SmartReach AI · Privacy-Safe Data Dictionary

We do not share the person,
the company, or the message.
We share what the interaction teaches an AI agent.

SmartReach does not license prospect identities, company names, emails, LinkedIn URLs, raw messages, raw subject lines, or raw replies. The licensed dataset is made up of privacy-safe, AI-extracted GTM conversation features — CTA type, tone, message length, personalization depth, reply intent, objection type, next best action, unsubscribe risk, and outcome labels.

Download PDF

169 fields · 15 sections

Section · 8 fields

1. Record Metadata

Field	Derived From	Description	Example Values
`record_id`	—	Anonymous record ID	`SR_000001`
`dataset_version`	—	Dataset version	`2026_Q2_sample`
`source_channel_type`	—	Channel category, without identifying account details	`email, linkedin_message, chat_style_message`
`privacy_status`	—	Privacy treatment of row	`privacy_safe_derived_only`
`raw_text_shared`	—	Whether raw message or reply text is included	`no`
`identity_fields_shared`	—	Whether names, company names, emails, URLs, or LinkedIn links are included	`no`
`classification_method`	—	How labels were created	`AI_classified, rule_based, human_reviewed, hybrid`
`label_confidence`	—	Confidence in the AI label	`high, medium, low`

Section · 7 fields

2. Timing Features

No time zone is shared. No exact personal timestamp is shared unless approved. The safe version uses timing buckets.

Field	Derived From	Description	Example Values
`send_day_of_week`	date sent	Day message was sent	`Monday, Tuesday, Wednesday`
`send_month`	date sent	Month of send	`January, May, September`
`send_quarter`	date sent	Quarter of send	`Q1, Q2, Q3, Q4`
`send_period_bucket`	date sent	Broad send period	`early_month, mid_month, late_month`
`weekday_vs_weekend`	date sent	Whether sent on weekday or weekend	`weekday, weekend`
`reply_time_bucket`	time from sent to reply	Reply speed bucket	`same_day, 1_3_days, 4_7_days, 8_plus_days`
`reply_speed_score`	time from sent to reply	Speed of response	`fast, medium, slow`

Section · 7 fields

3. Buyer Profile Categories

No prospect name, LinkedIn profile, email, company name, or exact title is shared. Titles are converted to broad, non-identifying categories.

Field	Derived From	Description	Example Values
`seniority_band`	title	Broad seniority category	`Founder, C-Level, VP, Director, Manager, Individual Contributor`
`department_category`	title	Functional department	`Sales, Marketing, Operations, IT, Finance, HR, Product, Executive`
`buyer_role_category`	title	Role in buying process	`economic_buyer, technical_evaluator, influencer, operator, budget_holder`
`authority_level`	title category	Estimated authority level	`high, medium, low`
`persona_cluster`	title + department	Buyer persona group	`operations_leader, technical_buyer, growth_leader, executive_buyer`
`likely_buying_power`	seniority + department	Likely commercial authority	`budget_owner, recommender, researcher, end_user`
`human_handoff_priority_by_persona`	seniority + outcome	Whether this type of buyer should be routed faster	`high, medium, low`

Section · 9 fields

4. Company and Market Categories

No company name, website, LinkedIn company URL, or exact account details are shared. Fields marked * are available only on some records. Revenue is a proxy only unless enriched and verified from an approved external source.

Field	Derived From	Description	Example Values
`industry_category`	industry	Broad industry category	`manufacturing, healthcare, SaaS, retail, finance, logistics, education`
`sub_industry_category`	industry	More specific, still non-identifying category	`industrial_machinery, pharma, cybersecurity, consumer_goods`
`employee_count_band*`	employee count	Broad employee range where available	`1_10, 11_50, 51_200, 201_500, 501_1000, 1000_plus`
`employee_count_available_flag*`	employee count	Whether employee count existed in source	`yes, no`
`estimated_revenue_band_proxy*`	employee count + industry	Revenue proxy, not verified revenue	`$10M_$50M_proxy, $50M_$250M_proxy, $500M_$5B_proxy`
`revenue_estimate_method*`	employee count + industry	How revenue proxy was created	`employee_count_proxy, industry_proxy, not_verified`
`revenue_confidence*`	data completeness	Confidence in revenue proxy	`low, medium, high`
`commercial_value_bucket*`	employee band + persona	Estimated opportunity value	`low, medium, high`
`enterprise_fit_score*`	employee band + industry	Fit for enterprise motion	`low, medium, high`

Section · 3 fields

5. Geography Categories

If geography is shared, it is kept broad and non-identifying. No time zone is shared.

