How can B-end products within enterprises effectively implement AI? Based on practical experience, the author analyzes the problems in detail and shares a set of practical methodologies to help the team achieve cognitive alignment and maximize business value in the process of AI implementation.
Before the text begins, here is a stack of armor.
1. I just throw out some methodological summaries in the practice process, and will not involve specific technical implementation and AI direction issues, which are not within the scope of discussion in this article.
2. My point of view may be wrong, because I just throw out some of the methodologies I summarized from my perspective, which may not necessarily apply to other scenarios, and you are welcome to discuss it rationally.
Various problems of AI implementation in internal B-end scenarios
Presumably, everyone who makes internal B-end products, after the outbreak of AI in the past two years, has encountered a problem: “How to integrate AI into our business?” ”
Everyone expects to use “powerful” AI to achieve the goals of “growth, profitability, cost reduction, and efficiency increase”, but in practice, it may be difficult.
I have divided the problems I will encounter here into two categories:
1. Human problems
This means that in the process of AI implementation, various stakeholders associated with us will throw us various problems, which will hinder our business progress. It can be divided into problems of the team itself, the business side, and the company’s top management.
First of all, there may be problems such as “positioning” and “path” at the team’s own level.
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
For example, people are often confused about what AI can do for us. What should be done? What can it be done?
If we can’t think about these problems clearly, we can’t start with the implementation of AI.
Secondly, there will be problems such as “AI direction”, “resource allocation”, and “cognitive alignment” from the business side.
For example, there are too many directions in the field of AI, and the technical routes are also very diverse. Sometimes, the team will be a little confused, not knowing which direction to exert force, not knowing which direction is the most suitable for our business and the most effective.
For example, the business will not cooperate with us for various reasons, such as “thinking that AI will replace us and not actively supporting it”, “imaginative ideas, maybe they are often brainwashed by various “industry-disrupting” short videos. As soon as you open your mouth, you will train a large industry model, and tomorrow you will upgrade the whole link of the business.”
If you can’t think about these problems clearly, it is difficult to implement them, and it is easy to go the wrong way.
Finally, you will also feel the problem of “cognitive alignment” from the “manager”.
Some managers may have too high expectations for AI, thinking every day about using AI to replace a certain team, kill a certain process, and pursue disruptive effects. This may also be related to the fact that many self-media advocate industry subversion every day, and many high-level executives are full of anxiety.
They may not clearly understand where the boundaries of AI’s current capabilities are, and there may also be deviations in their perception of the actual capabilities of the team and the complexity of project implementation.
If this piece is not handled properly, we may not be able to obtain resources to support the project. We may even be suspected of having problems with our abilities, which will affect our career and even lose our jobs. As a result, the vision of “AI truly landing in business” is impossible to talk about.
2. Problems with things
After aligning people’s cognition, it is a question of how to do things. In this process, you may encounter the following problems:
What should be the steps for AI to be implemented? Do you build the system first, or find a way to make the business work first?
When designing the system framework, how to design it can be combined with the existing business system without causing cost waste.
The lack of methodological guidance may make the implementation of AI difficult, resulting in the miscarriage of projects due to many reasons such as realization costs, human resources, and value presentation.
Preliminary work, cognitive alignment!
Next, let’s talk about my ideas for solving these two types of problems.
First of all, let’s talk about the “human problem”, which is actually a problem at the “cognitive” level. In order to solve these problems, it is necessary to carry out cognitive alignment in three directions: self, business, and upward, and also correspond to problems from the three directions. Self-awareness: Clarify the positioning and boundaries of your team
The first is self-awareness, and the first step here is to “clarify the positioning and boundaries of your team”.
We must first understand what is the positioning of our current team at the AI application level? I can roughly divide them into 4 types of teams and their level of AI exploration:
The first category is teams with no relevant development capabilities. It means that we do not have the relevant technical human resources to build AI-related functions. In this case, you can only focus on deploying mature open source projects, tools, or purchasing mature AI services from third parties. For example, MJ and SD are used directly in the image generation scene, and Keling and Jimeng are used in the video generation scene.
