Salesforce has long marketed itself as a company that builds with enterprises, not just for them. But with its AI roadmap, the company is taking that philosophy one step further: it’s effectively crowdsourcing what comes next by treating customer needs as a primary input into product direction. The underlying logic is simple and familiar in enterprise software—if one large organization runs into a problem, others are likely to face the same challenge—but the execution matters. Salesforce isn’t merely collecting feedback after the fact. It’s trying to turn customer thinking into a forward-looking signal that can shape which AI capabilities get prioritized, how quickly they move from concept to delivery, and which use cases become “default” bets rather than one-off experiments.
This approach is especially relevant now because AI roadmaps are no longer linear. In the past, product planning could assume a relatively stable path: gather requirements, build features, ship, iterate. Today, AI development is more like an ecosystem. Model behavior changes, integration patterns evolve, governance expectations tighten, and customers’ expectations rise faster than traditional release cycles. In that environment, internal roadmapping alone can feel like guesswork. Salesforce’s bet is that the best way to reduce that guesswork is to listen to the people who are already trying to operationalize AI inside complex organizations—where data quality, security constraints, workflow design, and adoption all collide.
At the center of this strategy is a feedback loop that treats enterprise customers as co-planners. The company’s premise is that real-world problems are rarely unique. When one customer asks for a capability—say, better automation of a specific kind of service workflow, or more reliable AI assistance for sales operations—it often reflects a broader pattern across industries and departments. Salesforce is trying to capture those patterns early enough that they can influence roadmap decisions before the market standardizes around someone else’s interpretation of what “AI for enterprises” should look like.
What makes this more than a generic “we listen to customers” statement is the implied shift in how priorities are determined. Traditional customer input tends to arrive as feature requests, bug reports, or enhancement requests tied to a particular account. Those inputs are valuable, but they can be narrow. Crowdsourcing the roadmap suggests a different mechanism: Salesforce is looking for themes that repeat across accounts, then translating those themes into product bets. That means the company has to do more than collect requests—it has to synthesize them, categorize them, and decide which ones represent durable demand rather than temporary pain.
In practice, that requires Salesforce to operate like a signal-processing engine. Customer requests have to be mapped to underlying product components: data ingestion and preparation, model orchestration, retrieval and grounding, workflow integration, user experience design, evaluation and monitoring, and governance controls. Two customers might ask for “AI that helps agents resolve tickets faster,” but the real roadmap question is whether the solution needs better knowledge retrieval, improved summarization, tighter context windows, new UI affordances, or stronger guardrails for compliance. The roadmap becomes less about a single feature and more about building the platform primitives that make multiple features possible.
That’s where the crowdsourcing idea becomes strategically interesting. If Salesforce only responded to individual requests, it would risk building a patchwork of capabilities that don’t scale. But if it uses customer input to identify shared primitives—capabilities that can be reused across many use cases—then customer-driven planning can accelerate platform maturity. In other words, the company can turn “many voices” into “one architecture.”
The enterprise AI challenge that makes this necessary is adoption. Enterprises don’t struggle with AI because they lack access to models; they struggle because AI has to work inside messy systems. It has to respect permissions. It has to handle incomplete or inconsistent data. It has to produce outputs that users can trust. It has to fit into workflows that already exist—often with strict compliance requirements and audit trails. And it has to be measurable, because executives want outcomes, not demos.
So when Salesforce invites customers to lead its AI roadmap, it’s also inviting them to define what “working” means. That can include expectations around accuracy, latency, explainability, and safety. It can include how AI should behave when it doesn’t know something. It can include how the system should cite sources, how it should avoid hallucinations, and how it should escalate to humans. These are not minor details; they determine whether AI becomes a tool people rely on or a novelty that gets ignored.
There’s another layer: enterprise customers are often the first to discover second-order problems. A sales team might request AI assistance for drafting emails, but the real issue might be that the drafts need to align with brand voice guidelines, incorporate deal context, and avoid sensitive disclosures. A service team might request AI summarization, but the real issue might be that the summarization must preserve policy-relevant details and remain consistent across languages or regions. Customers see these issues in the wild, and their requests can reveal where the platform needs to mature.
Salesforce’s crowdsourcing approach can therefore function as an early warning system. Instead of waiting for broad market adoption to surface common failure modes, the company can detect them through customer deployments and translate them into roadmap priorities. That’s particularly valuable in AI, where the cost of getting it wrong can be high—not just in user frustration, but in reputational risk and compliance exposure.