Field	Derived From	Description	Example Values
`country_category`	country	Country, if safe and approved	`United States, Canada, United Kingdom`
`region_category`	country	Broad region	`North America, Europe, APAC, Middle East, LATAM`
`geo_market_bucket`	country or region	Market category	`primary_market, secondary_market, emerging_market`

Section · 15 fields

6. Subject-Line Derived Features

No raw subject line is shared. Only extracted subject-line features are shared.

Field	Derived From	Description	Example Values
`subject_length_words`	subject line	Number of words in subject	`2, 4, 7`
`subject_length_chars`	subject line	Character count	`18, 42, 76`
`subject_length_bucket`	subject length	Subject length group	`very_short, short, medium, long`
`subject_personalization_present`	subject line	Whether subject contained personalization	`yes, no`
`subject_personalization_type`	subject line	Type of personalization	`first_name, company_reference, role_context, industry_context, none`
`subject_has_question`	subject line	Whether subject was phrased as a question	`yes, no`
`subject_tone`	subject line	Subject tone	`direct, curious, casual, formal, urgent, soft`
`subject_theme`	subject line	Subject-line theme	`quick_question, pain_point, value_prop, intro, follow_up, trigger_based`
`subject_specificity_score`	subject line	How specific the subject was	`1_5_score`
`subject_curiosity_score`	subject line	Curiosity level	`low, medium, high`
`subject_clarity_score`	subject line	Clarity level	`low, medium, high`
`subject_spam_risk_score`	subject line	Risk that subject appears spammy	`low, medium, high`
`subject_unsubscribe_risk_score`	subject line + outcome	Estimated risk that this subject pattern leads to unsubscribe or suppression	`low, medium, high`
`subject_pattern_id`	subject features	Reusable pattern label	`short_question_personalized, medium_value_prop, generic_follow_up`
`subject_outcome_label`	subject + outcome	What happened after this subject pattern	`no_reply, positive_reply, negative_reply, booked, unsubscribe`

Section · 22 fields

7. Outbound Message Derived Features

No raw outbound message is shared. Only extracted message features are shared.

Field	Derived From	Description	Example Values
`message_length_words`	message sent	Word count	`38, 74, 122`
`message_length_chars`	message sent	Character count	`220, 540, 890`
`message_length_bucket`	message length	Message length group	`very_short, short, medium, long`
`message_sentence_count`	message sent	Number of sentences	`2, 4, 6`
`message_paragraph_count`	message sent	Number of paragraphs	`1, 2, 3`
`message_structure_type`	message sent	Structure of message	`one_paragraph, multi_paragraph, bullet_style, question_led`
`message_readability_level`	message sent	Complexity level	`simple, moderate, complex`
`message_tone`	message sent	Tone of message	`direct, friendly, consultative, formal, casual, executive`
`message_formality_level`	message sent	Formality	`casual, neutral, formal`
`message_theme`	message sent	Main message theme	`cost_savings, growth, efficiency, visibility, risk_reduction, intro, demo_offer`
`message_hook_type`	message sent	Main hook type	`pain_point, curiosity, social_proof, trigger, value_prop, question`
`opening_line_type`	message sent	Type of opening	`personalized_intro, direct_value_prop, question, contextual_observation`
`value_proposition_type`	message sent	Type of value offered	`save_time, reduce_cost, increase_revenue, reduce_risk, improve_visibility`
`pain_point_present`	message sent	Whether a pain point was included	`yes, no`
`pain_point_category`	message sent	Type of pain point	`cost, growth, risk, efficiency, manual_work, visibility`
`benefit_count`	message sent	Number of benefits mentioned	`0, 1, 2, 3_plus`
`social_proof_present`	message sent	Whether message used proof	`yes, no`
`urgency_present`	message sent	Whether urgency was used	`yes, no`
`objection_prevention_present`	message sent	Whether message pre-handled objection	`yes, no`
`message_complexity_score`	message sent	Complexity score	`low, medium, high`
`message_quality_score`	message + outcome	Quality estimate	`1_5_score`
`message_pattern_id`	message features	Reusable pattern label	`short_soft_cta_pain_point, medium_consultative_value_prop`

Section · 9 fields

8. Personalization Derived Features

No actual name, company name, or personal detail is shared. Only the type and depth of personalization is shared.

