Salesforce's AI Strategy: How the CRM Leader Is Rethinking Customer Discovery in 2026

TL;DR#

Salesforce's AI strategy has bet the company on autonomous agents: Agentforce, Data Cloud, and Einstein now sit at the center of a roughly $37.9 billion business that grew 9% in fiscal 2025. Agentforce closed 5,000 deals — more than 3,000 paid — in its first 90 days, has handled over two million customer conversations, and CEO Marc Benioff says AI now does 30% to 50% of the work inside Salesforce, contributing to roughly 4,000 customer-service roles being cut. Data Cloud, the fuel for those agents, crossed 50 trillion records and more than doubled year over year. Yet the CRM giant's understanding of why customers churn, what they actually need, and which features matter still runs largely on structured CRM fields, dashboards, and surveys — data that captures the what but flattens the why. This is the strategic gap: Salesforce automates execution brilliantly, but the qualitative signal that should steer the agents still arrives in dropdowns and NPS scores. Conversational AI interviews close that gap by letting customers explain their reasoning in their own words — the discovery layer that even the world's largest CRM still lacks.

What is Salesforce's AI strategy in 2026?#

Salesforce's AI strategy in 2026 is to convert the world's largest CRM into a platform for "digital labor" — fleets of autonomous AI agents (Agentforce) that act on the unified customer data in Data Cloud and the predictions of Einstein to automate sales, service, marketing, and back-office work. The thesis, repeated by Marc Benioff across earnings calls and Dreamforce keynotes, is that every enterprise will soon manage a blended workforce of humans and agents, and that Salesforce should be the system of record and system of action for both.

It is an execution-first strategy. Agentforce resolves cases, drafts replies, books meetings, and reconciles records. What it does not do — what no agent layer can do on its own — is tell the company why a customer almost left, what job they were really hiring the product to do, or which roadmap bet will matter in eighteen months. That signal lives in conversation, and conversation is exactly what the CRM was never built to hold. For a deeper treatment of the underlying argument, see why AI-first customer research cannot start with a web form.

This article is for CX leaders, product managers, RevOps teams, and founders watching the most consequential enterprise-AI rollout in software and asking the harder question: once the agents are running, where does the understanding come from?

Salesforce at scale: the numbers behind the bet#

Salesforce is the largest pure-play enterprise software company in customer relationship management, and its AI investments are now material to the financials. In its fiscal 2025 results, Salesforce reported revenue of $37.9 billion, up 9% year over year, with subscription and support revenue of $35.7 billion. Those are the foundations the AI strategy is built on top of.

The AI-specific traction is the part worth studying:

• Data Cloud & AI annual recurring revenue reached $900 million, up 120% year over year, per the same fiscal 2025 filing — the fastest-growing line in the business.
• Data Cloud surpassed 50 trillion records and more than doubled year over year, becoming the unification layer that feeds every agent.
• Agentforce closed 5,000 deals in its first 90 days, including more than 3,000 paid, the fastest ramp of any product Salesforce has shipped.
• Agentforce has since handled more than two million customer conversations, according to Salesforce's own reporting.
• Nearly half of the Fortune 100 are now both AI and Data Cloud customers, a concentration of enterprise adoption that few competitors can match.

That scale is why Salesforce's choices set the tempo for the category. When the CRM leader treats autonomous agents as the future of work, the rest of the customer research and CX tool stack reprices around it.

Where Salesforce uses AI today#

Salesforce uses AI today across three tightly linked layers: Einstein for prediction, Data Cloud for unification, and Agentforce for autonomous action. Each layer compounds the next, and together they define what "AI-first CRM" means in 2026.

Einstein: prediction baked into the CRM#

Einstein is Salesforce's machine-learning layer, embedding predictive scoring, forecasting, and generative drafting directly into sales, service, and marketing workflows. It scores leads, predicts which deals will close, surfaces next-best-actions, and generates email and case-summary text. Einstein's job is to turn the structured data already in the CRM — opportunities, activities, case histories — into forward-looking signals reps and managers can act on.

Data Cloud: the unification engine#

Data Cloud is the real-time data platform that ingests and harmonizes customer data from across an enterprise so that agents and predictions operate on one profile instead of fragments. Crossing 50 trillion records, it is the reason Agentforce can answer a billing question with order context or a service question with subscription history. No unified data, no useful agent.