The second category is the team with basic development capabilities. It refers to the ability to deploy technical human resources only for system development, but there are no AI-related technical personnel. In this case, only AI capability-related APIs can be integrated with business systems, or internal-oriented AI systems can be developed separately. In addition, RAG technology can also be applied to carry out simple large model tuning to meet the customized demands of enterprises.
The third category is teams with small model development and fine-tuning capabilities. It refers to the basic AI technical talents in the team. In this case, you can try model fine-tuning and customized small model development for specific scenarios to meet business demands in specific scenarios.
The fourth category is teams with the ability to develop large models on their own. Teams with this ability are basically at the level of large factories and can explore deeper innovative applications.
The first and second categories focus on “integration”, that is, “by researching AI capabilities that are useful to business on the market”, and integrating business processes into the business to provide the efficiency improvement effect of “business + AI”. The third and fourth categories focus on “innovation and exploration”, which requires model development in combination with business demands and AI capabilities.
We can benchmark the current configuration of our own team, clarify which category we belong to, and understand how far we can achieve based on AI.
The key here is to clearly understand the team’s current real ability level and not blindly pursue lofty.
【Self-awareness】There are multiple stages of AI landing on the B-side
Based on our understanding of our own team positioning and boundaries, we can also roughly understand the multiple stages of AI landing on the B-side, so as to realize how far we can go.
Stage 1 is single access empowerment
It refers to the use of AI as an independent tool to solve specific, isolated business pain points. For example, if the business needs to generate an image and generate a video, we will access the relevant AI production capabilities to achieve single-point empowerment. In addition, such as intelligent document extraction, basic Q&A robots, automatic generation of meeting minutes, etc., can also be regarded as single-point capability empowerment.
This stage is characterized by ease of use, quick results, and narrow scope. The cost achieved is low, which can be met by deploying mature open source projects, tools, or purchasing mature AI services from third parties. It is the stage where the “first type of team without relevant development capabilities” mentioned above can be achieved.
The second stage is workflow integration and empowerment
AI capabilities are no longer isolated, but are embedded in existing core business processes or systems, providing automated or intelligent solutions for the entire link. Suppose we originally created an image production capacity, then this system is the inspiration library and material library of related businesses, automatically produce image materials, and then upload them to our delivery system. Thus, the closed loop of “demand-make-use” business process is realized.
At this stage, the “efficiency improvement” and “quality improvement” of AI can be maximized by fitting the business process. This stage requires the support of a “team with basic development capabilities” to develop the system, or use the company’s data assets for RAG tuning, so as to meet the customized demands of the enterprise.
The third stage is the intelligent scheduling stage of workflow
Just like the recently popular tools such as Manus and Lovart, they can input a sentence of demand through the user, let the large model understand the user’s demands, and plan the order of task execution, while scheduling the required AI tools, and finally complete the execution of the task.
This reflects a trend in AI tools today. There are more and more AI tools, and in the face of complex and numerous tools, people will be stuck in the dilemma of choice. And in some scenarios, the combination of multiple tools will be involved, which will undoubtedly further increase the cost of choice.
Therefore, the current trend of AI tools is a trend from “people looking for AI” to “AI looking for people”. Through AI, it understands the user’s “needs”, intelligently matches the corresponding AI tools, and completes task planning and execution.
At present, most of the solutions at this stage are general-purpose solutions, and if the internal customized scenario is implemented, it requires a certain “team with small model development and fine-tuning capabilities”.
Stage 4 is the digital employee stage
But this stage belongs to the “outlook for the future”. It means that the skills of employees are disassembled into AI workflows, and then by perceiving the business environment, then executing the plan, and finally taking actions to meet business demands.
But at present, this is just a prospect, so I won’t talk too much.
By separating these four stages, we can roughly correspond to the four types of teams mentioned above, and currently, stages 1, 2, 3, and 4 correspond to the upper limit of team types 1, 2, 3, and 4 to a certain extent.
