[Product Manager] Application and challenges of AI in SaaS products

With the popularity of the ChatGPT large model around the world, AI is rapidly helping to improve the efficiency of various industries. In the SaaS field, AI also has great potential.

Insert image description here
AI (artificial intelligence, the abbreviation of Artificial Intelligence) has been at the forefront of public opinion for the past year. With the popularity of the ChatGPT large model around the world, AI has finally changed its previous "unintelligent" image and quickly begun to form practical solutions in various industries. , to help improve the efficiency of various industries.

In the field of SaaS, AI also has great potential. AI has advantages in many fields such as natural language processing, image recognition, and intelligent recommendation. In the process of product development, we continue to explore the use of AI technology to help improve product experience.

This chapter will introduce the capabilities of AI, the application of AI in SaaS products, and the challenges faced by AI in SaaS.

1. Three main capabilities of AI

The overall technical architecture of AI is relatively complex and involves multiple technical fields. Here we mainly introduce the capabilities that AI can achieve.

AI has made great progress and development in computer vision, natural language processing, speech recognition, machine learning, etc., which provides a solid foundation for many systems to integrate AI.

1. Computer Vision

Computer vision is the study of how to enable computers to understand and interpret images and videos. It involves the use of computer algorithms and techniques to simulate and implement the functions of the human visual system. The goal of computer vision is to enable computers to perceive, understand and analyze the content in images and videos to achieve automated visual tasks, such as image classification, target detection, face recognition, image recognition, image segmentation, etc. Through computer vision technology, we can equip computers with capabilities similar to human vision, thereby achieving more efficient, accurate and intelligent image and video processing in various fields.

Computer vision has a wide range of application scenarios, for example:

Image recognition and classification: Computers can automatically identify and classify objects, scenes, faces, text, etc. in images through image recognition technology, such as face recognition, object detection, text recognition, license plate recognition, etc.
Video surveillance and security: Computer vision can be applied to video surveillance systems to achieve real-time analysis and processing of video streams, such as pedestrian detection, abnormal behavior recognition, target tracking, etc., to improve security effects and reduce labor costs.
Medical image analysis: Computer vision is widely used in the medical field and can assist doctors in disease diagnosis and treatment, such as tumor detection, lesion segmentation, medical image reconstruction, etc.
Autonomous driving and intelligent transportation: Computer vision is one of the cores of autonomous driving technology. It can realize the safety and efficiency of autonomous driving and intelligent transportation systems through the perception and analysis of roads, traffic signs, vehicles, etc.
Industrial quality inspection and robot vision: Computer vision can be applied to quality inspection and robot visual navigation in industrial production, such as product defect detection, parts positioning, object grabbing, etc., to improve production efficiency and quality.

The above are only some of the scenarios of computer vision applications. With the continuous development and innovation of technology, computer vision will play an important role in more fields.

2. Natural language processing

Natural Language Processing (NLP) is a discipline that studies the interaction between human language and computers, aiming to enable computers to understand, process and generate natural language. It covers many aspects, for example:

Language Understanding: Converting natural language into machine-understandable form by analyzing text or speech. This includes techniques such as lexical analysis, syntactic analysis, and semantic analysis.
Language Generation: Convert machine-generated information into natural language based on machine understanding for interaction with humans. This includes technologies such as text generation and speech synthesis.
Information Retrieval: Rapid retrieval and extraction of specific information through indexing and searching of large amounts of text data. This includes keyword extraction, text classification, text clustering and other technologies.
Machine Translation: Convert text or speech in one natural language into text or speech in another natural language. This includes techniques such as rule-based translation, statistical machine translation, and neural machine translation.
Sentiment Analysis: Understand people's emotional tendencies toward specific topics or events by analyzing the emotions, attitudes, and sentiments of texts. This includes techniques such as sentiment classification, sentiment lexicon, and sentiment inference.

3. Voice recognition

Speech recognition is technology that converts human speech into text. Its processing process includes the following steps:

Audio collection: Collect the user's voice input through a microphone or other recording device.
Preprocessing: Preprocess the collected audio, including noise reduction, noise removal, etc., to improve the quality of the speech signal.
Feature extraction: Extract features from preprocessed audio. Commonly used features include Mel Spectral Coefficients (MFCC), etc.
Acoustic model: Use a trained acoustic model, such as a deep neural network (DNN) or a hidden Markov model (HMM), to match features to a probabilistic model for speech recognition.
Language model: Use language model to correct and optimize the recognition results to improve recognition accuracy. The language model can be a statistical n-gram model or a neural network-based language model.
Decoding and post-processing: According to the results of the acoustic model and language model, decoding and post-processing are performed to obtain the final text output.

