Skip to main content
Truvisory
Federal

Contact-Center AI at the VA: The Honest SDVOSB Role

Tony Adams 7 min read

If you or a Veteran you know is in crisis: call the Veterans Crisis Line — dial 988, then Press 1, text 838255, or chat at VeteransCrisisLine.net/Chat. Free, confidential, 24/7, and you don’t need to be enrolled in VA health care to use it.

VA’s contact centers answer tens of millions of calls, chats, and messages a year, and they’re in active AI modernization — which makes “contact-center AI” a real and growing line of VA work. For a brand-new SDVOSB, the honest version of the opportunity is narrower than the marketing suggests: the realistic role is the surrounding AI engineering — agent-assist, self-service, summarization, analytics — almost always as a subcontractor, never as a 24/7 call-center prime, and never anywhere near the live Veterans Crisis Line. This spoke, under the VA AI modernization pillar, is the closest sibling to the RAG policy-assistant spoke, because the strongest contact-center wedge — helping a human agent find the right answer in VA policy — is RAG over VA policy.

What does VA’s contact-center stack actually look like?

The front door is 1-800-MyVA411 (1-800-698-2411) — the Veterans Experience Office’s 24/7 single access point, where pressing 0 reaches a live agent, 9 reaches a service-recovery specialist, and 1 routes to the Veterans Crisis Line. VEO itself was made permanent by the PRO Veterans Act of 2025. Behind that door, the operations split by mission: VA Health Connect consolidated 87 legacy facility call centers into 18 regional clinical contact centers handling scheduling, pharmacy refills, clinical triage, and virtual visits — nearly 41 million calls in FY21; VBA runs the national benefits lines (1-800-827-1000, plus education, loan guaranty, and insurance); and the VA.gov chatbot, a Microsoft Power Virtual Agent surface with roughly 240,000 users since 2022, handles unauthenticated, no-PII self-service. A VA OIG review of the clinical contact centers found an FY24 call-abandonment rate around 10% — about double the standard — and noted VHA spent more than $197 million upgrading the centers’ telephone and clinical technology. In other words: real volume, real pain, real modernization budget.

Where is AI already live or in pilot?

More than you’d guess. The 2025 VA AI use-case inventory lists 367 individual use cases, of which 215 are high-impact — more than any other federal agency — and several are contact-center entries. “VA VoiceBot for Call Center Modernization,” a Genesys Cloud natural-language front door to call routing, is listed as high-impact. The VA.gov chatbot handles public self-service. There’s an NLU IVR triage pilot, and the VA CTO’s office has run internal generative-AI drafting and summarization tools for staff. The pattern: VA is buying conversational AI, agent-assist, and analytics across the contact-center surface — and classifying the consequential pieces as high-impact, which matters for how an SDVOSB can ethically touch them.

The Veterans Crisis Line: the line you don’t cross

One part of this landscape is not a market opportunity. The Veterans Crisis Line is distinct, life-safety crisis infrastructure, operated by trained crisis responders, and an outside vendor should not propose generative AI into the live caller path. VA’s own posture is explicit: it does not use chatbots as a replacement for direct crisis intervention. The one well-documented piece of VCL-adjacent AI — ReflexAI’s simulated personas — is responder training only, and is deliberately built so it never engages a Veteran in crisis. That’s a specialist lane, and it isn’t a brand-new firm’s. The practical rule for any contact-center AI work at VA: carve the VCL out of scope explicitly. Even the general VEO front line, which screens for crisis and warm-transfers to the VCL, is a high-stakes path where AI belongs nowhere near the decision — only the human does the screening, and the human does the handoff.

Veterans Crisis Line: dial 988, then Press 1 · text 838255 · chat at VeteransCrisisLine.net/Chat. Free, confidential, 24/7.

What does M-25-21 and the PHI bar mean for this work?

A virtual agent that affects access to benefits, eligibility, or scheduling is, under OMB M-25-21, a presumptive high-impact AI use case — output that’s a principal basis for a consequential decision. That triggers a defined set of obligations: pre-deployment testing in context, an AI impact assessment, ongoing monitoring, real human oversight, a timely human review and appeal path for the Veteran, and end-user feedback — the governance layer covered in the M-25-21 spoke. On the data side, contact-center systems touch PII and often PHI, governed by the Privacy Act, HIPAA, and VA’s sensitive-records statutes (38 U.S.C. § 5701 and § 7332), so the consequential workloads run inside the VA Enterprise Cloud — a FedRAMP High environment. That’s the binding constraint on a Cloudflare-native firm: Cloudflare for Government is FedRAMP Moderate, with High in process and its AI platform targeted for 2026, which makes it a clean fit for the public-internet self-service surface and edge in front of VAEC systems — but not, today, for the PHI-bearing agent desktop, which sits inside a prime’s VAEC tenancy. (CMMC doesn’t enter into it — that’s DoD-only.)

