Customer-facing AI agents are moving into enterprise channels, not just direct sales

The channel shift

Two announcements point the same way for sellers of customer-facing AI. Parloa says its Agent Management Platform is now available as an SAP Solution Extension, so SAP customers can deploy AI agents connected to SAP business data. Sierra says its AI agent platform is now available on Google Cloud Marketplace, so enterprise customers can procure it through existing cloud commitments. In both cases the buyer does not start a new vendor relationship from scratch. The agent arrives inside a system and a budget the buyer already uses. m:…1339 m:…1303

This suggests the distribution question for customer agents is changing. A founder selling an agent no longer has to win only on model or demo quality. The founder also has to win a place in the procurement path the customer already trusts. Parloa cleared SAP's premium qualification process and is now available with official SAP support, according to Parloa. Sierra offers unified billing through the Google Cloud Console, according to Sierra. Those administrative details are the point for an enterprise buyer, because they reduce the work of approving, paying for, and governing a new tool. m:…1339 m:…1303

What buyers get in each channel

Through SAP, the pitch is connection to operational context rather than a standalone chatbot. Parloa says SAP customers can deploy AI agents that natively connect to SAP business data and operational context, and that its platform is one of the only conversational AI agent management extensions available for SAP Customer Experience solutions. The example Parloa gives is Fressnapf, which uses Parloa's voice agent to resolve customer questions and issues. An agent that can see order, account, and service context can resolve a case instead of merely answering about it. Parloa gives no deployment numbers or resolution rates for that example in the cited material, so a buyer should treat it as an illustration, not a measured result. m:…1339

Through Google Cloud Marketplace, the pitch is simpler procurement and pricing tied to results. Sierra says marketplace purchases count toward existing Google Cloud commitments, and that it charges on an outcome-based model, meaning it charges for agent results rather than token consumption. Sierra also says nearly 50 percent of the Fortune 50 work with Sierra, one in three leading banks work with Sierra, and five of the ten largest healthcare companies in the world work with Sierra. Those adoption claims come from Sierra itself. Still, the pricing detail matters, because outcome pricing changes what a customer compares: not cost per token, but cost per resolved case or completed task. m:…1303

The build alternative

HeyGen shows a third route, and it runs in the opposite direction from a marketplace listing. Instead of buying distribution through a platform, HeyGen built the model itself from traffic it already had. HeyGen says it routed approximately 6 billion characters of speech a week to third-party engines before launching its own model, and collected 644,284 blind user picks for voice cloning since late June 2026. HeyGen Voice entered its internal blind scoreboard in July 2026 and reached the top position by September 28, 2026, according to HeyGen. By October 9, 2026, HeyGen says, HeyGen Voice ranked first on an independent leaderboard, and as of October 21, 2026, HeyGen Voice Turbo and HeyGen Voice held the top two spots on the Artificial Analysis controlled-voice leaderboard. m:…3270

The economics HeyGen reports are the reason a founder might consider building rather than renting. HeyGen says its annual ElevenLabs usage cost decreased from 11.9 million dollars to 3.4 million dollars following the transition to HeyGen Voice, and that the share of voice quality ratings that are thumbs-up rose from 73 percent to 82.5 percent as users moved to HeyGen Voice. HeyGen Voice also overtook ElevenLabs in the share of speech generated within the HeyGen platform during the week of October 5, 2026, according to HeyGen. The same material notes a limit: ElevenLabs currently leads HeyGen in Chinese language voice generation. The build route does not automatically win every language or use case. It wins where the builder has enough of its own traffic and preference data to train and test against. m:…3270

What this means for a founder

Taken together, the three cases describe a choice rather than a single playbook. A founder can sell through a channel the enterprise already buys from, as Parloa does through SAP and Sierra does through Google Cloud Marketplace, and accept that channel's qualification, billing, and governance rules. Or a founder with large, repeated customer traffic can build the core model in-house, as HeyGen did, and turn a vendor bill into an asset the company controls. The middle path, buying a model and selling it direct with no channel and no owned data loop, looks harder in this light, because it competes on distribution and model quality at once without an advantage in either. m:…1339 m:…1303 m:…3270

For a founder deciding now, the practical questions follow from the evidence, not from a general theory of agents. First, where does the target buyer already pay, and can the product be listed, qualified, and billed there? Second, does the product's pricing match what the buyer wants to buy, meaning outcomes and resolutions rather than tokens? Third, is there enough of the company's own customer traffic to justify building a model, or would that effort simply reproduce a vendor's model at higher cost? It does show that companies winning enterprise attention answer them through channels, pricing, or owned models, rather than through a direct pitch alone. m:…1339 m:…1303 m:…3270