How Top AI Companies Are Leveraging Integrations to Power Next-Gen Applications

Introduction

Integrations are becoming the backbone of leading AI software. As AI products transition from novelty to daily utility, users expect them to integrate seamlessly into their workflows—whether that means automating tasks, accessing external data, or triggering actions across third-party tools.

We wanted to uncover the integration strategies behind the top enterprise AI companies (like OpenAI, Zoom, Notion), so we analyzed over 100 AI companies to share these insights with you.

Across hundreds of AI companies, four integration use cases consistently emerge:

1. Tool Calling

2. Data Ingestion for RAG (Retrieval-Augmented Generation)

3. Workflow Builder Actions

4. Bidirectional Syncs

After analyzing 100 of the top AI SaaS companies, we found that:

• 29% offer tool calling

• 38% support data ingestion

• 35% provide workflow builder actions

• 21% enable bidirectional syncs

These use cases aren’t just technical features—they’re strategic differentiators that define how AI products create value. Let’s break down each function in detail. In this report, each section is divided into 4 key parts.

1. Definition of the use case

2. Market adoption stats

3. Case studies

4. Best practices for implementation

Now, let’s dive in.

Tool Calling

What is Tool Calling?

At its core, tool calling enables an AI agent to recognize when it needs to use an action to complete a task or retrieve data to answer a question, select the appropriate tool for the job, and execute custom code using that tool. Examples of tool calling include agents sending a Slack message on behalf of their user or searching someone’s Google Calendar to find availability.

Market Adoption

Roughly 29% of companies have implemented tool calling, but adoption is highly uneven across industries. Contact center and CX platforms lead the way at 75–80%, reflecting their need to orchestrate complex, multi-system customer journeys in real time.

Interestingly, infrastructure and AI platform vendors (40%) are pushing the boundaries by exposing developer-defined tools via function calling — signaling that tool calling may eventually become a developer-first paradigm that trickles down into every vertical.

Sales and RevOps tools are the next most active segment at ~35%, as sales agents increasingly rely on AI assistants to automate CRM updates, meeting scheduling, and outreach tasks.

By contrast, productivity and HR tools show much lower adoption (5–10%), suggesting that while tool calling is powerful, these categories still emphasize collaboration and data centralization over autonomous action.

Most common integrations

Tool calling thrives where agents need to trigger real-world actions, and the top integrations reflect that:

• Slack (~60%) – The most common destination for agent-triggered notifications, updates, and insights.

• Salesforce (~55%) – Central for creating and updating leads, opportunities, and cases.

• ServiceNow (~40%) – Dominant in IT and CX for ticket creation and incident resolution.

• Zendesk (~35%) – A staple for support agents, used for ticketing and customer case updates.

• Google Calendar / Outlook (~25–30%) – Critical for scheduling, meeting availability, and event management.

• Custom APIs (~50%) – Especially in data/AI platforms, developers extend tool calling beyond off-the-shelf SaaS by exposing domain-specific actions.

Case Studies

Moveworks (CX + IT Agent Platform)

Moveworks positions itself as the “enterprise agent layer,” integrating deeply with IT and collaboration systems. Its agents can recognize how to act—whether resetting a password, escalating a ticket in ServiceNow, or updating Jira issues—and then execute those actions without human intervention. The standout capability here is multi-tool orchestration: agents aren’t locked to one workflow, but can fluidly move between Slack, Teams, Salesforce, and more, depending on context.

Key Integrations: Slack, Microsoft Teams, ServiceNow, Jira, Confluence, Salesforce, Google Calendar.

Kore.ai (Conversational AI + Contact Center Automation)

Kore.ai proves that tool calling can be generalized across industries. Their agents are modular, with tools for CRM lookups, ERP updates, HR approvals, and ticket creation. For instance, a customer might ask about an order, and the bot can dynamically check SAP, create a Zendesk ticket, and schedule a callback via Outlook—all in the same conversation. This modularity makes it attractive to enterprises with sprawling system landscapes.