Agentforce: autonomous digital labor#

Agentforce is Salesforce's fleet of autonomous AI agents that take action on Data Cloud and Einstein outputs without step-by-step human supervision. Since the September 2024 launch, Salesforce has shipped Agentforce 2.0, 2dx, 3, the Agentforce 360 platform, and Agentforce Operations for back-office work. Benioff frames the whole program as "digital labor," describing a $3 trillion to $12 trillion opportunity and stating that AI now performs 30% to 50% of the work inside Salesforce — a shift that contributed to roughly 4,000 customer-service roles being cut while support costs fell about 17%.

The results, on Salesforce's own terms, are real. Customers like OpenTable have resolved roughly 70% of inquiries autonomously, and 1-800Accountant hit a 90% case-deflection rate during peak tax week. This is the comparison playbook other enterprises are now running — much like the way we documented Klarna replacing 700 agents with conversational AI.

The blind spot: where forms, fields, and dashboards bottleneck understanding#

Salesforce's AI excels at execution but inherits a structural blind spot: it understands customers through structured fields, dashboards, and surveys, none of which capture why. Agentforce can resolve a ticket flawlessly and still have no idea the customer is one renewal away from leaving — because the reason lives in a sentence no one ever asked them to say.

Three failure modes recur at enterprise scale:

1. CRM fields flatten reasoning into categories. A "Closed Lost — Reason: Price" field tells you the rep's guess, not the customer's actual decision. Maybe price was the polite excuse and onboarding friction was the real cause. The dropdown can't hold that nuance, and the agent acting downstream inherits the wrong premise. Poor and incomplete CRM data is not a fringe problem: 44% of businesses estimate they lose more than 10% of annual revenue to inaccurate CRM data, and poor data quality costs the average organization $12.9 million a year, with 75% reporting they have lost customers because bad data drove the wrong outreach.

2. Dashboards report the what, never the why. A churn dashboard can show that a cohort dropped from 92% to 84% retention. It cannot tell you the cohort churned because a competitor shipped a feature, because a champion left, or because a price change broke trust. The dashboard is a smoke detector, not an investigator. Teams that try to retrofit the "why" onto dashboards end up with the same gap we describe in the complete guide to voice-of-customer programs in 2026.

3. Surveys and NPS measure sentiment but not cause. An NPS score of 32 is a number, not an explanation. Survey response rates are low, the highest-value customers are the least likely to answer, and the open-text box — when it exists — is a single shot with no follow-up. The reasoning that would actually change a roadmap gets compressed into a 1-to-10 rating. We unpack the limits of that approach in AI vs surveys: why conversations win for real customer research.

The irony is sharp. Salesforce has built the most sophisticated action layer in enterprise software, then pointed it at a data substrate — fields, dashboards, scores — that was designed to store transactions, not understand humans. The agents are only as smart as the discovery feeding them, and that discovery still starts, far too often, with a form.

How conversational AI interviews capture what CRM records miss#

Conversational AI interviews capture customer intent and reasoning by letting people answer in their own words while an AI interviewer follows up, probes vague responses, and explores context — producing the qualitative "why" that CRM fields, dashboards, and surveys structurally cannot. Instead of forcing a customer to translate themselves into a dropdown, the interview meets them in language.

The mechanism is the difference:

• Follow-up on the spot. When a customer says "the pricing felt off," an AI interviewer agent asks which part, compared to what, and what would have felt fair — turning a flat field into a decision narrative.
• Scale without flattening. A team can run hundreds of these interviews at once, the way enterprise CX programs run surveys, but each one stays conversational. This is the model Perspective AI built for product teams and CX teams who need depth and volume at the same time.
• Form replacement at the front door. A concierge agent can replace the static intake or onboarding form entirely, capturing intent at the exact moment a customer arrives — the same form-to-conversation shift that named-company case studies keep validating.

This is not a rejection of Salesforce's stack; it is the missing input to it. Imagine the loop closed: Agentforce resolves a wave of cancellation requests, those conversations are mined not just for resolution but for reasoning, and that reasoning flows back as a continuous discovery signal that retrains what the agents and the roadmap prioritize. The same pattern shows up across the enterprises we've profiled — from how HubSpot, a $30B CRM leader, approaches customer research to how Stripe runs customer research across 4 million businesses and how Notion decides what to build. The execution layer is increasingly commoditized. The discovery layer is where understanding — and durable advantage — actually lives.