Field	Derived From	Description	Example Values
`personalization_present`	subject/message	Whether personalization was used	`yes, no`
`personalization_type`	subject/message	Type of personalization	`first_name, company_reference, role_context, industry_context, trigger_based, none`
`personalization_depth`	message	Level of personalization	`generic, light, two_sentence, deep`
`personalized_sentence_count`	message	Number of personalized sentences	`0, 1, 2, 3_plus`
`two_sentence_personalization_flag`	message	Whether message had two or more personalized sentences	`yes, no`
`personalization_location`	subject/message	Where personalization appeared	`subject_only, opening_line, body, subject_and_body`
`personalization_specificity_score`	message	How specific the personalization was	`low, medium, high`
`personalization_risk_score`	message	Whether personalization may feel too invasive	`low, medium, high`
`personalization_outcome_label`	personalization + outcome	Outcome of that personalization pattern	`positive_reply, no_reply, unsubscribe, booked`

Section · 10 fields

9. CTA Derived Features

No raw CTA text is shared. Only CTA classification is shared.

Field	Derived From	Description	Example Values
`cta_present`	message	Whether a CTA was present	`yes, no`
`cta_type`	message	Type of CTA	`soft_cta, direct_meeting_ask, booking_link, reply_cta, question_cta, no_cta`
`cta_strength`	message	CTA intensity	`weak, medium, strong`
`cta_directness_level`	message	How direct the ask was	`soft, contextual, direct`
`cta_goal`	message	Intended outcome	`reply, schedule, book, click, buy, continue_conversation`
`cta_position`	message	Where CTA appeared	`opening, middle, closing, multiple`
`cta_question_based`	message	Whether CTA was phrased as a question	`yes, no`
`cta_friction_level`	message	How much effort the CTA requires	`low, medium, high`
`cta_outcome_label`	CTA + outcome	Outcome tied to CTA pattern	`positive_reply, booked, no_reply, unsubscribe`
`recommended_cta_for_agent`	outcome patterns	CTA an agent should use in similar case	`ask_question, offer_booking, send_link, continue_chat, handoff_to_human`

Section · 10 fields

10. Sequence-Step Features

No prospect identity or campaign identity is needed.

Field	Derived From	Description	Example Values
`sequence_step_number`	step	Step number in sequence	`1, 2, 3, 4`
`touch_type`	step	Type of touch	`initial_touch, follow_up, final_touch`
`follow_up_count_before_reply`	step + reply	Number of follow-ups before reply	`0, 1, 2, 3`
`reply_after_step`	step + reply	Which step generated reply	`step_1, step_2, step_3`
`positive_reply_by_step_flag`	step + outcome	Whether step produced positive reply	`yes, no`
`negative_reply_by_step_flag`	step + outcome	Whether step produced negative reply	`yes, no`
`unsubscribe_by_step_flag`	step + unsubscribe	Whether unsubscribe happened after this step	`yes, no`
`no_reply_by_step_flag`	step + no reply	Whether this step had no reply	`yes, no`
`sequence_fatigue_score`	step + outcome	Risk of too many touches	`low, medium, high`
`recommended_sequence_action`	step + outcome	What to do next in sequence	`continue, change_angle, wait, stop, human_review`

Section · 19 fields

11. Reply Derived Features

No raw reply text is shared. Only reply classifications are shared.