So where can an SDVOSB actually help?

In the surrounding engineering — and the table below is the honest version, with the sensitivity level and where the human agent stays in control made explicit.

// Contact-center work areas — sensitivity, human control, and honest SDVOSB fit
Work area Sensitivity Where the human stays in control Honest fit for a new SDVOSB
Agent-assist RAG over public VA policyLow–MediumAgent reads cited suggestion, decides, writes the recordStrongest fit — sub on SPRUCE/T4NG2
Public self-service virtual agentLow (always-available human handoff)“Press 0” / human fallback always present; bot makes no decisionsGood fit as a sub; small scope could prime via $5M sole-source
Call/chat summarizationMedium (PII transcripts)Agent reviews and edits before saveGood fit as a sub inside a prime’s VAEC enclave
QA / speech-analytics dashboardsLow–MediumSupervisors interpret; no auto-actionGood fit as a sub
Demand / volume forecastingLow (back office, no PII)Workforce-management humans set schedulesLowest-risk wedge — clean fixed-scope win
Conversational IVR / VoiceBot (authenticated)High-impact under M-25-21Live-agent handoff on ambiguityBounded sub-scope only under a prime’s ATO
Accessibility / Section 508 for chat and voiceLowAccessibility specialists decideGood fit as a sub
Live Veterans Crisis Line caller AILife safetyn/aDo not propose. Decline.
24/7 BPO operationsOperationaln/aNot a fit for a brand-new firm

Who’s actually winning this work?

Encouragingly for SDVOSBs, the contact-center primes at VA are increasingly SDVOSB. The May 2026 VEO Contact Center recompete went to Encore JV1 LLC — an SDVOSB joint venture — at a base-plus-options ceiling around $153 million, covering more than 360 agents and roughly 1.57 million inquiries a year. On the digital side, Oddcore (an Oddball/Wilcore SDVOSB joint venture) has taken a string of SPRUCE task orders, including the AI-enabled VA.gov chatbot (~$15M over three years), a VA VoiceBot task, and a $39.6M VA Innovation Unit technical-team task. The contrast worth noting: the giant federal contact-center BPO contracts — like the $6.6B CMS award to Maximus — are large-business territory, and not where a new VA-focused SDVOSB should aim. The realistic move is to sub to the SDVOSB primes already on SPRUCE and T4NG2, bringing a working demo of the surrounding AI — not to chase the BPO prime slot. That’s the teaming path, found through the forecasting workflow and the TAC/SAC buyer map.

Frequently asked

Would Truvisory run a VA call center?
No. Operating a 24/7 BPO isn't a fit for a new firm. The realistic role is the surrounding AI engineering, as a sub.
What's the cleanest first project?
Agent-assist RAG over public VA policy — it's the RAG-policy sibling, the corpus is public, and a Cloudflare-native demo is appropriate.
Can AI go into the Veterans Crisis Line?
Not from an outside vendor, not in the live caller path. The only documented VCL-adjacent AI is responder training, and that's a specialist's lane.
Does Cloudflare's FedRAMP Moderate cover the agent desktop?
No — authenticated, PHI-bearing workflows run inside VAEC (FedRAMP High). Moderate fits the public self-service surface.
Do I need CMMC?
No — CMMC is DoD-only and doesn't apply to VA work.

Working with Truvisory

Truvisory is a brand-new SBA-verified SDVOSB founded by a combat veteran. We have no VA contact-center past performance, and we’re not a BPO prime — and we say so plainly.

Where we can genuinely help is the surrounding AI engineering: RAG agent-assist over public policy, public self-service bots, summarization, QA analytics, and demand forecasting — Cloudflare-native, FedRAMP-aware, M-25-21-aligned, delivered in fixed 90-day scopes as a subcontractor on a contact-center or SPRUCE/T4NG2 prime’s task order. The Veterans Crisis Line stays out of scope, always. If you’re a prime or a VA program office doing this work and you need that kind of bounded sub, that’s the work we can do. For the rest of the map — the RAG sibling, the claims spoke, the compliance trilogy, and how a new firm teams in — start at the pillar.

Veterans Crisis Line: dial 988, then Press 1 · text 838255 · chat at VeteransCrisisLine.net/Chat. Free, confidential, 24/7.

§ Cluster

More in this series