But this is also the correspondence of the current stage, and with the “SaaS” tooling of the 2nd, 3rd, and 4th stage capabilities, teams with only the first type of capabilities can also apply the latest AI technology. Nowadays, there are more and more workflow tools, and there is already a certain trend.
Self-awareness: Barriers to the internal B-end
Although at this stage, most manufacturers are general scenario solutions rather than meeting the customized needs of enterprises. Therefore, for our B-end enterprises in a certain vertical industry, the efficiency of these AI capabilities is limited, and there is a “last mile problem” here.
At this stage, we can combine the situation of general solutions with our vertical industry needs to build relevant AI tools to improve the accuracy of AI in specific business, provide customized solutions for the business, and meet business demands most efficiently, so as to assist in solving the “last mile” problem of external AI capabilities.
However, with the development of external manufacturers, with the deep cultivation of their capabilities, scenarios, and industries, it is possible that one day a certain manufacturer will launch tools that “can meet our business needs”, “easy to use” and “cheap”, then the efforts of our internal B-end product team will be in vain.
Therefore, when we explore the direction of AI, we must also consider the feasibility of building barriers.
Personally, I think that judging whether barriers can be formed can have the following aspects:
1. Does it have enough customization advantages?
Third-party manufacturers (SaaS) are required by many companies, and they are standardized products, and there must be efficiency loss in specific scenarios, which is not easy to use compared with customized tools. And there are some special business scenarios that third-party manufacturers may not be able to empower. Even if third-party manufacturers launch low-code platforms, there is still a huge threshold for entry in a short period of time.
Therefore, we need to judge whether our company’s process has enough customized scenarios to distinguish it from the “empowerment” provided by third-party vendors to ensure that the use of internal tools can have a high enough efficiency improvement.
At the same time, we need to ensure that there are enough customized scenarios here and whether they will be covered by third-party manufacturers in a short period of time to ensure that we can be differentiated from third-party manufacturers, otherwise the boss has no reason to pay for the “internal B-end product team”.
2. Can you accumulate industry best practices?
Third-party (SaaS) is the industry average best practice, and in-house systems are the best practice for the enterprise. If our internal enterprises can form industry-leading best practices, then we do not need the empowerment of third-party vendors, but can explore and precipitate by ourselves, so as to achieve our own internal empowerment.
3. Can data risks and security issues be avoided?
This is a point of concern for leading companies and companies in highly competitive industries. Using an in-house system can avoid these risks to some extent. Therefore, we need to judge whether our industry can avoid this risk through internal systems.
If it is not a leading company, or an industry with low competition, the risk in this area is relatively small, then we may not need to develop our own internal system to avoid risks.
4. Is it possible to provide a cheaper solution?
Some SaaS offers high pricing that will package and sell some unnecessary additional services, resulting in an overall premium. Therefore, if the internal B-end product can provide just the right functions based on the actual needs of the business, and control the overall development cost. Then we also have certain advantages over SaaS.
The above four points are things we need to consider when promoting the implementation of internal AI and gradually do it. Otherwise, one day the external market will be disrupted, and our efforts may be in vain.
Business awareness: Scan business opportunity points
The above three points of self-awareness help us sort out the “question of what we can do”, and then we will explore “what we want to do”. It is necessary to find the direction of the business, and at this time, it is necessary to first conduct business cognition and scan the opportunity points of the business.
My idea is this:
First of all, we need to divide the field and take stock of the general business type. Let’s assume that the company will involve two fields: design and customer service.
Then we disassemble the business processes in these two areas and sort out the core links, such as:
The core process of design is “finding inspiration”, “making drafts”, “collecting materials”, “finishing drafts”, and “going live”.
The core process of customer service is “customer service training”, “understanding problems”, “solving problems”, “follow-up”, “feedback collection”, “recording reports”, “customer service management” and other links.
Next, we need to analyze the entry point of AI through these business processes, focusing on finding the pain points, such as repetition, inefficiency, error-prone, information overload, etc.
