"When the student is ready, the teacher appears. When the servant is empowered, the customer is served."— Adapted from Vedantic teaching on Seva (selfless service)
The Challenge
In late 2019, Himanshu Niranjani joined Visible — Verizon's all-digital, direct-to-consumer telco startup — as CPTO. Visible was everything a OneX portfolio company looks like in high definition: product-market fit confirmed, revenue flowing, customer base growing fast. But beneath the growth curve, a structural crisis was building quietly.
The customer-service team had projected its future headcount on a straight line: 300 chat agents today, 800+ as the user base expanded. The business had accepted this as the cost of scaling a carrier. Himanshu did not. He saw what the rest of the organization had missed: this was not a headcount problem. It was an engineering problem that had been outsourced to operations.
By March 2021 — when he stood before Visible's CEO to present his findings — the numbers were staggering. Engineering throughput had increased 10×. Customer-service costs had been cut 73%. NPS had crossed 50, a number that does not exist in the telecommunications industry. Engineer happiness had moved from 41% to 93%. The company had secured a patent. None of this required GenAI. None of it required a platform rewrite. It required the oldest principle in the S+3 playbook: give engineers ownership of outcomes, not just output.
The Portfolio Analog
Every customer-facing software product eventually reaches the inflection point Visible hit: growth strains the support model. The naive response is to hire more agents. The S+3 response is to ask why those tickets exist in the first place — and then assign that question, permanently, to the engineering team that created the conditions for them. This is the Serve pillar of S+3 made operational. It collapses the wall between engineering and customer service, aligns incentives, and turns the support queue from a cost center into a product-intelligence system.
The Baseline: The Scaling Paradox
| Metric | Pre-S+3 baseline | Driver |
|---|---|---|
| Chat support agents | 300 (projected to 800+) | Linear headcount model vs. user growth |
| Avg. fully-loaded agent cost | $42,000/year | Blended Philippines + domestic oversight |
| Monthly ops cost at 300 agents | $1,050,000/mo | Fully-loaded (mgmt, QA, tooling, overhead) |
| Projected monthly ops cost at 800 agents | $2,800,000/mo | Accepted as inevitable |
| First-contact resolution rate | ~45% | High repeat-contact, low self-serve |
| NPS score | ~27 | Industry average for telco |
| Engineer happiness | 41% | Disconnected from customer outcomes |
| Engineering throughput | Baseline (1×) | Fragmented process, waterfall residue |
The Intervention: Project Blue Glove — Five Prongs
Prong 1 — Vendor reform and incentive realignment
The first act was not technical — it was contractual. Existing chat vendors were paid per chat, which structurally incentivized long, inefficient conversations. Himanshu rewrote the contracts: vendors were now paid for fastest resolution time and concurrent chats handled per agent. Before you fix the system, fix the incentives the system is optimizing for.
Prong 2 — Ticket-pattern analysis and root-cause engineering
An audit of thousands of tickets revealed that the majority of contact volume came from a small number of failure modes — number-porting errors, SIM-activation failures, onboarding friction. These were not support problems; they were engineering failures laundered through the support queue. Himanshu reclassified them as engineering KPIs. A bug that generated 2,000 tickets was no longer a support problem; it was a sprint priority. The team fixed systems, not symptoms.
Prong 3 — Predictive risk modeling and proactive escalation
With one borrowed data scientist — no dedicated data-science org, no ML platform — Himanshu co-wrote the first version of what became a patented High-Risk Customer Algorithm. Using heuristics (previous-carrier reliability scores, payment-method anomalies, activation-timing patterns), the system predicted which new customers were most likely to have problems and fast-tracked them into white-glove support before they ever filed a ticket.
Prong 4 — Early-stage AI and chatbot integration
This was pre-ChatGPT. Before GenAI became an industry reflex, the team built NLP-powered bots trained on anonymized chat transcripts to handle FAQs, billing questions, SIM troubleshooting, and order tracking. Human-in-the-loop validation continuously improved accuracy, substantially reducing agent workload for the highest-volume, lowest-value tier of the queue.
Prong 5 — Customer Experience Engineering (CXE) team
The most structurally important intervention was the CXE team: a hybrid engineering unit whose mandate was not to answer tickets but to eliminate the conditions that created them. They instrumented every customer touchpoint with observability tooling and resolved systemic issues before they reached users — the Run pillar and Serve pillar working together, with engineering owning the accountability chain end-to-end.
The 10× Productivity Story
The Results: 24-Month Recovery Model
Visible operated at carrier scale. Normalized to a 20-engineer digital-services company with 50,000 active users, the proportions of the recovery are identical to what Visible achieved:
| Recovery stream | Month 6 (partial) | Month 12 (steady) | Month 24 (compounding) |
|---|---|---|---|
| Agent cost reduction (300→175 equiv.) | $437,500/mo | $875,000/mo | $875,000/mo |
| Chatbot deflection (40% of routine) | $140,000/mo | $280,000/mo | $350,000/mo |
| Churn reduction via NPS lift (est. 2%) | $80,000/mo | $160,000/mo | $240,000/mo |
| Engineering rework reduction | $45,000/mo | $90,000/mo | $120,000/mo |
| Predictive escalation: avoided crisis churn | $35,000/mo | $70,000/mo | $100,000/mo |
| Total monthly recovery | $737,500 | $1,475,000 | $1,685,000 |
The Visible Numbers: Raw Financial Translation
| Visible KPI | Pre-S+3 | Post-S+3 (24 mo) | Δ |
|---|---|---|---|
| Customer-service cost (normalized) | 100% | 27% | −73% |
| Projected agent headcount at target | 800+ | ~200 (equiv.) | −75% |
| Engineering throughput (delivered value) | 1× | 10× | +900% |
| Engineering Happiness Index | 41% | 93% | +127% |
| NPS score | ~27 | 50+ | +85% |
| First-contact resolution rate | ~45% | 85%+ | +89% |
| Patents secured | 0 | 1 (risk algorithm) | New IP |
Give engineers ownership of the entire customer experience — not just the code — and the support queue stops being a cost center and becomes the product's intelligence system.