What enterprise teams should take from the Salesforce playbook#

Enterprise teams should copy Salesforce's execution-layer ambition while fixing the discovery-layer gap it exposes — automate action with agents, but never let structured data be the only way you learn why customers behave the way they do. The lesson of the largest CRM in the world is that scale of action without depth of understanding compounds the wrong decisions faster.

A practical sequence:

1. Map where your "why" currently dies. Every closed-lost reason field, churn dashboard, and NPS comment box is a place where reasoning got compressed. List them.
2. Add a conversational discovery layer at the highest-leverage moment — cancellation, onboarding, post-purchase, or feature request — where intent is freshest. Start one study at the research builder.
3. Feed the reasoning back into the agents and the roadmap. Discovery that doesn't change a decision is just more data. Route it to the teams who own retention and prioritization.
4. Make it continuous, not quarterly. The CRM updates in real time; your understanding of why should too.

For a broader view of how modern teams assemble this, see the AI customer engagement software buyer's framework and the customer research stack product and CX teams actually use.

Frequently Asked Questions#

What is Salesforce Agentforce?#

Salesforce Agentforce is a platform of autonomous AI agents that take action across sales, service, marketing, and back-office workflows without step-by-step human supervision. Launched in September 2024, it has shipped through versions 2.0, 2dx, 3, and the Agentforce 360 platform, closed more than 5,000 deals in its first 90 days, and handled over two million customer conversations. It runs on Data Cloud for unified data and Einstein for predictions.

How does Salesforce use AI for customer research?#

Salesforce uses AI for customer-facing prediction and action far more than for qualitative customer research. Einstein scores leads and forecasts deals, Data Cloud unifies customer profiles, and Agentforce resolves cases — all operating on structured CRM data. The reasoning behind customer behavior, however, still arrives largely through surveys, NPS scores, and CRM fields, which capture outcomes but not the "why," leaving a discovery gap conversational interviews are designed to fill.

What is the difference between Salesforce Einstein and Agentforce?#

Salesforce Einstein is the prediction layer and Agentforce is the action layer. Einstein embeds machine learning into the CRM to score, forecast, and generate text, surfacing recommendations for humans to act on. Agentforce takes the next step, deploying autonomous agents that execute tasks — resolving tickets, booking meetings, reconciling records — based on those predictions and the unified data in Data Cloud, with minimal human supervision.

Why can't CRM data alone explain customer churn?#

CRM data alone can't explain churn because it records what happened, not why it happened. A "Closed Lost" field captures a rep's category guess, a dashboard shows a retention number drop, and an NPS score reports sentiment — but none capture the customer's actual reasoning. Industry research shows 44% of businesses lose more than 10% of revenue to inaccurate CRM data, underscoring how much decision-making rests on a thin, structured signal that misses the messy "why."

How do conversational AI interviews complement a Salesforce stack?#

Conversational AI interviews complement a Salesforce stack by supplying the qualitative reasoning that CRM fields and dashboards cannot. While Agentforce executes and Einstein predicts on structured data, an AI interviewer lets customers explain intent, constraints, and decision drivers in their own words, with real-time follow-up. That reasoning can flow back into the CRM as enriched context, improving both the agents' premises and the roadmap they ultimately serve.

Conclusion: execution is solved, understanding is the new frontier#

Salesforce's AI strategy is the clearest signal yet of where enterprise software is heading: a roughly $37.9 billion business reorganized around autonomous agents, with Agentforce, Einstein, and a 50-trillion-record Data Cloud automating action at a scale no competitor matches. The execution layer is, increasingly, solved. But the largest CRM in the world still understands its customers through structured fields, dashboards, and surveys — data that records the what and flattens the why. That is the gap every enterprise inherits when it bets on agents alone, and it is the gap conversational discovery is built to close.

The takeaway from the Salesforce playbook is not to slow down on automation. It is to refuse to let structured data be your only window into customer reasoning. Perspective AI runs hundreds of AI-led customer interviews at once — following up, probing, and capturing the intent and constraints that CRM records miss — so the agents you deploy are acting on understanding, not just data. Start a study or explore how it works to add the discovery layer your CRM was never built to hold.