This process is very good for our understanding of business, so we can explore the entry point through “business research”, “business interview”, “business rotation” and so on.
We can then assess whether these entry points can be addressed or optimized by current (team capability-based) AI technologies and analyze the feasibility of them.
Suppose we find the “industry intelligence collection” and “inspiration extraction” direction from the direction of “finding inspiration” in the design field, and we can analyze the AI technology required in it.
For example, industry intelligence collection requires information collection, that is, crawler technology, the basic core function here does not involve AI, AI will only play a role in “improving the accuracy of collection”, so the overall difficulty of this direction is “simple”.
For example, inspiration extraction requires the analysis of multimodal materials to extract inspiration for material design. At present, content extraction is relatively simple, but it is difficult to analyze the information required by the industry, so the difficulty of this direction is medium.
We can try to list all the entry points in the form of this table and take stock of the difficulty of implementing AI technology.
The difficulty of determining the implementation can be assisted by the following methods.
- How much public information is relevant to this AI technology? We can check how many academic studies, patents, and media reports have been published about this AI technology. Most of the relevant information shows that this direction has a certain feasibility. However, it may also indicate that this direction is only the content of cutting-edge research, and there is still a long way to go before it actually lands, so after researching this direction, we will also investigate the latter two directions.
- How many companies are working on this AI technology? We can check how many companies are making efforts to research this AI technology, or express their willingness to research in this direction. Most of the enterprises that make efforts show that this direction has a certain value and has high feasibility. Of course, there are also some companies here that are “doing difficult and right” things, so this judgment dimension is mainly used as an auxiliary reference.
- How much open source/commercialization capabilities does this AI technology have? We can check how many open source capabilities or commercial capabilities are available for this AI technology. Because this means that this AI technology has a certain degree of maturity, and we can directly stand on the shoulders of enterprises.
The core is to judge whether there are enough borrowing points to save us from rebuilding wheels.
Business cognition: Business value estimation and prioritization
We then move on to the next step of estimating the value of the business to prioritize the final business realization.
Because although some AI directions seem lofty, they are of no use when combined with business, so what is the purpose of researching and implementing these AI capabilities? Just to look awesome?
For example, we are an internal customer service system, AIGC has been very popular recently, and we have added the function of generating picture content to the customer service system. This may seem high-end, but it doesn’t actually solve the core demand of “user problem solving” in the customer service business.
For example, we are an internal video processing system, and recently let the character jump subject three very popular, in order to catch up with the hot spot, we quickly got a picture dancing AI ability on the system. It does seem to be quite hot, but how much of this demand will be used in actual internal business? It seems to be a waste of labor costs.
Therefore, we need to sort out a set of internal value evaluation standards to sort out multiple AI entry directions.
Value judgments have several dimensions:
1. Number of users: How many internal B-end users can this AI capability be needed? This determines the scope of AI’s role, i.e. how many users there are?
2. Frequency of use: How often can this AI capability be used? This determines whether AI capabilities can be used regularly or occasionally.
The two dimensions of “number of users” and “frequency of use” are used to judge the scale of demand in the direction of AI capabilities. If only one or two people can use it in the end or the overall frequency of use is very low, then it can be said that the space for this AI ability to play a role is very limited.
3. Efficiency improvement value: How much labor cost can be saved after using AI capabilities. Generally, under the same work content and work quality, the difference between manual processing time and machine processing time is compared.
4. Output value: How much value can the content produced by using AI capabilities play in the business. Here it can be measured by direct income or indirect ancillary income.
The two dimensions of “efficiency improvement value” and “output value” are used to measure the one-time value of AI capabilities.
We can use a formula to measure the value of the overall AI direction:
AI direction value = number of users * per capita usage frequency * single efficiency improvement value + number of users * per capita usage frequency * single output value
By using quantitative value and quantitative cost comparison, we can judge the cost-effectiveness and priority of this direction.
Here are some practical examples.
Suppose we are researching AI art-related capabilities, there are currently two directions: video editing and image editing, but the team has limited manpower, which direction should we choose first?
