"AI Large Models + E-Signature": The Future Direction of E-Signature Development
-
The data analysis and training capabilities brought by large models will amplify the data advantages of certain vendors, enabling them to differentiate themselves and break the homogeneous competition in the e-signature sector.
Since the explosion of AI large models, there have been numerous participants. In the field of e-signatures, this technology is also creating new possibilities. However, similar to the "difficulty in implementation" faced by large models in various scenarios, applying AI large models in the e-signature field is no easy task.
In fact, the e-signature sector is even more typical. From a process perspective, the steps in the e-signature process—such as creating signing templates, generating seals and signatures, initiating signatures, and executing signatures—are relatively straightforward, making it difficult to foster an environment conducive to large model implementation. Additionally, high-quality data is key to the decision-making capabilities of vertical models, and data collection, training, and learning require manual involvement and time accumulation. These are not only technical challenges but also issues of boundaries and compliance in the e-signature sector.
Even though some e-signature SaaS platforms have extended their services to pre- and post-signing scenarios, these areas are also battlegrounds for upstream and downstream partners in the e-signature industry. The "difficulty in implementation" of large models has become a consensus.
Under the impetus of large models, what might the future direction of e-signatures look like?
"To implement large models in e-signatures, we need to focus on pre- and post-signing scenarios," Liu Qian, General Manager of the Product Center at Fadada, told Chanyejia directly.
In fact, once a finalized document is turned into a template, initiated for signing, and completed with subsequent notarization, its content cannot be altered during this process. The product's capability is limited to recording whether the document has been tampered with and verifying the authenticity of identities and intentions. This makes it difficult to combine with the generative capabilities of large models and fully leverage their true value.
For AI large models, their value lies in collecting data, conducting training and learning, and providing intelligent decision-making to help businesses reduce costs, improve efficiency, and mitigate risks.
In the field of electronic signatures, integrating industry-specific content to generate and analyze materials based on needs is indeed a scenario where AI large models can add value. However, for pure electronic signing, compared to content generation and analysis, trustworthiness is the foremost priority.
In other words, for AI large models, simply combining with electronic signature products does not bring significant incremental value to customers.
A deeper consideration arising from Liu Qian's statement is that behind the current difficulty of implementing AI large models in the electronic signature field lies the need for electronic signature SaaS to find genuine large model application scenarios beyond just signing contexts.
In reality, this 'exploration' has been ongoing over the past few years. With the development of e-signature SaaS, the demand for online signing has gradually extended beyond just the signing scenario, expanding both forward and backward. Unlike the international trend of focusing on niche SaaS solutions, Chinese customers prefer a one-stop software solution that addresses all their needs.
Leading players in the e-signature SaaS field, such as eSign, Fadada, Qiyuesuo, and Shangshangqian, have been continuously expanding their product and service boundaries. For example, eSign's intelligent contract product leverages AI technology to provide enterprises with a full lifecycle service covering contract drafting, approval, signing, execution, archiving, and statistics. Fadada's iTerms intelligent contract review offers capabilities like contract review, collaborative review, text comparison, and intelligent archiving.
"The implementation of large models will enable everyone to utilize intelligent contracts," said Jin Hongzhou, CEO of eSign.
In the AI 1.0 era, artificial intelligence primarily relied on supervised learning, where models were trained on known input and output data samples to predict or classify unknown data to achieve a desired outcome. Large enterprises, with their standardized contract content and management, benefited more from supervised training, while small and medium-sized enterprises faced limitations in the deep application of intelligent contracts.
However, large models may change this landscape.
It is foreseeable that in these scenarios, AI large models can provide powerful decision-making support. So, what should e-signature vendors do?
From the current perspective, there are essentially two paths to implementing large models in the e-signature field: first, vendors building their own AI large models, and second, collaborating with general-purpose large models.
The former requires substantial funding, data, and AI technical support, making the latter a more feasible approach for e-signature vendors.
However, relying solely on general-purpose large models presents greater limitations for the e-signature sector compared to other domains. "Using general-purpose large models for niche applications will inevitably yield mediocre results," said Liu Qian. In his view, general-purpose models lack the capability to fully support e-signature and smart contract services and must be combined with local knowledge bases.
