We’re convinced the biggest beneficiaries of AI will be enterprise software companies. We’ve come to this conclusion in part from reviewing the earnings calls of 79 publicly traded SaaS companies. These include names like Snowflake, Salesforce, MongoDB, ServiceTitan, ServiceNow, Microsoft, AppFolio, Palantir, Atlassian, Paylocity, Doximity, Qualys, Bill.com, ZoomInfo, Dynatrace, Monday.com, Blackbaud, Cloudflare, Freshworks, Klaviyo, Datadog, Q2 Holdings, Shopify, HubSpot, Paycom, Unity, Twilio, Procore, JFrog, SPS Commerce, Waystar, SimilarWeb, Amplitude, Figma, Weave, RingCentral, Workiva, Five9, Alarm.com, Appian, HealthStream, Backblaze, Ziff Davis, Workday, Zeta, Digital Ocean, Kinaxis, The Trade Desk, Zoom Video, Nutanix, Alkami, Sprout Social, Definitive Healthcare, Sprinklr, Certara, Blend Labs, DoubleVerify, PubMatic, Asana, Gitlab, Box, Crowdstrike, Veeva, Okta, Wix.com, Zscaler, and Samsara.
What follows is a summary of what management teams are actually saying about AI’s impact on their businesses, their customers, and their competitive positioning.
We believe this is the most comprehensive analysis of enterprise software earnings calls published this cycle and what we’re learning cuts against the grain: enterprise software is going to be the biggest beneficiary of the move to agentic AI. Big take-aways are below:
AI is not profitable yet. The goal for the moment is driving margin neutral revenue. Microsoft, Salesforce, ServiceNow and nearly every other company described margin pressures from deploying AI product. AI workloads are just very expensive at the moment. They’re also complex. For instance Appian reported a significant jump in services revenue because these AI deployments at the enterprise level require deep integration work, data governance, and process design.
Direct revenue from AI is nascent for most, especially relative to total revenue. And is non-existent for some players like Doximity, or still in its very early stages but growing fast like at Freshworks. That said, enterprise customers are adopting and benefiting from the new AI products built by their already critical software providers like Datadog, Salesforce, and Monday.com. For instance, Braze AI Operator is actively used by two-thirds of customers after weeks of launch. There is also a category of companies that are building usage before monetizing, like Figma. No company we’ve observed is making AI product a significant part of any forecast, yet, even though they talk extensively about positive AI impact. The software companies are being very conservative in forecasting because the sales cycle is different, pricing is new, deal sizes are bigger, and implementation timelines are unknown. Both customers and software vendors are figuring out AI together.
The AI revenue coming out of legacy SaaS co’s is impressive. Some of the AI revenue looks nascent only because the existing SaaS businesses that are generating that AI revenue are huge. For instance, ARR at Workday is $8.8bln, of which $400mm is from agentic product. That’s only 4% of revenue from AI, but if Workday’s AI product was a standalone business, it would look large and incredibly fast growing. Similarly ServiceNow claims $600mm of agentic revenue, Salesforce claims $169mm with 800% YOYG, Docusign has $350 million in ARR growing 4.5x YOY representing 11% of total ARR, Ringcentral claims $100mm, and DigitalOcean is at $120mm. Again these are small numbers relative to total revenue at each of these companies, but it shows that existing SaaS companies may be the ones best positioned to scale AI products given they can sell into an existing customer base.
Monetization is changing. The pricing conversation for AI is largely going from “per seat” to “per outcome” or “per agent deployed.” Procore for instance wants to price its AI on construction dollar volumes, which would insulate it from employee count reductions. However, some like Atlassian are sticking hard to the per seat model. The shift to per outcome reflects the end customer’s skepticism of AI and the need to tie the investment in AI to outcomes. It also reflects the very high cost of AI workloads, and the uncertainty by the end customer and software vendor on how much AI workload they’re actually going to use. Finally, software companies are charged for their customers’ AI use on a workload basis, so it only makes sense that they would also charge by the workload or outcome, hopefully with some margin.
Infrastructure for AI versus AI products. Some like Snowflake, ServiceNow, and MongoDB, Twilio, Datadog, and Cloudlfare will win because AI is supported infrastructurally on their platforms. Some like Dynatrace believe AI makes their product more compelling than ever (in their case it’s observability and monitoring of AI).