Key Integrations: Salesforce, HubSpot, ServiceNow, Zendesk, SAP, Microsoft 365, Slack

Talkdesk (AI-Powered Contact Center Platform)

Talkdesk has doubled down on AI-driven customer support, embedding tool calling directly into its automation and agent-assist offerings. Their AI agents can create and update Zendesk or ServiceNow tickets, pull customer details from Salesforce or HubSpot, and even trigger Slack notifications for escalations. The company’s approach emphasizes real-time orchestration, where AI assistants handle transactional actions during a live call or chat, reducing average handle time and improving resolution rates.

Key Integrations: Salesforce, Zendesk, ServiceNow, HubSpot, Slack, Microsoft Outlook/Calendar.

Read 👉 Best practices for tool calling

Data Ingestion for RAG

What is Data Ingestion?

Data ingestion powers Retrieval-Augmented Generation (RAG) by pulling data from third-party apps into databases like vector, graph, or even relational for real-time search. Examples include ingesting documents from Google Drive or syncing knowledge bases from Confluence.

Market Adoption

38% of companies support data ingestion, making it one of the fastest-growing integration use cases. Unsurprisingly, contact center/CX vendors (85–90%) lead adoption, since effective support depends on surfacing knowledge base content, tickets, and account data instantly during conversations.

Infra platforms (70%) are another hotbed of activity, as frameworks like LlamaIndex and Databricks position ingestion as a foundational capability for building custom AI pipelines. Products built on infrastructure platforms like LlamaIndex and Databricks want to ship AI features themselves, so infra platforms want to easily enable AI capabilities for their end users.

The relatively lower adoption in HR (15%) and Finance (20%) highlights that ingestion is still most critical in knowledge-heavy workflows rather than transactional systems. However, as AI expands into compliance, employee engagement, and spend optimization, ingestion will likely become a baseline requirement in those sectors too. In many ways, ingestion is the entry point to AI-enablement, turning static third-party data into dynamic knowledge accessible at prompt time.

Most Common Integrations

• Google Drive (~60%) – Still the dominant source of documents and enterprise knowledge.

• Microsoft SharePoint / OneDrive (~50%) – Essential for enterprise content and collaboration.

• Slack (~40%) – Popular for pulling unstructured chat data into vector databases.

• Confluence / Atlassian (~35%) – A top source for organizational knowledge and documentation.

• Zendesk & ServiceNow (~30–35%) – Fueling AI agents with ticketing and support knowledge.

• Databases (Snowflake, BigQuery, Postgres, MySQL) (~15%) – Increasingly important as infra providers and dev frameworks incorporate structured data.

Case Studies

Coveo (Enterprise Search Leader)

Coveo is a cloud-native, AI-powered relevance platform that provides enterprise search capabilities. Coveo’s entire business is built on ingestion. It provides dozens of connectors—Google Drive, Salesforce, Jira, databases—and handles complexities like incremental sync and permission-aware crawling. Its platform has become the blueprint for what enterprise ingestion should look like: robust, secure, and deeply integrated.

Key Integrations: Google Drive, Microsoft SharePoint, Salesforce, Zendesk, Jira, Box, Confluence, Snowflake.

Zendesk (Helpdesk + AI Deflection)

Zendesk leverages ingestion to unify content across help centers, CSV uploads, and even web crawlers. That ingested data powers its AI assistants, which can reduce ticket volume by surfacing the right answers before a human ever gets involved. This is ingestion directly tied to tangible business outcomes: lower support costs and faster customer resolution.

Key Integrations: Zendesk Help Center, Salesforce, Freshdesk, CSV/web crawler, Google Drive.

LlamaIndex (Developer Framework)

LlamaIndex makes ingestion approachable with LlamaHub loaders for popular sources like Slack, Notion, and databases. It handles chunking, embedding, and vector storage out of the box, giving developers the building blocks to stand up a production-grade RAG pipeline in hours instead of weeks.