In other words, similar to other fields, vendors need to integrate vast amounts of contract data into general-purpose models to develop specialized models for the e-signature industry.
But not all vendors are capable of achieving this.
First, collecting customer contract data into general-purpose models risks exposing sensitive client information.
As is well known, data is highly sensitive in the field of e-signatures. Most e-contract service providers offer public cloud SaaS-based e-contract products, where data is stored in cloud data centers, and users' e-contract signing and data are hosted on public servers.
Although platforms provide various authentication and verification methods to ensure data security and prevent tampering with contract data, users with high requirements for information security and contract data sensitivity still view data security risks as their primary concern.
Therefore, an e-signature proprietary model must be built on a private cloud to ensure the security of contract data.
This is also crucial for the selection of general-purpose large models. Liu Qian told Industrialists that Fadada is currently collaborating with multiple general-purpose large model providers, integrating the strengths of these models with product application scenarios. This allows the aggregated contract data to maximize its value while ensuring security.
Additionally, to improve the accuracy of intelligent decisions made by e-signature proprietary models, vendors need to rely on manually annotated high-quality data for training and learning. Data annotation for e-signatures not only requires technical capabilities but also expertise in legal knowledge, contract standards, and industry experience.
More importantly, whether e-signature vendors possess high-quality contract data is somewhat a 'false proposition'. Compared to traditional e-signature vendors, most e-signature SaaS providers started relatively late, resulting in weaker capabilities in high-quality data integration.
From this perspective, for e-signatures, the difficulty in implementing AI large models lies not only in the scenarios but also, more critically, in the data. Compared to other fields, the data barrier is even higher.
However, it is undeniable that under the "AI large model + e-signature" model, some fundamental changes may occur.
Specifically, in the entire lifecycle of contract signing, apart from the security and compliance issues during the signing process, contract drafting and review are the most critical stages for enterprises. As customer signing demands increasingly extend to pre- and post-signing scenarios, these requirements present new challenges for vendors.
In the past, most e-signature vendors, with the support of AI, have partially addressed issues like intelligent contract drafting and error correction through smart contract products.
However, there remains a gap between the quality of contract drafting and error correction and the ideal state. This gap is partly due to limitations in data quality, quantity, and computational power.
Under the 'AI Large Models + E-Signature' model, leveraging the capabilities of foundational general-purpose large models, combined with sufficient computing power and data volume, along with high-quality contract data from e-signature providers, enables more accurate intelligent decision-making in contract drafting and review processes. This helps enterprises shorten contract signing cycles, reduce error rates in contract texts, and truly realize the potential of 'smart contracts.'
Secondly, the 'AI Large Models + E-Signature' model also brings changes to delivery models. Due to the high customization demands of large domestic enterprises—such as significant differences in signing requirements across different businesses within the same industry—the SaaS delivery model for e-signatures in China is generally resource-intensive, requiring substantial human, financial, and effort investments from service providers.
By empowering large models to automate certain stages of the contract signing lifecycle, this pressure can be significantly alleviated, accelerating the industry's move toward standardization. For example, under the 'Large Models + E-Signature' model, most small and medium-sized enterprises (SMEs) can achieve self-service capabilities.
"We have integrated various large models into our services," as seen in eSign's solutions, where large model capabilities have become the foundational layer of its ePaaS platform.
From a broader perspective, the domestic SaaS sector often faces intense competition and homogenization due to market fragmentation, and the e-signature industry is no exception. With the support of large models, e-signature providers with extensive experience and data accumulation in specific verticals will see significant improvements in service capabilities.
In other words, the data analysis and training capabilities brought by large models will amplify the data advantages of certain providers, enabling them to differentiate themselves and break the cycle of homogeneous competition in the e-signature sector.
Whoever accumulates deeper may take the lead first.
More importantly, large models may become a foundational capability. On top of these models, electronic signature providers can leverage their data and computing power to create integrated, full-stack services, driving electronic signature products toward standardization and scalability.