Copilots are out. For instance ServiceTitan and Salesforce are building AI product which actually executes context-aware actions. Sentinel One now has an agentic security analyst that can investigate threats autonomously. AI as an assistant or Copilot has been de-emphasized by nearly every company we’ve reviewed. It’s now about AI which actually takes action.
Trust is an issue for AI. Doximity has stopped releasing AI product until they get it perfect, because it’s not ok for AI to mis-diagnose a patient. Qualys makes a similar point in cybersecurity; being the agentic remediation layer requires a level of trust that generic AI tools can’t establish. AI errors in healthcare and cybersecurity are near unacceptable, and thus the bar for accuracy is higher, which also means uplift from AI will be delayed.
Performance is strong. Quite a few of these companies like MongoDB, Cloudflare, and ServiceNow, and Datadog closed some of largest deals ever in Q4. Others like Atlassian had record quarters. Zeta beat guidance for an 18th consecutive quarter. Companies that are selling AI products to their customers report better growth and retention among those customer cohorts. On the other hand, there are companies like Zoominfo and Ziff Davis experiencing serious disruption, with growth falling to near zero.
The moat is very high. Moats for enterprise software exist around: proprietary non-public and sensitive historical customer data, workflows, governance, security, integrations, compliance, vendor trust especially in highly regulated industries like healthcare (Waystar and Healthstream cited this), and operational knowledge of the existing customer. AI cannot standalone, but rather needs to sit on top of software which manages all the above in an enterprise-friendly manner. Additionally, any of these software companies building agents have serios edge, because those agents are trained on enormous repositories of historical customer data that an AI startup will not have. For a customer to simply move their data to an upstart or competitor isn’t realistic: hyperscalers charge significant egress fees to move data and then there’s the risk of doing so.
SMB software is in trouble All that said, SaaS companies focused on SMB customers, which have much less internal complexity, could face a serious threat from AI that allows customers to do internal builds or from AI startups. Monday.com, Asana, EverCommerce and Zoominfo cited issues in their SMB customer bases. SMBs have simpler workflows, less institutional complexity, and less switching cost.
Sitting near the data is especially advantageous. A number of companies like Five9 and Appian report their customers have to solve data hygiene and silo issues before AI can be deployed effectively. This is a pre-requisite, so if you’re a software company sitting near or on top of the data already, you’ve got edge. Additionally SaaS companies with proprietary data sets like Zeta, Waystar, and SPS Commerce will see the value of those data sets increase as AI agents and tools scale into that data. Any software company with a hardware component also now becomes more valuable with AI. For instance, Samsara can now predict equipment failure more accurately for every customer because it already has the sensors in the field across tens of thousands of prior deployments, and can learn about predicting future failures at scale.
Internal AI improves overall margins. Many of the companies themselves such as Blackbaud and Klaviyo are using AI in their own operations to improve margins and productivity. For instance, Workday’s engineering output grew 22% in the last six months measured by code delivered. Figma built and shipped Sana (Workday’s AI layer) from project start to GA in three months. HubSpot had 97% of code commits using AI assistance. JFrog accelerated key API development 30x. The value of AI to SaaS is not just a revenue story but also a margin expansion and product velocity story. It’s one of the reasons we believe enterprise software companies will be the biggest beneficiaries of AI, since they can monetize new agentic product they release but also push coding/dev costs down.
More AI will increase the need for existing software. AI needs software to operate more efficiently. As the Appian CEO put it, “AI without workflows is chaos.” For companies that sit on the measurement, observability, and analytics like Amplitude, Datadog, Dynatrace, and Snowflake, AI-driven development increases demand for incumbent software. AI agents are also becoming a new customer or user. Crowdstrike’s security products will be needed for every AI agent operating with elevated permissions because they’re potential targets for adversarial manipulation. Okta has a similar dynamic; AI agents will need governance and permissioning overseen by a SaaS layer. The consumer of enterprise data platforms is no longer just a human analyst or developer, it’s an AI agent querying autonomously, continuously, at scale. That consumption will become monetization for software companies, and indeed is already happening at Sailpoint where non-human identities accounted for 25% of SailPoint’s SaaS identity growth in Q4. Even the foundational AI companies themselves are customers: Datadog has 14 of the top 20 AI-native companies as customers. Amplitude has 25 AI-native customers above $100K ARR with one frontier lab at seven figures. New markets are also opening up for software companies that can counteract AI. For instance, DoubleVerify’s ad products which ensure an ad isn’t being placed next to AI generated slop or is not being served to a bot is a new market for that company.