The technologies involved in speech recognition include signal processing, machine learning and natural language processing. Among them, deep learning has been widely used in speech recognition, such as using convolutional neural network (CNN) or recurrent neural network (RNN) for feature extraction and modeling. At the same time, speech recognition also needs to combine language models and post-processing technology to improve the accuracy and fluency of recognition.

2. The main learning methods of AI

1. Machine Learning

Machine learning is a subfield of AI that enables computers to learn and improve from data without having to be explicitly programmed, through the use of algorithms and statistical models. The applications of machine learning are very wide. In addition to the image recognition, speech recognition, natural language processing, etc. mentioned earlier, it also includes recommendation systems, predictive analysis, automatic driving, etc.

There are many methods of machine learning, the common ones include supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning.

Supervised Learning: By given input and corresponding output label, train the model to learn the mapping relationship between input and output. During training, the model adjusts its parameters by comparing them to labels to make accurate predictions on unknown data.
Unsupervised Learning: In unsupervised learning, the model can only obtain input data and does not have corresponding output labels. The goal of the model is to discover hidden structures and patterns in data, such as clustering, dimensionality reduction, anomaly detection, etc.
Semi-supervised Learning: Semi-supervised learning is a learning method between supervised learning and unsupervised learning. It utilizes labeled data and unlabeled data for training to improve the performance and generalization ability of the model.
Reinforcement Learning: Reinforcement learning is a learning method that learns optimal behavior strategies through interaction with the environment. It learns optimal behavioral strategies through the interaction between the agent and the environment. In reinforcement learning, an agent observes the state of the environment, takes actions, and adjusts its behavior based on feedback (rewards or punishments) from the environment to maximize long-term accumulated rewards. Reinforcement learning has wide applications in many fields, such as robot control, game strategy, autonomous driving, etc.

Each method of machine learning has its applicable scenarios and algorithms. The choice of which approach depends on the nature of the problem and the data available.

2. Deep learning

Deep learning is an application of artificial neural networks and one of the branches of machine learning. It simulates the human nervous system by building a multi-layer neural network to achieve automatic classification and prediction of large amounts of data. Deep learning is a new research direction in the field of machine learning, mainly by learning the inherent laws and representations of sample data. levels, allowing machines to have human-like analytical learning capabilities. The ultimate goal of deep learning is to enable machines to recognize and interpret various data, such as text, images, and sounds, so as to achieve the goal of artificial intelligence.

Deep learning has achieved remarkable results in many fields, such as speech and image recognition, natural language processing, recommendation and personalization technology, etc. Deep learning has a wide range of applications, such as search engines, data mining, machine translation, multimedia learning, etc. It solves many complex pattern recognition problems by imitating human audio-visual and thinking behaviors, and brings many advances to the development of artificial intelligence technology.

3. 4 application scenarios of AI in SaaS products

Now, AI has penetrated into all walks of life from early conceptual products and has become an effective way to improve efficiency in various industries. For SaaS products, AI can provide assistance in many aspects such as SaaS product marketing, after-sales consulting, and product capability improvement. It can even reshape the functional experience of some products and greatly improve work efficiency.

1. Product capability improvement

AI has a high degree of recognition in speech and image processing, and can already demonstrate efficiency advantages in data input.

Our finance and taxation SaaS products need to process a large number of images. Traditional OCR recognition technology has always been inaccurate due to issues such as the printing clarity, angle, and template standardization of bills. Previously, the information on these pictures mainly relied on manual input or manual inspection, such as product items, amounts, bill numbers, and a series of other information. It takes tens of seconds to process a bill quickly, and a few minutes at the slowest time. When encountering some list bills, , it will take longer and manual processing is error-prone.

Using AI image processing technology, the recognition accuracy is greatly improved, and a batch of images can be uploaded and processed in batches. In the past, a large number of employees entered the work, and now it is automatically processed by the system, freeing up a lot of employees' energy.

Machine learning capabilities can help automate business rule processing.

For example, our invoice SaaS product has the concept of tax classification codes. There are only more than 4,000 tax classification codes, and there are millions or even tens of millions of various product categories. In order to improve invoicing efficiency, we generally Product items are automatically matched to tax classification codes (this is a tax policy requirement). Matching based on traditional rules is actually very difficult to exhaust. There will always be a variety of new products emerging, which cannot be fully automated and requires human participation.