Key Integrations: Google Drive, Slack, Notion, Confluence, GitHub/GitLab, Snowflake, BigQuery, websites, vector DBs (Pinecone, Weaviate)

Read 👉 Best practices for RAG ingestion

Workflow Builder Actions

What are Workflow Builder Actions?

Workflow builders are ways to visualize and apply deterministic behavior to AI actions. They allow users to customize automations with language and defined workflow steps that perform actions in 3rd-party platforms. Workflow builder actions are not tool calling, as user’s define the workflows, versus the agent deciding how to act.

Examples include sending Slack messages based on scraped web data , or sending calendar invites based on calendar availability.

Market Adoption

About 35% of companies have introduced workflow builders. CX vendors again lead with ~70% adoption, since workflows let enterprises customize how AI fits into their customer journeys with precision and compliance controls. Productivity (40%) and RevOps (35%) also show strong uptake, reflecting the need to balance automation with human oversight.

Lower adoption in data/infra (20%) and security/IT (25%) is notable. In these categories, workflows tend to be more developer-driven or embedded into existing ITSM frameworks (like ServiceNow or Jira). Users are also more “code fluent”, and prfer automations as code, rather than via workflow builders. Github Actions is a good example. Still, as enterprises demand more self-service automation, even these platforms are beginning to invest in low-code builders. The big picture: workflow builders represent the practical bridge between AI flexibility and enterprise governance.

Most common integrations

• Slack (~70%) – Ubiquitous as a notification and approval layer.

• Salesforce (~60%) – Core to sales and CX workflows.

• Microsoft Teams (~55%) – An alternative to Slack, particularly in enterprise IT environments.

• HubSpot (~40%) – Common in marketing and RevOps automations.

• Zendesk & ServiceNow (~35%) – Popular for automating support ticket flows and ITSM processes.

• Google / Outlook Calendar (~25–30%) – Enabling meeting scheduling and approval workflows.

• Workday & Okta (~20%) – Emerging as integration points for HR and identity workflows.

Case Studies

HubSpot (RevOps Automation)

HubSpot’s Workflow Builder is central to its product strategy. Users drag and drop triggers (e.g., form submission), conditions (e.g., lead score > 80), and actions (e.g., send email, notify Slack). With AI layered in—predictive lead scoring, smart enrichment—the workflows become both deterministic and intelligent.

Key Integrations: Slack, Gmail/Outlook, Zoom, Salesforce, LinkedIn Ads, Google Ads, custom webhooks.

Okta (Identity Lifecycle Management)

Okta Workflows shows how workflow builders can tackle high-stakes domains like identity and security. When a new hire is created in Workday, Okta Workflows can provision accounts across Slack, Jira, and ServiceNow, while AI monitors for anomalies or risky patterns. The result: compliance and security without manual overhead.

Key Integrations: Workday, Slack, ServiceNow, Jira, Box, Salesforce, Google Workspace, Microsoft 365.

Copy.ai (AI-Powered Sales & Marketing Automation)

Originally known for AI copywriting, Copy.ai has evolved into a full RevOps automation platform. Its workflows go beyond text generation: the platform can automatically draft outbound emails, log them in Salesforce or HubSpot, enrich contact records, and trigger Slack notifications — all in one flow. By pairing AI-driven personalization with deterministic workflow actions, Copy.ai helps sales teams scale outreach without losing context or accuracy.

Key Integrations: Salesforce, HubSpot, Slack, Gmail/Outlook, LinkedIn Sales Navigator, Apollo.io.

Read 👉 Best practices for workflow builders

Bidirectional Syncs

What is Bidirectional Sync?

Bidirectional sync goes beyond simple one-way integrations by ensuring that data remains consistent across multiple systems of record, no matter where updates originate. Instead of pushing data in one direction (e.g., exporting leads into Salesforce), bidirectional-syncs enable continuous two-way communication so changes made in one system automatically propagate to others.

