2 Utah companies are quietly changing how AI actually works for real businesses

Conversion rates roughly triple when practices switch their website chat to MyAdvice’s Maya, and review response times improved from an average of >20 days to under 24 hours with nearly 100% of reviews receiving a response. MyAdvice builds custom small language models per client to reduce hallucinations and improve accuracy; GCommerce (managing ~2,500 properties) is developing protocols to push hotel inventory directly into AI systems to bypass OTAs and reclaim direct bookings. Both firms flag trust, staff impacts and regulatory concerns but argue that honest, well-deployed AI will create a compounding competitive advantage.

Analysis

Vertically specialized, client-specific LLM deployments create a qualitatively different moat than generic chatbots: they turn proprietary, verified business data into a persistent yield stream (higher conversion, lower churn) rather than a one-off marketing lift. That moat is defensible only if the vendor solves operational scale — model retraining, latency, compliance and labeled-data plumbing — which creates an addressable market for MLOps, managed model hosting and verticalized SaaS rather than pure-play consumer LLMs.

For hotels, direct inventory feeds into LLMs are a credible pathway to recapture customer relationships and reduce intermediated commissions, but the economics are two-layered: hotels must invest in standards/APIs and reconciliation systems up-front, while gatekeepers (OTAs or large LLM vendors) can monetize placement via pay-to-play or preferential routing. Expect initial share shifts in niche segments and chains that move fast, with a more material reallocation of gross bookings only after standardized protocols and integration stacks scale across hundreds of properties.

Key near-term catalysts and risks are asymmetric. Catalysts: pilots with national chains, published standards for inventory APIs, and a few high-visibility wins showing unit-economics improvement (6–18 months). Risks: deep-pocketed OTAs or hyperscalers building native integrations, regulatory pushback over data access or anti-competitive bundling, and the nontrivial engineering cost to maintain real-time accuracy and fraud/reconciliation (any of these can delay meaningful revenue migration by 12–36 months).

Contrarian edge: the market may be overstating immediate OTA vulnerability. OTAs control demand aggregation and have balance sheets to subsidize placement, and hyperscalers can internalize hotel inventory faster than hotels can standardize it. A staggered, multi-year drift toward direct LLM bookings is more likely than an abrupt disintermediation; that favors investors focused on durable implementation winners (consultants, cloud infra) over binary bets on OTA collapse.

Key Decisions for Investors

• Pair trade (12 months): short EXPE / long MAR. Thesis: EXPE faces commission pressure if hotels win direct LLM channels; MAR participates via higher direct booking margin and brand capture. Target 3:1 upside/downside — size short 60% of pair, stop-loss if EXPE outperforms by 15% intrapair within 3 months; take profit if pair moves 20% in favor within 12 months.
• Long ACN (6–12 months): buy ACN outright or a 9–12 month call spread. Thesis: enterprise customers will pay for systems integration and MLOps as they operationalize vertical LLMs in healthcare and travel. Risk: project timing/execution; reward: 12–20% re-rating if ACN wins multiple large rollouts; cap position to 2–4% portfolio.
• Long cloud/infra exposure (12–24 months): overweight MSFT or GOOGL for model hosting and vector DB demand. Thesis: hosting thousands of per-client small LLMs raises demand for reliable cloud GPU, embeddings, and search infra. Risk: margin pressure if hotels/SMBs choose on-prem or cheaper providers; reward: durable revenue growth and multiple expansion as enterprises standardize on a small number of clouds.
• Tactical option hedge (6–18 months): buy EXPE 12–18 month puts as tail-risk insurance rather than outright short if you are doubtful of timing. Cost-justify as insurance (pay ~3–6% premium for downside protection); use if you want exposure to potential OTA share loss but respect execution and timing risk.