Still, crowdsourcing a roadmap is not without tension. The biggest risk is that customer input can skew toward the loudest or most urgent accounts rather than the most broadly valuable capabilities. Enterprises vary widely in maturity, data readiness, and internal appetite for change. Some customers may push for advanced automation because they have strong process discipline and clean data. Others may need foundational improvements first. If Salesforce simply followed the most aggressive requests, it could end up optimizing for a subset of customers rather than the majority.
To manage that, Salesforce has to balance two competing goals: responsiveness and generality. Responsiveness means moving quickly when customers identify a real need. Generality means ensuring that what gets built can serve many organizations, not just one. The roadmap has to be shaped by patterns, not anecdotes. That’s why the “crowdsourcing” framing matters: it implies that Salesforce is aggregating inputs across customers and using repetition as a proxy for broader demand.
Another challenge is timing. AI roadmaps can be derailed by dependencies outside the company’s control: model availability, regulatory shifts, changes in data privacy expectations, and evolving integration requirements. Even if customers agree on what they want, the path to delivering it can be constrained by technical realities. Salesforce’s approach likely requires a disciplined product management process that can separate “what customers want” from “what is feasible now” and “what will be feasible soon.” Otherwise, crowdsourcing can create unrealistic expectations and lead to churn when timelines slip.
There’s also the question of differentiation. Many enterprise vendors are racing to add AI features. If Salesforce’s roadmap is heavily customer-led, it could risk becoming a follower rather than a leader—building what customers ask for rather than anticipating what customers will need next. The counterbalance is that customer-led roadmaps can still be visionary if Salesforce uses customer input to identify emerging categories. For example, customers might not ask for “AI agents” in those exact terms, but they might request workflows that require multi-step actions, tool use, and orchestration. Those requests can reveal a shift in how customers want AI to operate. In that sense, customer input can help Salesforce lead by recognizing the direction of travel earlier than competitors.
This is where Salesforce’s unique position matters. As a platform embedded in sales, service, marketing, commerce, and operations, Salesforce sits at the intersection of data, workflow, and user behavior. That gives it a different vantage point than a standalone AI vendor. When customers request AI capabilities, they’re often requesting them in the context of existing Salesforce objects, processes, and integrations. That context can help Salesforce build AI that is not just accurate in isolation, but useful in the flow of work.
The crowdsourcing strategy also aligns with how enterprise buyers evaluate AI. Buyers don’t just ask, “Can it generate text?” They ask, “Can it improve my KPIs?” They ask, “Can it reduce time-to-resolution?” “Can it increase win rates?” “Can it lower costs?” “Can it comply with our policies?” “Can it integrate with our systems?” Customer-led roadmapping can keep those evaluation criteria front and center, because customers are the ones who will measure outcomes.
If Salesforce is doing this well, the roadmap should start to reflect a clearer connection between AI features and business metrics. Instead of shipping AI as a collection of isolated tools, the company can prioritize capabilities that unlock measurable improvements across departments. That might mean focusing on AI that reduces manual effort in service resolution, improves lead qualification, accelerates content creation with brand governance, or helps teams navigate complex customer histories. It might also mean prioritizing the infrastructure that makes those improvements sustainable: monitoring, evaluation, and continuous learning loops that keep performance stable as data changes.
One of the most compelling aspects of this approach is the potential for faster iteration on trust. In AI deployments, trust is not a one-time checkbox. Users learn over time what the system does well and where it fails. Enterprises also refine their governance policies as they gain experience. If Salesforce is actively incorporating customer input into its roadmap, it can treat trust-building as a product priority rather than an afterthought. That could include better transparency features, improved confidence signaling, more robust citation and source grounding, and clearer controls for administrators.
It also raises an interesting possibility: customers might influence not only what Salesforce builds, but how Salesforce measures success. For example, a customer might want AI to be conservative—prioritizing fewer but more reliable recommendations. Another might want higher recall—accepting more suggestions but requiring human review. Those preferences can shape evaluation frameworks and product settings. Over time, Salesforce could develop a more flexible approach to AI behavior that adapts to organizational risk tolerance and workflow needs.
Of course, there’s a delicate balance between personalization and standardization. Too much customization can fragment the product