For example:

• A new lead created in HubSpot is synced into Salesforce → any updates to that lead in Salesforce (like deal stage changes) flow back into HubSpot.

• A support case opened in Zendesk is mirrored in ServiceNow → resolution details logged in ServiceNow are pushed back into Zendesk so agents don’t lose context.

• An expense logged in Expensify automatically appears in NetSuite → reconciliation updates in NetSuite are reflected back in Expensify.

This doesn’t just apply to pairs of platforms — often, bi-sync spans entire workflows across 3rd-party platforms, connecting CRMs, ERPs, HR systems, ITSM platforms, and communication tools.

Market Adoption

Only 21% of vendors currently offer bi-sync, making it the least common of the four use cases. That said, adoption is concentrated in high-stakes categories: CX (70%) and RevOps (45%). In these domains, inconsistent data directly impacts customer satisfaction and revenue forecasting, making real-time sync a competitive differentiator.

Categories like HR (30%) and Finance (30%) are emerging areas, as platforms like Rippling and NetSuite demonstrate the value of keeping employee and financial records consistent across multiple systems.

The relatively low adoption in Productivity (10%) and Data/Infra (10%) categories doesn’t mean integrations aren’t important there — in fact, they’re critical. But these platforms typically rely more on contextual ingestion than on automated writes.

For example:

• Productivity tools like Figma or Notion require explicit user action to create or modify content (a designer builds a new mockup, a writer drafts a doc). AI integrations here generally pull data in to provide context, rather than pushing new artifacts out.

• Data/Infra platforms (like Databricks or Kubernetes) similarly depend on DevOps engineers or developers to create and update services. The system is not designed for AI to autonomously “push” new jobs or pipelines without user input.

This reliance on user-driven creation means there’s little need for the “push” side of bidirectional sync. Instead, these categories emphasize pull-based integrations (data ingestion), giving AI systems the context they need to operate while leaving authoritative record-keeping to user-initiated actions.

The relatively low adoption of bi-directional syncs (21%) doesn’t mean the feature is less valuable — it means it’s fit for narrower, more defined use cases.

Unlike tool calling or ingestion, which apply broadly across categories, bi-sync requires a 1:1 mapping between record types in different systems. For example:

• Salesforce ↔ HubSpot works well because both CRMs manage contacts, opportunities, and deals in similar structures.

• Salesforce ↔ Scratchpad works because Scratchpad is designed as a UI layer for Salesforce data.

• But for a product like you.com, what exactly would it sync with Salesforce? At best, it might push updates to a record — which looks more like tool calling or workflow automation, not a true sync.

This explains why bi-sync adoption is high in CRM, CX, and HR/IT (where records overlap) but much lower in categories like productivity or data infra. In those, user-created objects (docs, dashboards, pipelines) don’t have a natural equivalent in another system, making syncing irrelevant or risky.

The real takeaway: bi-directional sync isn’t inherently “better” than other integration approaches. It’s just the right tool for very defined use cases — where two systems share overlapping data models and businesses need that data to stay perfectly aligned.

• Use workflows when you want deterministic, multi-step processes that mix automation and human approval.

• Use tool calling when you want AI agents to autonomously act on records in another system.

• Use bi-sync when you need “copy-paste with memory” — records mirrored across systems, updated continuously without user intervention.

In other words: bi-sync is an extremely valuable automation pattern, but only when the problem space calls for it.

Most Common Integrations

• Salesforce (~80%) – The dominant anchor system for sales, CX, and finance syncs.

• HubSpot (~50%) – Paired with Salesforce or used independently in mid-market sync strategies.

• Zendesk (~45%) & ServiceNow (~40%) – Ensure cases and incidents remain aligned across support and IT platforms.

