• Post 4 (optional):
• Post 5 (optional):

If you do not post publicly, write "n/a, do not post" and we will adjust.

Section E. Sales call

• Recording or transcript provided? (yes/no)
• If yes, format and location:
• If no, will you provide one? When?
• If no and never, do you have an alternate (Loom, role-play recording)?

Section F. Brand guide

• Brand guide provided? (yes/no)
• If yes, location:
• If no, are there words or claims you require or forbid in outbound? List below.

Section G. Voice characteristics, your read

This section is your subjective read on yourself. The model will form its own read from the samples, but yours is useful for cross-checking. Rate each from 1 to 10 and add a one-sentence note.

• Directness (1 = heavy hedging, 10 = blunt). Score and note:
• Formality (1 = very casual, 10 = corporate formal). Score and note:
• Warmth (1 = transactional, 10 = warm and personal). Score and note:
• Humor (1 = no humor in business, 10 = jokes routinely). Score and note:
• Technicality (1 = avoid jargon, 10 = lean into industry terms). Score and note:
• Length default (1 = always short, 10 = comfortable with long messages). Score and note:

Section H. Phrases that are unmistakably you

List five phrases or sentence patterns you use often that, when you hear them in your own writing, feel like a fingerprint. They can be small. Examples to seed your thinking, not to copy: "to be honest," "the simple version is," "quick gut check," "I might be wrong, but," "stupid question."

• Phrase 1:
• Phrase 2:
• Phrase 3:
• Phrase 4:
• Phrase 5:

Section I. Phrases that are not you

List five phrases or sentence patterns that you would never write but that appear in generic outbound regularly. We will block these. Examples: "circle back," "touch base," "exciting opportunity," "leverage synergies."

• Block phrase 1:
• Block phrase 2:
• Block phrase 3:
• Block phrase 4:
• Block phrase 5:

Section J. Final notes

Anything else we should know. Specific topics that are off-limits. Specific people whose names should never appear. Specific competitors whose names should be handled with care. Specific cultural sensitivities. Anything.

Once this worksheet is complete and the supporting files are uploaded, your account manager moves to day one of the build process described in section 2.7.

SECTION 203.2 What happens in the first month after launch

The profile goes live. We start sending messages. Here is what your week-by-week experience should look like.

Week one. First batch of messages goes out. You receive copies of every message we send, in real time, in a feed your account manager shares with you. Read at least the first twenty. Flag anything that does not sound like you. Your account manager applies the flags as targeted profile patches.

Week two. Reply data starts coming in. Reply rates are usually a little below your eventual steady state because the profile has not yet been calibrated against actual prospect responses. This is normal.

Week three. First reply-batch analysis. Your account manager pulls the messages that got replies versus those that did not, and identifies any voice patterns associated with each. Profile is patched.

Week four. Reply rate stabilizes. You should be seeing consistent message quality. The "does this sound like me" question should be answerable with a confident yes when you read a random sample of ten messages.

If at week four the answer is anything other than yes, escalate to your account manager. Do not let it ride. The fix is almost always more inputs, not more iteration on the same inputs.

SECTION 213.3 The quarterly refresh routine

Once a quarter, your account manager will reach out to schedule a refresh. This takes thirty to sixty minutes of your time and dramatically extends the useful life of your profile.

The refresh has four parts.

Part one. New voice memo, five minutes

We do not need a full fifteen-minute memo each quarter. Five minutes is enough. The prompt list is shorter:

1. What has changed in your business since the last refresh?
2. What is something you are saying differently now than you were three months ago?
3. Tell me about your best sales conversation this quarter.

That is it. Three minutes minimum, six minutes maximum. Send the file.

Part two. Three new email samples

Pick three emails from the past quarter that you are proud of. Use the same filter as in onboarding. We are looking for evolution since the last batch. The new samples replace the oldest samples in the profile.

Part three. Reply-data review

Your account manager pulls the reply data for the quarter and walks you through it for ten minutes. Which campaigns landed. Which did not. Which phrases drew the most engagement. Which segments responded best. You give input on what you want to lean into next quarter.

Part four. Profile update

Based on parts one through three, your account manager produces an updated profile. You review three sample messages from the new profile. Sign off, or flag, and the new version goes live.

The total time investment is under an hour per quarter. The reply-rate impact is measurable. Clients who do quarterly refreshes hold steady or grow over time. Clients who do not refresh slowly drift back toward generic over six to twelve months as the profile becomes stale relative to their evolving voice.

SECTION 223.4 Edge cases and exceptions

You are running outbound for multiple personas

Each persona gets its own voice profile. The cost is in the inputs, not the system. Each persona needs the full five-element checklist filled out by that specific person.