Based on machine learning technology, we changed the matching rules from manual matching to AI matching. The processing process is roughly as follows:

Data collection and processing: In the first step, we process a large number of commodity item and tax classification coding data accumulated in the database. Clean and preprocess these data to remove duplicate data, handle missing values, etc.
Feature Engineering: Next, we transform the data coded for product categories and tax classifications into features that the machine learning model can understand. Use bag-of-words models, TF-IDF, word embedding and other technologies to process text data.
Model selection and training: With the feature data, we select a suitable machine learning model for training. We mainly focus on text, and tried to use several models such as naive Bayes, support vector machine, and deep learning model (such as recurrent neural network).
Model evaluation and adjustment: After each model training is completed, we use some evaluation indicators to check the performance of the model, and evaluate it based on accuracy, recall, F1 score, etc. When running a model for the first time, you often need to adjust the parameters of the model. If it is still not suitable, try using other types of models.
Model deployment and use: After many explorations, we selected the neural network model. This model has high accuracy and can meet the requirements. Next, we deploy it to the production environment, and then encapsulate the model capabilities into interface services for business systems. Call directly to realize the automated processing flow of the system.

Finally, we designed a system feedback mechanism. If users find that the matching is inaccurate, they can modify it. These modification information can be considered as the original data matching is inaccurate, and the data is saved for subsequent model improvement.

After several months of practical exploration, I deeply realized that AI has great advantages over traditional methods in handling such problems. First, the accuracy is better, and second, it supports automated processing of the system, making the user process smoother and improving the experience significantly.

This kind of rule matching with large amounts of data that is difficult to exhaustively is very suitable for AI processing, such as accounting account matching in financial and taxation products. You can pay more attention to some product pain points in your daily work, and it is very likely that AI will be appropriate to solve them.

2. SaaS Product Marketing

In terms of SaaS product marketing, AI can also provide some assistance. For example, personalized recommendations, product usage guides, quick help, etc.

Personalized recommendations: AI algorithms can analyze user behavior and preferences, provide users with personalized recommendations, and help companies provide customers with a more personalized experience.
Product Usage Guide: We have found that users’ understanding and usage proficiency of the product greatly affects subsequent experience and renewal, and our service team cannot accurately understand the problems currently faced by users. We learned through the questionnaire that generally new users are prone to face many problems in the first three months of using the product. In the first three months, each user faces different situations, which poses a huge challenge to the service team. The challenge is that comprehensive training is not accurate enough, and partial training makes it difficult to accurately understand customers’ problem points. I think the best service is to provide it when he needs it. If we want to provide good services, we need to analyze user operations, abnormal problems encountered, and matching solutions. We have already begun to do some work. AI is better than humans in terms of accuracy and timeliness of problem judgment. The response speed is much faster, hopefully providing better service to users. This is what we need to continue to work towards.
Provide quick help: Provide timely responses through the official website, APP, product and other portals. If manual communication is required, we can automatically transfer the pre-sales, customer success or after-sales teams for support in a timely manner based on the user's situation.

3. Intelligent customer service

Intelligent customer service has been used on a large scale in many industries and has relatively mature solutions. Although we sometimes still complain that intelligent customer service is not intelligent, with the accumulation of more data, the accumulation of knowledge base, and the application and understanding of AI The improvement of capabilities is gradually improving the quality of AI customer service. Excellent AI customer service has significantly improved the user experience and has a significant cost reduction effect on the enterprise. For the customer service team, they no longer need to deal with a large number of low-value problems and can focus on personal growth and team progress. Some excellent customer service personnel gradually Turn into an AI "trainer" and continuously provide high-quality knowledge content to the AI model.

With the development of AI infrastructure, the cost of training AI customer service robots is getting lower and lower. For example, we can use some large models to train enterprise-specific models based on the large models, which can avoid high technical investment and ensure data security. It is a relatively low-cost solution.

4. Data analysis

AI's ability to process data can help us apply it to the field of data analysis, such as data preprocessing, data exploration, predictive modeling, decision support, etc. AI can also be used for label processing of user portraits, etc.

Data preprocessing: Artificial intelligence can be used to automatically identify and deal with problems such as noise, missing values, and outliers in the data, improving data quality and accuracy.
Data exploration and visualization: Artificial intelligence technology can automatically analyze and explore large-scale and complex data sets, use unsupervised learning and other methods to perform cluster analysis, discover hidden patterns and trends, and generate interactive data visualization. Allows users to understand data more intuitively.
Predictive modeling: Through machine learning and deep learning technology, AI can build models based on historical data and predict future events or trends.
Support decision-making: By combining data analysis and machine learning technology, AI can provide decision-makers with real-time, data-based suggestions and decision support.

