Testing H! header

Outbound prospecting with AI seems to be everywhere and nowhere all at once — not because teams aren’t trying, but because the workflow is fundamentally stacked against them.

Vendors popping up daily selling new, often high cost SaaS solutions. Meanwhile, Reps are bouncing across LinkedIn, company websites, internal notes, Salesforce records, and “enablement resources” to craft a single message.

• Leaders want personalization at scale.

• Finance wants predictable cost.

• Engineering wants to avoid yet another SaaS integration.

• Everybody wants pipeline.

So we built something that solves all of it — inside Salesforce, driven by cheap OpenAI API, with zero reliance on expensive vendor contracts.

This post explains what we built, why it matters, and how any organization can replicate the approach with the right architecture and a few dollars in API usage.

High-Level Summary

We created a Salesforce-native AI engine that generates:

• personalized outreach strategies

• tailored messaging

• persona-aware positioning

• company-size-aware CTAs

• industry-informed value props

…all from the fields already sitting in your CRM.

Cost per full outreach plan? Pennies.
Time-to-first-usable-message? Seconds.
Vendor lock-in? Zero.

This is how outbound should feel in 2025.

What Leaders Care About (TL;DR for Busy Executives)

For Sales Leaders

TL;DR: Reps get high-quality personalized outreach without slowing down.
No more generic templates or 20-minute research cycles before hitting send.

For CFOs

TL;DR: You get enterprise-grade personalization for the cost of spare change.
A few pennies in OpenAI usage per lead, no new SaaS contracts, no seat licenses.

For CTOs / VPs of Engineering

TL;DR: Clean architecture, no shadow IT, no brittle Zapier chains.
All logic in Salesforce, governed by your existing security and audit model.

For CEOs

TL;DR: You accelerate revenue motion without adding headcount or technical debt.
Outbound becomes more professional, more consistent, and more aligned with your brand.

Why We Didn’t Buy a Tool for This

You can easily spend six figures annually on “AI-powered sales engagement” platforms.

Most are:

• rigid

• opinionated

• hard to customize

• dependent on vendor uptime

• priced per seat or per sequence

• slow to adapt to your tone and ICP nuances

We wanted something different — something that acts like a seasoned operator writing bespoke messaging for each prospect:

• grounded in business context

• role-aware

• industry-aware

• respectful of company size

• aligned to our brand voice

• maintainable by us

• practically free to run

So we built it ourselves.

And the truth? Any company can — with the right blueprint.

What the System Actually Does

1. Pulls in dozens of Salesforce fields:
   industry, revenue, employee count, business model, department, management level, website, recent activity, etc.

2. Analyzes the lead’s role, domain, and company size to infer likely pains.

3. Generates a tailored outreach plan with:

   • summary

   • strategy

   • a multi-step outreach sequence

   • subject lines

   • short, clean message bodies (no walls of text)

   • reasoning for each touch

4. Stores everything back on the Lead record.
   No browser tabs. No manual formatting.

5. Costs literally pennies per plan.

6. Runs entirely within your Salesforce governance model.

This isn’t “AI email writer” nonsense.
It’s role-driven, ICP-aware, contextually intelligent outbound strategy generation, done with the data you already own.

How We Tuned the Messaging to Fit Our Brand

We shaped the engine to reflect how we communicate:

• straight to the point, respecting folks time

• grounded in operational reality

• no hype language or buzzwords

• clear narrative based on our value proposition

• CTAs that flex based on company size (enterprise ≠ SMB)

This is crucial:
AI is only powerful when the persona behind it is sharp.
And any company can define its own persona — operator, technical leader, sales strategist, industry expert — and embed it directly into the generation engine.

We simply encoded ours.

Why This Is Valuable for Different Personas

Sales Reps

• Stop guessing what to say.

• Stop copying templates.

• Start sounding like someone prospects want to talk to.

Sales Enablement

• No more maintaining long playbooks.

• No content drift.

• Everything stays inside CRM.

Revenue Leaders

• Better messaging → better meetings → more pipeline.

• Consistency without micromanagement.

Marketing Leaders

• ICP alignment at the message level.

• Clear feedback loop on tone, value props, and patterns.

IT & Security

• No external data sprawl.

• No new vendors to review.

• Secure keys with Named Credentials.

Like technical details?

Some readers come for the business strategy.
Others come for the implementation details.

If writing Apex, building LWCs, handling malformed JSON, or wrestling with OpenAI’s response formatting sounds like a weekend you’d rather skip — feel free to stop here and click to schedule a discovery call with me so we can talk about outcomes instead of technical details.

If you're the kind of CTO who alphabetizes your AWS accounts or has opinions about schema registry patterns…
welcome to the fun part.

Technical Deep Dive (For Engineers, Architects & Admins)

Architecture Overview

We built a Salesforce-native system consisting of:

• Apex service layer

  • builds a structured context block

  • enforces persona rules

  • makes OpenAI API call through Named Credential

  • normalizes malformed JSON

  • persists results on the Lead

• Lightning Web Component (LWC)

  • surfaces summary, strategy, reasoning

  • formats messages into readable blocks

  • adds copy buttons for subject/body

  • auto-loads saved plans

• OpenAI gpt-5-mini

  • cheap & fast

  • supports response_format: json_object, which cuts down cleanup work

Contextual Intelligence

We feed in Salesforce fields such as:

• Industry

• Title / Department / Job Function

• Revenue / Employee count

• Ownership type (public/private)

• Website

• Recent tasks

Blank or unknown fields are ignored, preventing hallucinations.

Persona & ICP Logic

The prompt architecture handles:

• role interpretation

• industry-aware adjustments

• company-size segmentation

• CTA strength modulation

• tone consistency and brand positioning

JSON Reliability Layer

We implemented a custom normalizer to handle:

• missing brackets

• extra commas

• code fences

• truncated arrays

• unexpected type returns

This makes the system production-safe even with varying model responses.

LWC UX Details

The component renders:

• summary

• high-level strategy

• omni-channel steps

• copy-to-clipboard buttons

• reasoning in muted formatting

It feels like a native Salesforce feature — because it is one.

Why This Matters for Forward-Looking Organizations

AI isn’t a replacement for authenticity.
But it is an accelerant for clarity, consistency, and context.

The real innovation here isn’t “AI writes emails.”
It’s:

• using your own CRM data as the personalization engine

• running the entire workflow inside Salesforce

• giving reps operator-level messaging without operator-level headcount

• empowering technical teams to expand, tune, or integrate however they want

• doing it all at near-zero cost considering typical CAC/CPA spend

This is modern outbound:
data-driven, permission-aware, contextually intelligent, and infinitely scalable.

Want This in Your Org — Without Writing 800 Lines of Apex?

What we built for ourselves is repeatable — tailored to any industry, product, go-to-market motion, or leadership voice.

If you want:

• personalized messaging that isn’t generic

• outbound that respects ICP nuance

• AI integrated safely inside Salesforce

• extremely low operating cost

• no dependence on external outreach vendors

• an architecture your engineering/security teams actually like

Proksel Consulting can build the entire system for you — or help your technical team do it themselves.

Curious what this could look like inside your org?