Forty-eight percent of MSPs ranked AI and automation as the number one client need for 2026, ahead of security and backup. Thirteen percent are generating meaningful revenue from it. That is Kaseya's 2026 State of the MSP Report, more than 1,000 shops.
That split is the whole problem in agency AI service delivery. Clients are asking. Almost nobody is staffing the week after the demo: reliability, integrations, and a human who can take the exception. That is the part clients pay retainers for.
This week at ChannelCon, GTIA CEO Dan Wensley put the inside-the-shop numbers next to it. Ninety-seven percent of IT service providers already use AI internally. Twenty percent have meaningful governance. Twenty-five percent have a formal revenue strategy.
Pax8 and GTIA launched the Managed Intelligence Alliance on that stage because there is no shared standard for the thing agencies are already selling. SMBs are experimenting. They are stuck turning that experiment into an operating model.
You can sell the agent in an hour. Owning the outcome is the product.
The drop after the demo
MIT NANDA's GenAI Divide (research January through June 2025) looked at 300 public initiatives and dozens of org interviews. Directional, not audited filings. Thirty to forty billion dollars in enterprise GenAI spend. Ninety-five percent of organizations reported zero measurable P&L return.
Generic chat tools got deployed by about 40%. They help a person work faster. They do not move the books. Custom and vendor-sold tools: 60% evaluated, 20% reached a pilot, 5% reached production. Buying it from outside succeeds at about twice the rate of building it in-house.
One CIO in that report: "We've seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects." For work that has to be right, 90% of users still prefer a human.
Gartner spent 18 months naming the same four causes: poor data, weak risk controls, rising cost, unclear value. In July 2024 they forecast at least 30% of GenAI projects abandoned after proof of concept by the end of 2025. In January 2026 they said at least 50% were. S&P Global found the share of companies abandoning most AI initiatives went from 17% to 42% year over year.
The model in the room is rarely the thing that dies. The run after go-live is.
The cheap hour and the paid hour
Klarna is the public version. February 2024: their assistant handled 2.3 million conversations in a month, 75% of chats, work they compared to 700 agents. May 2025: the CEO told Bloomberg they were hiring humans again. The bot was cheaper. The output was lower quality. He said there will always be a human if the customer wants one, and that investing in the quality of human support is the way of the future for them.
People who tested it early already described the bot as a filter to reach a person. The cheap hour was the bot. The paid hour is the person who owns the exception.
Then the legal version. Moffatt v. Air Canada, 2024 BCCRT 149. A chatbot told Jake Moffatt he could book full fare for a bereavement trip and apply the discount after. Policy required applying before. Air Canada argued the chatbot was a separate legal entity. The tribunal said no. A company is responsible for the information on its site. Damages: C$812.02.
If you sell a client-facing agent, the client's brand is on the hook, and so is whoever they hired to run it. Handoff, knowledge-base accuracy, and who answers when the bot is wrong are the product.
Salesforce surveyed 3,075 service professionals in March and April 2026. Agentic AI in customer service went from 39% in 2025 to 66% in 2026. Seventy percent of adopters saw measurable value within 60 days. The number one improved metric was customer satisfaction, not handle time. Seventy-two percent of service operations people called data readiness a major blocker, against 59% of leaders. The people who live in the tickets see the gap first.
Adopting the bot and owning the ticket are different jobs.
White label technical capacity is a named owner
The white-label pitch is more accounts without hiring. Mid-market retainers run $2,500 to $10,000 a month. Rebadged agent platforms run $49 to $500. Headline gross gets quoted at 60 to 80%. Per-agent resale often sits at $300 to $500 a month plus a usage markup.
The honest line in those decks: software resale is a support business. Support for the rebranded product lands on the agency. Fulfillment partners will do the work while you keep the relationship. Quality control moves one step away. White label technical capacity only counts if a named person owns the exception.
Promethean Research asked 119 agency leaders. Average after-tax net in 2025: 13%. Shops that narrowed their mix grew 13% and posted 30% net. Some of the recent compression, per founder Nick Petroski, is clients expecting cheaper work because of AI. If you cannot show who owns the ticket after the demo, the client will take the discount and keep the risk.
Building the agent is about 30% of the work. Deployment, maintenance, and chasing API changes eat the rest. Clients buy a named result. MIT's 2x finding is the quiet offer. What matters is who answers the overnight ticket.
Agency support capacity is the week-after contract
Before the next AI demo, write the support contract for the week after go-live. Name who owns the ticket, the response time in writing, the human escalation path, the knowledge-base owner, and what happens when an API or a policy changes. If you cannot staff that, do not sell the demo. If you can, that is the product, and that is what the retainer is for. That block is agency support capacity.
live_agent:
owner: named person on the account
human_response: first person response, in writing
escalation: a human is always reachable
liability: client brand owns the words; you own the runbook
systems: [crm, helpdesk, billing, identity]
data_owner: whoever lives in the tickets
change_budget: model updates, prompt drift, API breakage
sold_as: outcome
Internal use and a service line are different products. Governance is the missing one. Your client's staff already has a good-enough demo on a personal chat account. What they will pay a retainer for is the version that is allowed, logged, wired into the stack, and backed by a human.
We run production agent systems, including white-label work where the agency keeps the client. Three of those engagements are live under NDA, with a fashion and marketing agency, a marketing and PR firm, and a technology partner. The job is the ticket, the integration, and the handoff. 3am is a retainer, not a default.
We do the work. First conversation is free. No commitment.
Members get the one-page support contract as a PDF at kief.studio.