Branching from assessment of [[Case Studies/Pylon/Assumptions & Problems Addressed]], a valuable tool would be to establish a formula for how the "Business" Agent assess support costs - linear scaling with ticket volume (ext. of customer base). This is a thought exercise, designed to abstract the levers of support costs, not a literal accounting model; frame the operational challenges for myself.
**The "Outdated" Formula:**
$C_{\text{total}} = (V_{\text{total}} \times C_{\text{agent}}) + (V_{\text{esc}} \times C_{\text{eng}}) + C_{\text{tools}}$
**The "Pylon" Redefined Formula:**
$C_{\text{new}} = (V_{\text{total}} \times (1 - R_{\text{ai}})) \times (C_{\text{agent}} \div E_{\text{agent}}) + (V_{\text{esc}} \times (1 - R_{\text{esc}})) \times (C_{\text{eng}} \div E_{\text{eng}}) + C_{\text{platform}}$
The magic is in the **four new variables** that Pylon introduces:
## Defining the variables for the "outdated" or current formula:
- $C_{\text{total}}$ = Total Cost of Support
The final number the "Business Agent" is trying to control.
- $V_{\text{total}}$ = Total Volume of Tickets
Core linear driver. As the customer base grows, $V_{\text{total}}$ grows, and every other variable tied to it explodes.
- $C_{\text{agent}}$ = Average Agent Cost Per Ticket
This is the fully-loaded cost (salary, training, overhead) for a support agent to handle one ticket. The only "fix" in this old model is to hire more agents as $V_{\text{total}}$ increases.
- $V_{\text{esc}}$ = Volume of Escalated Tickets
This is the percentage of $V_{\text{total}}$ that requires engineering. It also grows linearly with the total volume.
- $C_{\text{eng}}$ = Average Engineering Cost Per Escalation
This is the "Engineering Support Tax" quantified—the fully-loaded cost (an engineer's valuable time) to resolve one escalated ticket.
- $C_{\text{tools}}$ = Total Tooling Cost
The (often per-seat) licenses for the helpdesk, Jira, etc., which also scales up as the number of agents increase
## And expanding/evolving the formula with/for Pylon:
- **$R_{\text{ai}}$ (The AI Resolution Rate)**
- It directly attacks **$V_{\text{total}}$ (Total Volume)**; the key lift, represents Pylon's AI Agents and automated deflection.
- If AI resolution rate is 50%, you've _halved_ the number of tickets that even need to touch a human. The linear relationship ($V_{\text{total}}$) is broken. (Nice)
- **$E_{\text{agent}}$ (Agent Efficiency Gain)**
- This lever attacks **$C_{\text{agent}}$ (Agent Cost)**.
- This represents Pylon's AI Assistant (suggested replies, surfacing knowledge).1 If an agent can solve tickets 2x faster, you've effectively cut their per-ticket cost in half (or doubled their capacity), fighting burnout and scaling the _existing_ team.
- **$R_{\text{esc}}$ (Escalation Reduction Rate)**
- This lever attacks **$V_{\text{esc}}$ (Escalation Volume)**.
- By giving L1 Support agents a unified view, better context, and AI-surfaced knowledge, they can solve more complex problems without escalating (reduced "engineering tax").
- **$E_{\text{eng}}$ (Engineering Efficiency Gain)**
- This lever attacks **$C_{\text{eng}}$ (Engineering Cost)**.
- For the escalations that do get through, they are now "clean" (with full context, logs, and user history). This eliminates the back-and-forth, slashing the engineering time-per-escalation from (for example) 45 minutes down to 10.
**Takeaway:** Pylon's pitch isn't just "we're a cheaper $C_{\text{tools}}$." The platform fundamentally re-architects the entire cost structure of support by introducing exponential efficiency (deflection and assistance) to a traditionally linear problem.
## Mock Scenario: "Anto-Novation", a Mid-Sized B2B Company
Modeling / drawing it out to show impact; the numbers tell the story.
This is fun for me.
*(Disclaimer: This is a thought exercise and for my own enjoyment. The goal is to model the levers of change, not to be a literal accounting sheet. The numbers are for illustration.)*
**Baseline Assumptions (Monthly):**
First, establishing baselines under the "Outdated" model.
- **Total Tickets ($V_{\text{total}}$):** **2,000**
- Avg. Agent Cost Per Ticket ($C_{\text{agent}}$): $20
(Based on a fully-loaded agent salary, training, and overhead)
- Escalation Rate: 20%
(1 in 5 tickets needs an engineer)
- **Volume of Escalations ($V_{\text{esc}}$):** 20% of 2,000 = **400**
- Avg. Eng. Cost Per Escalation ($C_{\text{eng}}$): $100
(Based on 1-1.5 hours of a fully-loaded engineer's time for discovery, back-and-forth, and fixing)
- Tooling Cost ($C_{\text{tools}}$): $1,500
(Seats for Zendesk, Jira, etc.)
**Calculation:**
- **Support Agent Cost:** (2,000 tickets $\times$ $20) = **$40,000**
- **Engineering "Support Tax":** (400 escalations $\times$ $100) = **$40,000**
- **Tooling Cost:** = **$1,500**
**Total Monthly Cost: $81,500**
> **Key Insight:** In this common scenario, the "hidden" Engineering Support Tax ($40k) is _just as expensive_ as the _entire_ visible support agent payroll ($40k). This is the core problem an engineering leader feels.
### The "Pylon" Model: Redefined
Applying Pylon's levers, assumed, to the _exact same_ 2,000 tickets.
**Pylon Levers (Assumptions):**
- AI Resolution Rate ($R_{\text{ai}}$): 40%
(40% of total tickets are deflected by AI, never touching a human)
- Agent Efficiency Gain ($E_{\text{agent}}$): 2x
(AI-assist helps agents solve tickets twice as fast, halving their cost-per-ticket to $10)
- Escalation Reduction Rate ($R_{\text{esc}}$): 50%
(Better-equipped L1 agents solve 50% of the issues that would have been escalated)
- Eng. Efficiency Gain ($E_{\text{eng}}$): 2x
(Clean escalations cut engineering time in half, reducing their cost-per-escalation to $50)
- New Platform Cost ($C_{\text{platform}}$): $5,000
(A higher, all-in platform cost)
**Calculation:**
- **Human-Handled Tickets:** 2,000 $\times$ (1 - 0.40) = **1,200 tickets**
- **New Agent Cost Per Ticket:** $20 $\div$ 2 = **$10**
- **New Support Agent Cost:** (1,200 tickets $\times$ $10) = **$12,000**
- **Original Escalations:** 400
- **New Escalation Volume:** 400 $\times$ (1 - 0.50) = **200 escalations**
- **New Eng. Cost Per Escalation:** $100 $\div$ 2 = **$50**
- **New Engineering "Support Tax":** (200 escalations $\times$ $50) = **$10,000**
- **New Platform Cost:** = **$5,000**
**Total New Monthly Cost: $27,000**
### Bottom-Line Comparison
What story do the numbers tell?
|**Metric**|**"Outdated" Model**|**"Pylon" Model**|**% Change**|
|---|---|---|---|
|**Total Support Cost**|**$81,500**|**$27,000**|**-67%**|
|Engineering Support Tax|$40,000|$10,000|**-75%**|
|Support Agent Cost|$40,000|$12,000|**-70%**|
|Tickets Handled by Humans|2,000|1,200|**-40%**|
|Tickets Escalated to Eng.|400|200|**-50%**|
Nice, business isn't just saving money here, I can propose it's reclaiming 75% of its engineering capacity that was being lost to the "support tax."