We do not recommend running outbound under a name that is not a real person on your team. Aside from compliance issues, the messages sound off because the voice has no underlying source.

You are a white-label partner running outbound for many end-clients

Same logic. Each end-client gets their own voice profile. You can streamline by walking them through this playbook, sending them the worksheet, and structuring the onboarding call as a voice-memo capture. Some white-label partners record onboarding calls and use the recording itself as the voice memo input. Works well.

For end-clients who will not engage with the worksheet, you have two choices. Run them on a generic profile and accept the lower reply rate, or do not take them on. We recommend the second. A generic-profile client undermines your white-label brand because the reply rate fails to demonstrate the system working.

Your voice has changed dramatically and the old profile is wrong

A common case is a founder moving from a casual scrappy phase into a more polished growth phase, or vice versa. Reply data drifts. Messages that used to sound right start to feel wrong.

Do not wait for the next quarterly refresh. Tell your account manager. We run a full re-onboarding. The old profile is archived, the new one is built fresh. Four to seven days, same as initial onboarding.

You want the system slightly more polished than you actually are

Rare but real. Some clients want the AI to write a slightly more curated version of their voice rather than a perfect mirror. Legitimate request. We can dial the voice-match strength on the six tone axes, leaning a half step toward the brand-guide voice. We do not recommend dialing this very far. If the dial moves too much, you are back in generic territory.

SECTION 233.5 Three real anonymized cases

Three short examples of voice profiles working in production. Names and details changed.

Case A. The hardware founder

Founder at a hardware startup. His voice is technical, dry, and self-deprecating. His voice memo had three running jokes about engineers being bad at sales. His emails were short and punchy.

The profile leaned on three signature moves. It opened with a specific technical observation about the prospect's product, not a pleasantry. It used self-deprecating asides. It closed with a non-pushy ask, often a question. Reply rate stabilized at 5.8 percent on cold outbound to engineering leaders at mid-stage hardware companies.

The technical specificity in the opener flagged "real hardware person, not a typical SDR." The self-deprecation flagged "comfortable, not desperate." Both signals neutralized the cold-outbound classifier in the prospect's brain.

Case B. The fractional executive

Fractional COO who works with services businesses. Her voice is warm and organized. She uses lists naturally even in emails. Her voice memo had clear numbered structure even when answering open-ended questions.

Her profile encoded the list habit. About 30 percent of her cold messages contain a numbered list of two or three items. This is unusual for cold outbound. Her list structure became a voice fingerprint prospects could not pattern-match to template.

Reply rate stabilized at 7.1 percent on a tough audience. The list structure made the messages feel less salesy and more like a peer reaching out.

Case C. The agency owner

Agency owner. Informal, fast-paced, lots of contractions. Voice memo was full of one-word sentences.

The profile preserved the one-word-sentence habit. About one in four cold messages from his account opens with a one-word sentence. Sometimes "Look." Sometimes "Real talk." Sometimes a name as a one-word call-out.

Reply rate stabilized at 6.4 percent. Multiple replies said some version of "this didn't sound like the usual AI message." That comment from prospects is one of the strongest signals we track.

What unites all three. None of these voices were invented. Each started in the client's voice memo and emails and was preserved through the build. The job was not creativity. The job was faithful capture.

SECTION 243.6 Final note

You will get out of this what you put in. The system is good. The model is good. Your account manager is good. None of those compensate for thin inputs.

A complete five-element checklist, submitted in the first week, refreshed quarterly, will produce outbound that sounds like you for as long as you run it. A half-finished checklist will produce outbound that sounds like a slightly improved generic vendor, which is still better than nothing but not why you signed up.

The single highest-leverage action you can take in your first week as a Sales Connector client is to spend ninety minutes on inputs. Fifteen on the voice memo, twenty on selecting and forwarding emails, ten on social posts, thirty on a sales call recording, fifteen on the worksheet. Ninety minutes total, once, for a profile that will run for years.

That is the deal. We build the rest.

Do this now

1. Open the worksheet in section 3.1, copy it into a doc, and start filling it out today. Do not wait for the perfect time block.
2. Schedule your first quarterly refresh on your calendar for ninety days from your launch date. Title the event "Voice profile refresh" with a link to this playbook.
3. Forward this playbook to anyone on your team whose voice will also run through Sales Connector outbound. Each of them needs their own profile.

This playbook is maintained by the Sales Connector content team. If your experience differs from what is described here, tell your account manager. The playbook gets updated when the system changes or when client feedback identifies a gap.