Chapter 28 Key Takeaways: Customer-Facing Work: Sales, Support, and Outreach

Core Principles

The authenticity imperative is non-negotiable. Customer-facing communication that feels automated, generic, or inauthentic actively damages relationships. In a world of increasing AI-generated outreach, genuine human attention and specific knowledge stand out.
"AI drafts, humans send" is the right operating model for most customer-facing communication. AI generates speed and scale; human review provides quality, relationship sensitivity, and accountability. The review step is not overhead — it's the core quality control mechanism.
The personal review step is not optional. Before sending any AI-assisted customer communication, a human must read it and take personal accountability for the content. The question is not "is this accurate?" but "would I be comfortable if this customer knew exactly how this was created?"
Not all customers warrant the same personalization investment. Tiered personalization — deep manual research for top priorities, AI research with verification for mid-tier, category-level personalization for the rest — is a rational allocation of limited attention.

Research before writing is the difference between genuine personalization and template insertion. "Personalized" outreach that inserts only the company name is recognizable to experienced recipients. Specific research — recent news, business model specifics, relevant context — makes personalization genuine.
Verify at least one fact per AI research summary before using it in outreach. AI generates plausible-sounding company-specific information that is occasionally wrong. A reference to a product launch, funding round, or executive hire that didn't happen destroys credibility immediately.
The "1,000 people who need different messages" problem requires a tiered approach. Not every prospect warrants individual research. Define which 20% of your list deserve full personalization and design efficient category-level approaches for the rest.
Cold email sequences should feel like a progression, not four versions of the same message. Each email should make a different case, use a different angle, and reflect that time has passed since the last contact. The break-up email serves a real function: it closes the cycle cleanly and sometimes triggers responses from previously unresponsive prospects.
The LinkedIn note must fit in 200 characters and have a genuine reason for connection. The small format demands genuine specificity — there is no room for generic praise or vague shared interests.

Meeting prep briefings should anticipate objections, not just summarize context. The most valuable preparation output is: what will this person push back on, and what should I be ready to say?
Discovery questions should have follow-up probes. The first question opens a topic; the follow-up probe goes deeper into what matters. AI can generate both, but the follow-ups require judgment about which initial answers warrant deeper exploration.
Honest competitive analysis — including where competitors are genuinely stronger — is more useful than defensive comparison. Understanding where you lose helps you identify which prospects are not good fits, which objections to prepare for, and which topics to avoid.

Audit your ticket population before building AI workflows. Knowing what percentage of your tickets are routine vs. complex determines whether AI assistance will provide significant value and where to invest in knowledge base quality.
The knowledge base quality determines AI response quality. AI responses based on well-written, accurate knowledge base articles are much better than AI responses generated from scratch. The upfront investment in knowledge base quality pays for itself quickly.
Explicit escalation criteria, defined in advance, are the most important design element of any AI support workflow. The criteria should be written down and reviewed regularly. Common escalation triggers: strong emotion, legal references, repeat contact on the same issue, enterprise accounts, upcoming renewals.
Human review before sending is what allows AI drafts to be sent without quality loss. Without review, quality degrades. With review, quality can actually improve because the review process catches gaps that AI missed.
The human-in-the-loop imperative for complex cases is not just ethical — it's measurably effective. Research shows that customer satisfaction scores for human-reviewed AI responses match fully human responses, while fully automated responses score lower on complex and emotional issues.

Account review preparation should lead with what the customer needs from the meeting, not what you want to present. The agenda should be organized around their decisions and their value realization — not your quarterly metrics.
Renewal conversations should open with specific value delivered, not with the contract renewal date. The most effective renewal framing starts from a place of demonstrated partnership value, then moves to what the next phase looks like.
Upsell identification is about timing and trigger events, not just product fit. AI can identify accounts with product-fit for expansion. Knowing when to have the conversation — what trigger events make it natural — requires relationship knowledge AI doesn't have.

Integrated CRM AI tools (Salesforce Einstein, HubSpot AI) produce better insights than standalone tools because they work with real customer data. The quality advantage comes from context, not writing capability.
The automation vs. augmentation choice reflects a philosophy about risk tolerance. Automated tools (Intercom Fin) are appropriate for high-volume, low-complexity scenarios with well-designed escalation. Augmentation tools (Zendesk AI) are appropriate when human judgment is more important than response time.

Disclosure is clearly required when customers are having real-time interactions without knowing they're talking to AI. This is a basic honesty standard, not just a compliance consideration.
Disclosure of AI assistance in reviewed, human-sent communications is a judgment call that depends on relationship type, industry norms, organizational policy, and the specific nature of the communication.
The credibility damage from customers discovering undisclosed AI use is real and asymmetric. The damage to trust from a single discovered undisclosed AI interaction typically outweighs the efficiency gained from the AI assistance. Err on the side of transparency in established relationships.

Response rate research supports genuine specificity as the primary driver. AI outreach that achieves genuine specificity through the research-then-write workflow matches or exceeds manually personalized outreach.
Human-reviewed AI support responses maintain customer satisfaction. The review step is what preserves quality — not just the quality of the AI output itself.
Customer ability to detect AI-generated communication is improving. As AI patterns become more familiar, the authenticity bar rises. This argues for more thorough editing, not less.
Partially personalized AI communication may be worse than clearly generic. The "uncanny valley" effect applies: almost-but-not-quite personal triggers lower trust than either clearly automated or clearly human. Thorough editing matters.