• Workday & Okta (~25%) – Critical in HR/IT for employee lifecycle and identity sync.

• Microsoft Dynamics 365 (~25%) – An enterprise alternative to Salesforce for CRM/ERP syncs.

• Slack & Calendars (~20%) – Niche but growing, especially for mirroring conversations and scheduling across environments.

Case Studies

Salesforce (CRM)

Salesforce sits at the center of most bi-sync strategies. Through APIs and MuleSoft connectors, it keeps data flowing between marketing, support, finance, and HR systems. Vendors like Outreach and Clari build directly on Salesforce’s APIs to guarantee their apps are in lockstep with CRM records. Salesforce exemplifies bi-sync as the system of record.

Key Integrations: HubSpot, Outreach, Salesloft, Zendesk, ServiceNow, NetSuite, Workday.

Rippling (HR Platform)

Rippling’s entire value proposition hinges on keeping employee data synchronized across HR, IT, and payroll. A title change in Workday cascades into Okta permissions, payroll adjustments, and Slack group updates. Rippling is a masterclass in bi-sync’s power to drive efficiency and compliance in employee lifecycle management.

Key Integrations: Workday, Okta, ADP, Gusto, Slack, Google Workspace, Microsoft 365.

NetSuite(Finance and ERP platform)

NetSuite is widely used as the system of record for financials, procurement, and ERP workflows. For many enterprises, keeping financial and operational data in sync across tools is critical for reporting accuracy and compliance. Bi-directional sync ensures that when an invoice or purchase order is created in NetSuite, it also appears in spend management platforms (like Coupa or Expensify) or CRM systems (like Salesforce).

Updates in those downstream systems then flow back into NetSuite to maintain consistency. By enabling these syncs, NetSuite reduces data silos between finance, sales, and operations teams, ensuring that everyone works from the same version of truth.

Key Integrations: Salesforce, Coupa, Expensify, QuickBooks, Workday, ServiceNow.

Read 👉 Best practices for bidirectional sync

Conclusion

AI companies are not just viewing integrations as add-ons but as core product pillars:

• Tool Calling enables agents to act.

• Data Ingestion fuels retrieval and context.

• Workflow Builders blend deterministic and intelligent actions.

• Bidirectional Syncs keep systems aligned in real time.

To be an AI SaaS leader, mastering these use cases is no longer optional—it’s table stakes.

That’s where Paragon comes in. As the all-in-one integration platform, Paragon provides the infrastructure to build these use cases fast, reliably, and at scale—so you can focus on delivering value to your users, not wrangling APIs.

Companies we analyzed to create this report:

Ada.cx

Airtable

Aisera

Antrhopic

Anyword

Calendly

Celonis

Clickup

Coda

Cognism

Cohere

Dialpad

Fireflies

Gainsight

Gong

Grok

Guru

HubSpot

Instabase

Intercom

Krisp

Llamaindex

Mistral

Moveworks

Nice

Notion

OpenAI

Pandadoc

Retool

Ringcentral

Salesforce

Seismic

Stackoverflow

Talkdesk

Vedrai

Vidmob

Writer

Yellow.ai

Zendesk

Zoom

Zendesk

Zoom

Asana

Monday.com

Smartsheet

Box

Dropbox

Lucid

Miro

Freshworks

6sense

Outreach

Salesloft

Clari

People.ai

Genesys

Liveperson

Gladly

Workday

Rippling

Gusto

Deel

Papaya Global

Coupa

Bill.com

Expensify

Okta

Wiz

Lacework

Databricks

Pinecone

Elastic

Scale AI

Parloa

Tommoro

Highspot

Thoughtspot

Logz.io

Kore.ai

Puzzel

Otter.ai

Callminer

Anyscale

Cresta

Polyai

Odin AI

aiXplain

Descript

Averi

Avenue

Atomicwork

zoovu

Zip

Awell

Vectra AI

Zenhub

Decagon

Azira

Suzy