Salesforce acquisitions 2025: the $10B blueprint behind Agentforce
If you're trying to operationalize AI agents in the enterprise, you've probably learned the hard lesson: agents are easy to demo, hard to deploy.
Not because the models are weak, but because enterprise reality is messy. Fragmented data. Unclear process ownership. Inconsistent governance. Integration sprawl. 80% of AI projects fail due to data issues, a stat that aligns with what most Salesforce teams see once they leave the pilot phase.
That's why Salesforce's 2025 acquisition spree matters. In 2025, Salesforce acquired 12+ companies for ~$10B+, and the pattern is clearer than the press headlines: this wasn't "buying AI." It was buying the prerequisites for Agentforce to work in production.
What Salesforce bought (and where it fits in Agentforce)
| Acquisition | Primary capability | Agentforce layer | Why it matters | What you should do now |
|---|---|---|---|---|
| Informatica (~$8B) | MDM, governance, catalog, data quality | Data trust | Agents depend on trusted, connected context | Baseline data quality/MDM + governance ownership |
| Own Company (~$1.9B) | Backup, recovery, data protection | Data resilience | AI increases data criticality and blast radius | Define retention, recovery, sandbox strategy |
| Zoomin (~$450M) | Unstructured content to AI-ready knowledge | Knowledge/context | Agents need clean, searchable knowledge sources | Inventory KB/docs and align knowledge strategy |
| Waii (undisclosed) | Natural language → SQL | Data access | Unlocks self-serve analytics for non-technical users | Harden semantic layer + metadata management |
| Apromore (undisclosed) | Process mining + simulation | Process intelligence | You can't automate what you can't see | Pick 1–2 processes; instrument before automating |
| Regrello (undisclosed) | Supply-chain workflows | Operational workflows | Agentic ops needs collaborative, governed flows | Define cross-team ownership + SLA/controls |
| Spindle AI (undisclosed) | Forecasting + scenario modeling | Decision intelligence | Agents become useful when tied to business outcomes | Define KPIs; connect data to forecasting inputs |
| Convergence.ai (est. $150–200M) | Autonomous agents for complex tasks | Agent execution | Moves agents from chat → action | Define "allowed actions," guardrails, audit trail |
| Doti (est. ~$100M) | Enterprise search ("enterprise brain") | Retrieval/search | Agents need retrieval across systems | Map data sources; decide access control model |
| Qualified (undisclosed) | AI SDR + website conversion | GTM automation | Revenue workflows are a prime agent use case | Align marketing/sales handoffs + routing rules |
| Bluebirds (undisclosed) | Prospecting automation | GTM automation | Automates top-of-funnel busywork | Define ICP + compliance for outreach |
| Moonhub (acqui-hire) | AI talent | R&D capacity | Talent hubs are now product strategy | Track roadmap maturity + support readiness |
Values are based on public reporting where available; some deal values are undisclosed or estimated.
Pattern 1: The Informatica "Walk Away → Buy Later" Move
Salesforce's most important move wasn't announcing an acquisition. It was walking away.
In April 2024, Salesforce reportedly abandoned the Informatica deal after three rejected offers ($33, $35, $36 per share).
Both stocks tanked: Salesforce dropped 7%, Informatica cratered 59%.
Fourteen months later, Salesforce came back at $25/share and closed the deal. Robin Washington, Salesforce's COO/CFO, called the approach "methodical, patient, and decisive."
What this signals: If Salesforce is willing to wait 14 months and still pay $8B for data management, the Agentforce roadmap assumes enterprise-grade data trust is non-negotiable.
Pattern 2: Data Trust Is the Foundation (Because Agents Hallucinate Without It)
Benioff summarized it bluntly: "Without clean, connected, trusted data there is no intelligence, only hallucination."
That's the logic behind the data cluster:
Informatica for governance, catalog, MDM, data quality
Own for backup/data protection
Zoomin for unstructured knowledge processing
Waii for natural language data access
Data readiness checklist:
Pick 1–2 high-value agent use cases (customer support deflection, sales routing, renewal assistance)
Identify authoritative systems for each entity (Account, Contact, Product, Order, Case)
Define data ownership + stewardship (business + IT), not just pipelines
Put governance in place early (PII, consent, retention, access controls)
Instrument quality: duplicates, missing fields, stale attributes, conflicting keys
Pattern 3: Process Intelligence Is the "Middle Layer"
Apromore is the strategic tell most teams missed.