Code complexity will drive the need for software. Any software company operating in the monitoring and observability of code/tech stacks gets an extra benefit. AI-generated code ships faster, in higher volumes, and with less human review than human-written code which means more bugs, more security vulnerabilities, and more system complexity per unit of time than ever before. Datadog, Dynatrace, Amplitude, and Qualys all noted versions of this dynamic. The developer productivity gains from AI coding tools directly translate into more demand for the monitoring, testing, and security tools that catch what AI-generated code gets wrong.
AI is allowing software companies to really show their ROI. AI is allowing software companies new ways to show ROI. Examples include Waystar ($15B in prevented denials), ServiceTitan (18-point EBITDA margin improvement for Max customers), Klaviyo (50% higher open rates, 40% higher revenue per campaign), and HubSpot (2x meetings booked for Customer Agent users). They are using those outcomes to justify both higher prices and faster expansion within accounts.
Build vs Buy. Customers are concluding quickly they don’t want to build AI internally, especially those in regulated industries. RingCentral notes the engineering talent and customer compliance issues make building AI in-house unattractive; why not trust an existing software vendor? Waystar noted a similar dynamic. Customers are not building AI point solutions, rather they’re turning to their existing software vendors.
The legacy business is funding the excitement. Quite a few of the publics have a dynamic whereby their legacy business which is slow growing but generates significant free cash flow is funding an exciting but nascent AI platform or product. Dropbox, RingCentral, BOX, SPS Commerce all fit into this.
AI is probabilistic. Mission-critical systems of record that require 100% accuracy are not substitutable for probabilistic AI models. Software that is deterministic therefore becomes the governance layer through which AI operates. For any enterprise where security, consistency, and accuracy matters, software is the base on which AI sits. Agentic solutions cannot stand on their own. Veeva and Crowdstrike touched on this extensively, since the former operates in a regulated industry and the latter absolutely cannot allow hallucinations or guesses of any kind. Dynatrace also put it well: purely probabilistic AI, which are LLMs making inferences, produces unreliable outputs in production environments where precision matters. UI Path said something similar whereby software that brings together “deterministic automation, agentic AI, and enterprise-grade orchestration together on a single platform” will be the winner.
No one is standing still. Especially not those that will be impacted by new AI product. For example, Wix, which is a website builder has released its own AI products that makes website building even easier for the user and opens up their market to less savvy users. As such they’re expanding their market, not just waiting for an AI upstart to obliterate them.
Go ahead, try and build it. One thing that stands out about the earnings calls is how truly complex enterprise software is. The idea that AI is going to replicate these companies cheaply and quickly is laughable. Let’s use Alkami as an example: Alkami serves community banks and credit unions who sign 5–7 year contracts, and embed thousands of regulatory requirements into their digital banking deployments. Alkami integrates into over 450 different financial technology systems, few of which have publicly available integration specifications and all of which carry potential legal liability. Customers use Alkami as a system of record to codify fraud mitigation practices, money movement thresholds, customer due diligence, enhanced due diligence, and decision logic including approvals, outcomes, and supporting reasons. Implementation takes 9–12 months. So good luck replicating them quickly with AI. Also note Alkami has put out AI product which is getting very strong adoption among new logos especially. Rubrik’s CEO also slammed the question about AI replicating their suite: “Rubrik is a very large and complex piece of software, and it is an enterprise-scale code with about 12 years of soaking time with thousands of customer feedback and customer use cases in a large enterprise environment that has been built into this. It is not something that you can write code on or an LLM can solve. We are the system of record of last resort around data and identity when a large bank or large hospital faces a ransomware attack.”
Revenue multiples in real time can be seen for all these companies at https://www.softwaremultiples.com/. Also visit https://www.blossomstreetventures.com/ for detailed financials and metrics data for all these companies. If you would like summaries of the actual earnings calls so you can run your own analysis, please email me.
Thank you for your readership. Email the author directly at sammy@blossomstreetventures.com