MENU
Home Expertise Our Work The Method About Team Research PestPilot LotPilot Annual Letter Book a Call Request an Evaluation
RESEARCH & QUALIFICATIONS · DRS

The work, the thinking, and the proof.

DRS Automation Group publishes what it ships. Capabilities validated in production. Active research threads we're studying. The technical stack we're qualified to operate against. And field notes — short observations from real engagements, written for operators considering artificial intelligence.

01 / CAPABILITIES TESTED

Validated in production, not just on paper.

Every capability listed here has been deployed, tested, and operated for real clients. No pilots. No proofs-of-concept. Production systems with real customers on the other end of the line.

VOICE AI

Inbound voice agents

Full conversational AI handling inbound calls — greeting, qualification, scheduling, dispatch routing, escalation. Live in production with sub-second latency.

PLATFORMVAPI
STATUSDEPLOYED
SMS AI

Two-way SMS conversation agents

Lead-qualification and multi-day follow-up agents over native SMS. Handles inbound, drips outbound, books appointments directly into client calendars. Sub-30-second response time on first reply.

CHANNELTWILIO
STATUSDEPLOYED
DISPATCH AI

Optimal-technician routing

Real-time skill-matrix routing with day-off detection, geographic radius enforcement, and load balancing. Pulls live spot-truth from the field-service system; never books a tech who isn't working.

METHODPHASE-B
STATUSDEPLOYED
SOCIAL DM AI

Instagram & Messenger agents

Brand-voiced conversational AI on Instagram DM, Facebook Messenger, and WhatsApp. Handles inbound qualification, routes to booking, escalates hot prospects. Three brands in production.

PROTOCOLMETA GRAPH
STATUSDEPLOYED
CRM AUTOMATION

Field-service integrations

Bidirectional integrations with major field-service platforms — FieldRoutes, HubSpot, Zoho, custom CRMs. Real-time spot reads, write-back of confirmed appointments, lead capture into the right pipeline stage.

PATTERNAPI + WEBHOOK
STATUSDEPLOYED
OPS & MONITORING

Production observability

Heartbeat monitoring, OpenRouter credit alerts, FieldRoutes rate-limit tracking, daily metrics aggregation, automated daily snapshots. Watchdog runs every 15 minutes.

STACKSUPABASE · CRON
STATUSDEPLOYED
02 / ACTIVE RESEARCH

What we're studying right now.

A handful of research threads we're actively investigating. Each one ties to a specific operator problem we believe artificial intelligence will reshape in the next 18-36 months.

THREAD
01 / 2026
ACTIVE

Vertical AI versus horizontal AI — the precision differential.

We're documenting performance differentials between vertical-AI systems (engineered for one industry) versus horizontal-AI receptionists deployed at the same operator. Early data suggests vertical agents close 30-50% more qualified bookings per hundred calls handled. Working paper in progress.

PESTPILOT · LOTPILOT
THREAD
02 / 2026
ACTIVE

Model selection for operator-AI: cost-to-precision frontiers.

Benchmarking Claude, GPT, and open-source models across the specific tasks operators care about — lead qualification, scheduling logic, dispatch decisions. Optimizing for the precision-per-dollar curve at production volume, not for benchmark scores.

VAPI · OPENROUTER
THREAD
03 / 2026
SHIPPED

Phase-B intelligent dispatch — spot-truth routing at scale.

Built and deployed a Phase-B optimization layer that auto-assigns the optimal technician on every booking — skill matrix, day-off detection, real spot occupancy load balancing. Eliminated dispatch overhead for the first PestPilot deployment.

PESTPILOT · DISPATCH
THREAD
04 / 2026
EXPLORING

Voice-AI failure modes in production — error taxonomy.

Building a public taxonomy of failure modes we've observed in production voice deployments — silent failures, schema-drift errors, latency cliffs, escalation logic gaps. The goal is a shared vocabulary the industry currently lacks.

VOICE · QA
THREAD
05 / 2026
EXPLORING

Which verticals productize and which stay custom.

Mapping the characteristics that determine when a custom build is productizable into a vertical operating system versus when it should stay bespoke. Pattern recognition across pest control, automotive, mentorship, and field services.

FIRM RESEARCH
03 / QUALIFICATIONS

The stack we're qualified to operate against.

DRS isn't a certifications-on-the-wall firm. We're qualified by the systems we've shipped and continue to operate. Every technology listed here is in active production for at least one client.

The bias is intentional: we go deep on the platforms that matter to operator-led businesses, instead of broad on every framework. When we recommend a stack, it's because we've run it under load.

If your business runs on something we haven't operated against, we'll tell you — and either bring in a vetted specialist or refer you to someone better suited. Honesty about scope is part of the qualification.

PRODUCTION STACK
VapiVOICE AI
TwilioSMS · TELEPHONY
Claude (Anthropic)LLM
GPT (OpenAI)LLM
OpenRouterMODEL ROUTING
n8nWORKFLOW
Supabase / PostgresDATABASE
Vercel / Next.jsFRONTEND
FieldRoutesFIELD SERVICE
HubSpot / ZohoCRM
Meta Graph APISOCIAL
Cal.comSCHEDULING
04 / FIELD NOTES

Short notes from the field.

Working observations from real engagements. Written for operators considering artificial intelligence — and for ourselves, so the firm's institutional knowledge compounds.

MAY 2026FIELD NOTE

The 30-second response is the whole game.

Across every operator-led business we've worked with — pest control, auto dealerships, field services — speed of first response correlates more strongly with closed business than any other lever. Sub-30 seconds beats every other variable, including pitch quality.

SHORT · 2 MIN READ
APR 2026FIELD NOTE

Most "AI failures" are integration failures.

The thing operators call "the AI didn't work" is rarely the AI — it's the connection between the AI and the system of record. CRM write-backs failing silently. Calendar slots that don't exist getting offered. The model is fine; the plumbing isn't.

SHORT · 3 MIN READ
APR 2026FIELD NOTE

Vertical AI compounds. Horizontal AI plateaus.

A generic AI agent gets good at the first 80% of pest-control conversations quickly and then stays there. A vertical agent built for pest control keeps improving — because every edge case the firm encounters becomes part of the system. The narrower the scope, the steeper the learning curve.

SHORT · 4 MIN READ
MAR 2026FIELD NOTE

What "live before invoice" actually costs.

We ship every system into production before sending the final invoice. It looks risky on paper. In practice it accelerates client decision-making and weeds out the operators who wouldn't have signed off anyway. The system has to actually work for the relationship to actually start.

SHORT · 3 MIN READ

Want the firm's research applied to your operation?

Book a 30-minute discovery call or request an in-depth evaluation. We'll bring the thinking, the stack, and the production track record to bear on your specific business.