The company has been named a Leader in Gartner's Magic Quadrant three times (the only process mining vendor with that distinction besides Celonis). Process intelligence is the new competitive frontier.
The principle: If you automate before you measure, you scale the wrong workflow. AI agents don't fix broken processes; they accelerate them.
Where to start:
Choose one cross-team process (lead → opportunity → quote, case → resolution, order → fulfillment)
Map the "happy path" and the 3 most common exceptions
Decide what the agent can do vs. what needs human approval
Define auditability: what changed, why, and who approved
Pattern 4: Agents Are Moving From "Chat" to "Do"
Convergence.ai (London) and Doti (Israel) point to the next phase: autonomous execution and enterprise retrieval.
Convergence was founded in April 2024 and acquired 14 months later; Salesforce described it as "the nucleus of a world-class AI lab" in London. Doti is positioned as an enterprise search "brain," strengthening Salesforce's Israel AI R&D hub.
Winning designs will be "bounded autonomy":
Clear allowed actions (create/update, send, approve, escalate)
Tight permissioning (role, record, field)
Strong retrieval strategy (what the agent can see and cite)
Monitoring + feedback loops (quality, drift, escalation rate)
Pattern 5: The AI Talent War Is Geographic
Three acquisitions are Israeli-founded (Own, Zoomin, Doti). According to LinkedIn's AI Talent Concentration Index, Israel ranks #1 globally with 1.98% of its workforce in AI roles, ahead of Singapore, Luxembourg, Estonia, and the rest of the top 10.
In parallel, Salesforce is building a London AI lab around the Convergence team (12 ex-DeepMind, 8 Meta AI veterans), paired with a $6B UK investment commitment through 2030.
What this signals: Roadmap risk isn't only product features. It's whether Salesforce can staff and ship the integration work fast enough.
Pattern 6: Acquisition Velocity Changes Your Due Diligence
Doti: ~1 year from founding to ~$100M acquisition
Convergence.ai: 14 months from founding to acquisition
Fast acquisitions are great for innovation. They also mean you should plan for:
Roadmap churn (renames, packaging changes, merged SKUs)
Integration lag (connectors, permissions model, admin UX)
Security review timelines (especially for regulated industries)
Pricing uncertainty (bundles vs add-ons)
Action: When selecting an Agentforce use case, pick one where you can tolerate product evolution without operational disruption.
What This Means for Your 2026 Roadmap
If you're a CTO or CRM leader planning 2026, start with your data foundation. Everything else depends on it:
Choose your data strategy: MDM, governance, and stewardship are now program-level concerns.
Instrument before you automate: Add process mining/measurement before deploying agents at scale.
Design "bounded autonomy": Define what agents can do, when they must escalate, and how you audit actions.
Plan integration as architecture, not plumbing: Expect deeper convergence across Data Cloud, MuleSoft, Informatica.
Treat change as a constant: Bake in quarterly reviews for product maturity, packaging, and controls.
For teams ready to move from strategy to execution, our Agentforce implementation guide breaks down phases and timelines.
Salesforce has the pieces. The open question is how fast the platform experience converges, and whether your foundation is ready when it does.
FAQ
Do I need Informatica to succeed with Agentforce?
Not necessarily. But you do need the capabilities Informatica represents: governance, quality, lineage, and trusted identity across systems. For many enterprises, that's the bottleneck.
What's the minimum "data readiness" for enterprise AI agents?
An authoritative system per key entity (Account/Contact/Product), consistent permissions, and measurable data quality. If your team can't explain where a field comes from and who owns it, your agents will struggle.
Why does process mining matter if I already have automation?
Automation tools execute; process mining explains reality. It shows variants, exceptions, bottlenecks, and where humans actually intervene. That visibility is critical before letting agents act.
What should I watch for as Salesforce integrates these acquisitions?
Packaging changes, admin UX maturity, permission models, logging/audit capabilities, and support/reference architectures.
What's a sensible first Agentforce use case?
Pick a workflow with clear inputs/outputs and a safe fallback (e.g., case triage, knowledge retrieval + summarization, lead routing with human approval).
Key Sources
Official: Salesforce 2025 recap | Informatica completion | Convergence.ai | UK investment
Analysis: Everest Group | CNBC | Constellation Research
Talent: LinkedIn AI Talent Index | Globes on Doti
Ready to Build Your Agentforce Foundation?
If you're planning an Agentforce program and want a practical readiness review (data foundation, process visibility, or agent design), we can help.