Field	Derived From	Description	Example Values
`reply_received_flag`	reply/no reply	Whether a reply was received	`yes, no`
`reply_length_words`	reply	Reply word count	`4, 18, 72`
`reply_length_bucket`	reply length	Reply length group	`very_short, short, medium, long`
`reply_sentiment`	reply	Sentiment classification	`positive, neutral, negative, unclear`
`reply_intent`	reply	Main intent of reply	`interested, not_interested, scheduling, question, objection, unsubscribe, referral`
`reply_category`	reply	Reply category	`pricing_question, timing_question, product_fit_question, competitor_question, not_relevant, send_info, book_meeting`
`reply_contains_question`	reply	Whether reply asked a question	`yes, no`
`reply_question_type`	reply	Type of question	`clarification, pricing, scheduling, product_fit, competitor, technical, availability`
`reply_objection_present`	reply	Whether reply contained objection	`yes, no`
`reply_objection_type`	reply	Type of objection	`price, timing, need, authority, trust, competitor, current_vendor`
`reply_urgency_level`	reply	Urgency level	`low, medium, high`
`reply_complexity_level`	reply	How complex response needs to be	`simple, moderate, expert_required`
`buying_intent_level`	reply + outcome	Commercial intent	`low, medium, high`
`reply_requires_answer`	reply	Whether response is required	`yes, no`
`reply_requires_clarification`	reply	Whether a clarifying question is needed	`yes, no`
`reply_requires_human_review`	reply	Whether human should review	`yes, no`
`recommended_reply_style`	reply	Best response style	`short, detailed, consultative, apologetic, direct`
`recommended_next_response_type`	reply	What the next response should do	`answer, clarify, schedule, suppress, route_to_sales, handoff_to_human`
`reply_classification_confidence`	reply classification	Confidence in reply classification	`high, medium, low`

Section · 14 fields

12. Outcome Fields

These are safe because they describe interaction outcomes, not identities.

Field	Derived From	Description	Example Values
`no_reply_flag`	no reply	Whether no reply occurred	`yes, no`
`unsubscribe_flag`	unsubscribe	Whether unsubscribe occurred	`yes, no`
`positive_reply_flag`	positive reply	Whether reply was positive	`yes, no`
`negative_reply_flag`	negative reply	Whether reply was negative	`yes, no`
`active_conversation_flag`	in talks	Whether conversation became active	`yes, no`
`scheduling_flag`	scheduling	Whether scheduling was reached	`yes, no`
`booked_flag`	booked	Whether meeting was booked	`yes, no`
`conversation_state`	outcome fields	Current/final state	`no_reply, active, scheduling, booked, negative, suppressed`
`outcome_label`	outcome fields	Final label	`no_reply, positive_reply, negative_reply, unsubscribe, scheduling, booked`
`conversion_stage`	outcome	Funnel stage reached	`sent, replied, active_conversation, scheduling, booked`
`sales_ready_flag`	reply + outcome	Whether conversation is sales-ready	`yes, no`
`qualified_interest_flag`	reply + outcome	Whether interest appears qualified	`yes, no`
`agent_success_label`	outcome	Whether interaction achieved goal	`success, partial_success, failure, suppression`
`outcome_confidence`	outcome	Confidence in outcome label	`high, medium, low`

Section · 9 fields

13. Suppression and Compliance Features

No unsubscribe message text is shared. Only the suppression signal and guardrail label are shared.

Field	Derived From	Description	Example Values
`suppression_required`	unsubscribe/negative outcome	Whether contact should be suppressed	`yes, no`
`agent_stop_signal`	unsubscribe/negative outcome	Whether AI should stop outreach	`yes, no`
`safe_follow_up_allowed`	outcome	Whether follow-up is allowed	`yes, no`
`negative_outcome_type`	unsubscribe/negative reply	Type of negative outcome	`unsubscribe, not_interested, complaint, wrong_persona`
`unsubscribe_stage`	step + unsubscribe	Step where unsubscribe happened	`step_1, step_2, step_3`
`unsubscribe_trigger_likely_reason`	message features + step	Likely reason for unsubscribe	`too_direct, wrong_persona, too_many_touches, poor_fit, generic_message`
`message_risk_pattern`	message features + negative outcome	Risk pattern in message	`too_pushy, too_long, weak_relevance, direct_CTA_too_early`
`sequence_risk_pattern`	sequence + negative outcome	Risk pattern in sequence	`fatigue_after_step_3, negative_after_direct_CTA, unsubscribe_after_follow_up`
`guardrail_training_label`	outcome	AI guardrail label	`continue, slow_down, stop, human_review, suppress`

Section · 12 fields

14. Next-Best-Action Fields

Among the most valuable fields for licensing.

