SaaS × Large Models? Embrace Them, But Don't Rush | Yunqi Capital · Riding the AGI+ Wave

To be updated)

Imagination knows no bounds; productivity is still being honed. Cover image: Yunqi Capital × Midjourney

The change has already happened. The powerful semantic understanding, information extraction, and text-to-image generation capabilities of large language models are helping SaaS companies further iterate their product architecture and business models.

At the AGI+ Wave series of salons, we discussed with over 200 entrepreneurs, industry experts, frontline programmers, and product managers in Beijing, Shanghai, and Hangzhou how large models are reshaping the industry. Multiple Yunqi Capital portfolio companies shared how they're using new approaches to efficiently solve old problems — and how to define the right problems.

Chen Yu, Partner

"In future applications, roughly 80% will be 'upgraded versions' of existing apps integrated with large models, while 20% will be entirely new applications built on the novel capabilities of large models. We encourage and support our portfolio companies to continuously upgrade and iterate by leveraging large model capabilities, and we look forward to the new opportunities that large models will give rise to."

Han Yi, Executive Director

Success for SaaS companies hinges on three critical factors: first, delivering standardized tools or products to ensure customers receive quality service experiences and sustained value, thereby driving renewals; second, building robust operational and business systems, including strong sales, operations, and service teams, plus efficient internal collaboration; third, deep and thorough penetration of specific scenarios, which enables rapid initial traction and subsequent service expansion.

Move fast, but stay steady. At the new starting line that large models have drawn, who will maintain the lead? We've compiled standout perspectives from this salon series along with recent deep thinking from Yunqi Capital investors, hoping to spark fresh insights.

---

01 Optimizing Product Strength

Safer, Smarter, More Comprehensive

Before integrating large models, re-examining application compliance and data security is "Step 0" for beginning iteration. In April, the Cyberspace Administration of China released draft measures for managing generative AI services for public comment, signaling that regulatory direction will likely tighten. Prioritizing compliance infrastructure early and designing applications with dynamic adjustment capabilities are prerequisites for ensuring products can ascend steadily and sustainably.

Behind the generation and mutual transformation of audio, text, images, and video lies the re-structuring of customer business logic. Content production will become more intelligent with AI's assistance, but to bring AI closer to producing quality content, beyond manual annotation and expert scoring, there's a clever path: obtaining user feedback through interaction and aligning with user experience to optimize the aesthetic direction of content.

General-purpose large models and vertical models are reinforcing each other and ascending in tandem, with growing product integration as a key trend. Large models are breaking down certain "skill barriers," smoothly connecting disparate business scenarios within enterprises. As original traffic chains "lengthen," ToB SaaS companies need to reconceptualize their service delivery scenarios amid these changes.

02 Making Data "Move"

Faster Speed, More Feedback, Better Iteration

At this stage, data quality is a decisive factor in model performance, which is why we're actively encouraging more SaaS companies to integrate large models as soon as possible.

On one hand, getting users onto products faster means obtaining data feedback faster, finding optimization directions faster, and thus pushing product iteration faster — establishing a "first-mover advantage." On the other hand, continuously exploring the rhythm of introducing private-domain data through practice can drive the co-evolution of both product and team.

03 Maintaining "Flexibility"

Staying Vigilant About New Models

New models and algorithms emerge constantly — how to maintain application flexibility and extensibility? This requires building systematic model evaluation capabilities: small-scale testing first, then after parallel comparison across multiple new and existing models, selecting the optimal model and deployment approach.

Additionally, large models currently still have low "controllability," so establishing contingency mechanisms for unexpected issues is also essential.

04 Team "Growth Capacity"

Building Trust, Continuous Learning

After the arrival of large models, technical barriers have been lowered. Continuously strengthening customer trust and securing user repurchases are core capabilities that SaaS companies serving enterprise customers need to keep prioritizing.

Moreover, encouraging open-minded thinking within teams and establishing efficient human-machine collaboration mechanisms are equally important — these will become "hidden critical factors" for strong enterprise growth capacity.

05 Cultivating Both Internal and External Capabilities

Also Defining "Boundaries"

In discussions, multiple entrepreneurs mentioned that their companies are also fully integrating GPT internally, actively exploring applications in daily operations, data analysis, and knowledge system construction — improving efficiency while discovering new usage scenarios.

Another shared consensus: "A SaaS company's most important goal remains helping customers create value; AI is merely a powerful tool." Therefore, companies must also pay attention to resources invested in AI iteration, always keeping customer needs first and methodically migrating product capabilities step by step.

Below are selected excerpts from our conversations — more delightful details from AGI+, enjoy～

In what ways have large models already helped companies and customers reduce costs and improve efficiency?

Online 3D Cloud Design Platform

Huang Xiaohuang, Co-founder/Chairman, Coohom

AIGC-related applications have served as new traffic entry points, bringing us over 40% conversion rates. Our AIGC Lab, established in April, focuses on researching "spatial AIGC" scenario applications, including AI design generation and iterative creation for home furnishing, commercial spaces, real estate and architecture, and other full-space domains. Many users have already used this new tool to generate design plans for their new homes. Additionally, we've comprehensively introduced AI across data analysis, sales, and marketing processes, with corresponding adjustments to our talent structure — efficiency has improved substantially.

Integrated Intelligent Contract Management Platform

Wu Dan, Founding Partner/COO, Zhenling Technology

Large models precisely identified insufficiently rigorous expressions during contract review and made improvements. By integrating large language models (LLM), natural language processing (NLP), and knowledge graph technologies into review scenarios, we've transformed contract review from a "subjective question" into a "true/false question," significantly accelerating work speed and improving quality. Meanwhile, our marketing department and automation testing teams have both begun using GPT in daily work — in writing articles, generating test scripts, and coding scenarios, GPT has helped us improve efficiency and accuracy.

Innovative Design-Production-Packaging Service Platform

Chen Yan, Founder/CEO, Xiangzhihe (Elephant Intelligence)

AI has helped us achieve efficient, agile, flexible delivery from design to production. We've already released an industry-level solution combining AI intelligent design with nano digital printing packaging, and will continue exploring in this direction. In the future, creativity or new ideas may become commercial entry points. In an era of content explosion, AI will be an important productivity tool, working with us to truly realize a "what you see is what you get" flexible production model.

Enterprise WeChat SCRM System Platform

Qin Pengfei, Co-founder, Weiling Technology

With audio, records, and text generation fully interconnected, customers' efficiency and precision in marketing processes have significantly improved. Our micro-GPT, launched by integrating ChatGPT capabilities, can automatically generate conversations, texts, and articles. Going forward, we'll continue iterating rapidly deployable functions for frontline marketing and sales teams.

B2B Intelligent Marketing Platform

Feng Shicong, Founder & CEO, Bailing Intelligence

Customers always care more about results than methods. We launched the Aideas intelligent marketing app marketplace, aggregating full-network online data and proprietary offline data based on NLP, image recognition, recommendation & prediction algorithm models. Starting from enterprise marketing and customer acquisition scenarios, we extract and refine real-time data insights through algorithm models, further deconstructing and integrating our B2B intelligent marketing capabilities into modular marketing applications.

At the salon event, multiple other entrepreneurs also shared their experiences using large models internally and their directions for product optimization after integrating large models.

The commercialization of general-purpose large models has only just begun, with more unknowns awaiting answers from practitioners in practice. As Alan Turing said: "We can only see a short distance ahead, but we can see plenty there that needs to be done."

Our AGI+ Wave series salons will continue — we look forward to deeper exchanges with more wave-riders, jointly embracing the new AGI+ era through the collision of ideas.