IBM channel chief on AI maturity

Enterprise AI is having its grown-up phase. That was basically Kareem Yusuf’s message at IBM Think in Boston this week — not that AI is new, but that the easy part is over. The demo stage got everyone’s attention. Now buyers want proof that a system changes how work gets done, who makes decisions, and whether the result is actually worth the money. IBM is trying to show that shift inside its own channel before asking partners to sell it outside. ### Who is Kareem Yusuf here? Yusuf runs ecosystem, strategic partners and initiatives at IBM, so he sits right where vendor strategy meets the channel. In the interview, he framed AI maturity less as a model-quality story and more as an operating-model story — how partners sell, how IBM incentives line up, and how internal teams decide what to automate versus what to leave to people. That lines up with IBM’s broader Think 2026 pitch, which centered on an “AI operating model” for moving beyond scattered pilots. (sixfivemedia.com) ### What changed in the last year? The tone changed. Yusuf said maturity across the ecosystem accelerated significantly over the past 12 months, and the center of gravity moved from proving that AI can do something to proving financial benefit. That is a big difference. A flashy assistant can win a meeting. It does not win a budget cycle unless someone can show faster deal flow, better conversion, lower service cost, or some other operational gain. (sixfivemedia.com) IBM’s own event language made the same point — enterprises have invested heavily in AI, but only a few think the spend is paying off. ### Why does Kareem.ai matter? Because it is IBM using AI on itself. Kareem.ai is an internal channel health analytics tool now in active pilot, and Yusuf described it as a way to manage partner performance rather than as some generic chatbot. That matters because channel chiefs talk all the time about AI changing sales ecosystems. Much fewer actually point to a live internal system with a defined job. IBM is trying to avoid the “sell it first, figure it out later” trap. (sixfivemedia.com) ### What was the hard part? Turns out the problem was not getting enough data. It was deciding which data deserved to exist in the model at all. Yusuf’s test was blunt: if an answer would not change what he does next, the data does not belong. That is a useful filter because enterprise AI projects often die from overeating — too many dashboards, too many signals, not enough decisions. (sixfivemedia.com) ### Which metrics survived that filter? Two of them carried most of the weight: business partner-originated opportunity and business partner-owned opportunity. Those are channel-specific measures, but the bigger lesson is general. Good enterprise AI usually starts by shrinking the problem. Instead of asking a model to explain the whole business, you give it a narrow lane tied to an action. Think less “tell me everything about partner health” and more “show me the indicators that change coverage, investment, or follow-up.” (sixfivemedia.com) ### Where do humans stay in the loop? Right at the point of consequence. Yusuf’s framing was not “replace judgment.” It was closer to “surface the signal, then let people decide.” That is the practical version of human override in enterprise AI — not a philosophical slogan, but a workflow rule. Especially in channel management, a recommendation only matters if a person can inspect it, trust it, and act on it without losing context. (sixfivemedia.com) ### Why does this matter for partners? Because IBM wants more revenue to come through partners — Yusuf told CRN in December that IBM is aiming for 50 percent ecosystem revenue, up from about 40 percent reported earlier by CEO Arvind Krishna. If that goal is real, partner programs cannot just reward resale volume. They have to reward measurable influence, repeatable services, and AI deployments that survive contact with real operations. (sixfivemedia.com) ### Bottom line? The interesting part of this interview is not that IBM says AI is important. Every vendor says that. The interesting part is the narrower claim — enterprise AI is maturing when companies can name the workflow, the decision, the metric, and the human backstop. IBM is now trying to make that case with its own channel as the test bed first. (sixfivemedia.com) (crn.com)