Field	Derived From	Description	Example Values
`next_best_action`	reply + outcome + step	Recommended next action	`follow_up, answer_question, clarify, send_booking_link, route_to_sales, suppress, wait, human_review`
`next_best_action_reason`	row features	Why action is recommended	`positive_reply_with_scheduling_intent, unsubscribe_requires_suppression`
`allowed_next_actions`	outcome + compliance	Safe actions agent can take	`answer_question; route_to_sales, suppress; stop_sequence`
`disallowed_next_actions`	outcome + compliance	Actions the agent should not take	`continue_outreach, send_follow_up, push_for_meeting`
`agent_should_continue`	outcome	Whether agent should continue	`yes, no`
`agent_should_stop`	unsubscribe/negative	Whether agent should stop	`yes, no`
`agent_should_ask_question`	reply + stage	Whether agent should ask a question	`yes, no`
`agent_should_answer_directly`	reply question	Whether agent should answer directly	`yes, no`
`agent_should_offer_booking`	positive/scheduling	Whether booking CTA should be offered	`yes, no`
`agent_should_handoff_to_human`	reply + intent + complexity	Whether human should take over	`yes, no`
`human_handoff_reason`	reply + outcome	Why human handoff is needed	`complex_question, high_intent, scheduling, unclear_reply, negative_or_compliance`
`routing_destination`	next action	Where the conversation should go	`sales, support, expert, compliance, nurture`

Section · 15 fields

15. Chat Agent Training Fields

Especially valuable for product-specific LLMs, chat agents, revenue AI workflows, and human handoff logic.

Field	Derived From	Description	Example Values
`agent_use_case`	message/reply/outcome	Type of agent this row trains	`sales_chat_agent, customer_engagement_agent, expert_marketplace_agent, support_agent`
`conversation_stage_chat`	reply + outcome	Stage of chat conversation	`discovery, comparison, objection, closing, support, handoff`
`customer_intent`	reply/outcome	What customer appears to want	`browsing, interested, comparing, ready_to_buy, needs_help, not_interested`
`question_type_chat`	reply/message	Chat-style question type	`product_fit, pricing, availability, comparison, technical, scheduling`
`objection_type_chat`	reply	Objection category	`price, timing, trust, features, competitor, authority`
`best_response_style`	reply + outcome	Best style for agent response	`short, detailed, consultative, direct, empathetic`
`response_complexity_needed`	reply/question	How advanced response should be	`simple, moderate, expert_required`
`clarifying_question_needed`	reply	Whether agent should ask clarification	`yes, no`
`recommended_clarifying_question_type`	reply	Type of clarifying question	`use_case, budget, timeline, technical_requirement, quantity`
`buyer_readiness`	reply/outcome	Readiness stage	`browsing, evaluating, ready, not_ready`
`purchase_intent_score`	reply + outcome	Purchase/sales intent score	`1_100_score`
`recommended_cta_for_chat`	state + intent	Best CTA for chat agent	`ask_follow_up, send_link, book_call, buy_now, human_handoff`
`chat_handoff_trigger`	reply/outcome	Trigger for human handoff	`complex_question, purchase_ready, scheduling, complaint, unclear_intent`
`marketplace_agent_training_category`	use case	Marketplace category	`business_expert, product_seller, advisor, support_expert`
`synthetic_sales_brain_label`	full row	What this row trains	`intent_detection, objection_handling, CTA_selection, handoff_logic, suppression_logic`

Section 16

Fields We Do Not Share

Excluded from the licensed sample unless separately approved under a strict agreement.

Field	Share?	Reason
Prospect name	No	private identity
Prospect email	No	private contact information
Prospect LinkedIn URL	No	private/prospect identity
Company name	No	account identity
Company website	No	account identity
Raw subject line	No	may contain private/company/person details
Raw message sent	No	may contain private/company/person details
Raw reply text	No	private communication
Sender email	No	sender identity
Sender LinkedIn URL	No	sender identity
Time zone	No	not part of licensed dataset
Exact campaign/account identifiers	No	can reveal customer/account identity

We do not share the person,the company, or the message.We share what the interaction teaches an AI agent.

1. Record Metadata

2. Timing Features

3. Buyer Profile Categories

4. Company and Market Categories

5. Geography Categories

6. Subject-Line Derived Features

7. Outbound Message Derived Features

8. Personalization Derived Features

9. CTA Derived Features

10. Sequence-Step Features

11. Reply Derived Features

12. Outcome Fields

13. Suppression and Compliance Features

14. Next-Best-Action Fields

15. Chat Agent Training Fields

Fields We Do Not Share

We do not share the person,
the company, or the message.
We share what the interaction teaches an AI agent.