AI will have more and more application scenarios in SaaS products. As the SaaS team deepens its understanding of AI and advances in technology, AI will be deeply integrated with SaaS products in multiple directions to help product development.

4. Three challenges faced by AI in SaaS products

In the process of combining AI with SaaS products, you may face the following challenges, and we also try to give some methods to deal with them:

1. Data Privacy and Security

AI technology requires large amounts of data for training and improvement. This means that when combining AI with SaaS products, data security and privacy must be ensured. Our customers definitely don’t want their data to appear in public Q&A, which could hurt their competitiveness.

Ensuring data security is an important responsibility of SaaS companies. In the process of applying AI, we must adopt strict data security measures, such as data desensitization, data encryption, access control and security auditing. At the same time, strict data usage regulations need to be established to ensure that user data will not be leaked and abused. During the model training process, the enterprise-specific model training method based on large models mentioned above can be used to ensure that the data training process is completed within the enterprise to avoid the leakage of sensitive data.

2. Technology integration

When integrating AI technology with SaaS products, you may encounter technical challenges. The general AI technology stack is different from the technology stack of normal product iterations. In addition, some SaaS companies may not have enough understanding of AI, which will cause the use of AI to face dual obstacles in recognition and investment.

Before starting integration, you need to understand the functionality and architecture of AI technology and SaaS products in detail. Determine feasible integration solutions and develop detailed technical implementation plans. In addition, it may be necessary to hire a professional technical team or cooperate with an external AI company to complete AI integration tasks.

3. Data quality issues

The accuracy and reliability of AI algorithms largely depend on the quality of data. Even for SaaS companies, there are issues with data quality such as deviations, missing or incomplete data, which may affect the output of AI algorithms. In addition, data collection work may also become complex or inefficient, hindering the application of AI technology in SaaS products.

Before applying AI algorithms, we need to ensure the accuracy and reliability of the data. This requires steps such as data cleaning, data preprocessing, and data validation. In addition, some technical means may be needed to improve data quality, such as data mining, data analysis, and data visualization. In addition, in order to solve some problems of missing data, it is necessary to carry out product iteration or use some monitoring systems to complete the collection of original data. For some professional problems, manual data cleaning or data annotation may even be required.

5. The role of AI in the development of SaaS

In the process of product development, it is an important development direction to use AI capabilities to improve product efficiency, marketing efficiency, and after-sales efficiency.

Improve efficiency: AI technology can help SaaS companies automatically handle some heavy, repetitive and ineffective tasks, thereby improving the company's work efficiency.
Reduce costs: Through AI technology, SaaS companies can reduce the investment in human resources and reduce labor costs. At the same time, they can also improve the efficiency of resource utilization, thereby reducing operating costs.
Improve customer satisfaction: AI technology can automatically handle customer service, quickly respond to customer needs, improve the efficiency and quality of customer service, and thereby improve customer satisfaction.
Enhanced ease of use: Through natural language interaction with products, such as text or voice commands, AI technology makes SaaS products easier to access and use, improving user efficiency and productivity.
Improve data utilization efficiency: AI technology can help SaaS companies form applicable algorithms and models more easily, thereby improving data utilization efficiency and promoting enterprise development.

6. AI assists SaaS development

The future development direction of the combination of AI and SaaS may be in the following directions:

Provide personalized services: Through technologies such as AI and NLP, human voice patterns and voice control can be automatically processed to provide more personalized services.
Improve the intelligence level of SaaS products: In the future, the intelligence level of SaaS products will become higher and higher, and the automation and intelligence level of SaaS products will be improved through AI technology.
Spawn new SaaS formats: The combination of AI and SaaS may spawn some new SaaS formats, such as intelligent customer service based on AI technology, enterprise-level AI training platforms, etc.
Improve the security of SaaS products: Through AI technology, the security and reliability of SaaS products can be enhanced and user data and privacy can be effectively protected.
Optimize user experience: Through AI technology, the user experience of SaaS products can be optimized and user efficiency and productivity improved.

Like the Internet, AI will become the infrastructure of the entire society and be deeply integrated with all walks of life. The integration of AI and SaaS is only a matter of time. We hope that practitioners in the SaaS industry can recognize, understand and embrace AI as soon as possible.