Then let’s break it down according to the above four dimensions:
The above data is a fictitious example.
You can look at this table to calculate the specific values of the two contents separately.
The value of the tool for AI video editing is multiplied by 400w, which of course is a hypothetical value.
The value of AI image editing tools is multiplied by 110w
Then it is obvious that the value of AI video editing tools is high overall. At present, both technologies have borrowing, so the cost is similar. Therefore, without discussing other factors, AI video editing tools are more worth exploring.
In this process, we can retain some directions that “the current team capabilities cannot meet but are of high value” and can be reached after appropriate “replenishment of manpower”. Used to fight for resources in the future.
Because we are ultimately serving the value of AI, not needing to be limited to current team capabilities.
【Business cognition】Based on the existing party and align with the team first
After completing the direction sorting, we need to align with the business team again, which can be in the form of making a rough explanation document by making the direction. It depends on the team’s comprehension, which can be a text description or a high-fidelity prototype.
The main things that need to be done here are:
First, by aligning the existing direction with the team, let the business imagine the scenarios when using these functions, and identify which are pseudo-requirements, so as to correct our direction design and avoid the wrong direction caused by “low business understanding”.
Second, to carry out the preliminary verification of business value to a certain extent, we need to obtain “frequency estimates” and “value estimates” to correct our initial value judgments and optimize our functional ranking.
Third, we need to vaccinate our businesses and reduce concerns about “AI replacing them”.
To reduce business concerns, it is necessary to establish the understanding that “AI is co-creation with people”.
AI is a summary of “content that can be summarized into laws”, which is essentially lagging behind the frontier fields of “continuous innovation and development”.
We must know that AI can do what humans can do!
AI is like an intern, with various skills, but not how to use it, and we need to summarize a set of SOPs to guide them in their business. And they will continue to grow in the process, as long as we can take them and teach them new knowledge and content, then we will not be replaced by them.
I think the cooperation between AI and us can actually be divided into three layers.
1. The lowest level is the low-complexity basic work, which occupies the highest workload.
These tasks can be done completely automatically by AI without human effort.
2. The second level is challenging work, in which AI can assist us in some repetitive work, but the most important key lies in our own judgment and execution.
At this level, we have a cooperative relationship with AI, and AI acts as our assistant here.
3. The top level is the most complex work, which is often innovative and breakthrough.
These jobs are essentially irreplaceable by AI. Unless we are far behind the forefront of the industry. The industry’s large model knowledge is far ahead of us.
This field is also our core barrier, as long as we can get ahead of AI, then AI will be difficult to catch up with us.
Upward cognition: Upward management based on alignment results
At this point, we already have an AI direction based on “self-awareness” and “business awareness”.
At this time, we need to carry out upward management.
First, we need to align the “expectations”. Notice required:
- Team boundaries: What level of content can our team achieve, and what kind of team do we need to build if we want to achieve this level of content.
- Direction and value estimates: We need to inform what we can do and the value.
- Risk estimation: We need to inform which directions are currently “difficult to achieve”, and we need to inform the content that is difficult to implement, and there is a risk of “uncertain ROI”.
By aligning these contents, we need to:
- Confirm the expectations of superiors and adjust priorities. Because the direction we come to often lacks a high-level strategic perspective. So we need to get information from the process of upward management to correct our direction.
- Strive for resources and set up an AI special team with this.
Why set up an AI task force?
First of all, it is possible to obtain human resource advantages, align goals, and supplement manpower. This allows us to promote the implementation of AI more efficiently.
Secondly, through the special team, we can use this to schedule support for other line businesses. Because AI is only a technical capability, all businesses are + to AI, so it cannot be rectified, it needs business fundamentals, needs support from business, needs to fully schedule support from each business line, and if there are special circumstances, it can get support from superiors.
Then, through the group method, we can establish a “decreasing marginal cost” cost advantage to create AI middle office services within the enterprise. A unified team undertaking AI research tasks can significantly reduce duplication, research, learning, and deployment costs, and can serve the entire team with a single investment. Moreover, a dedicated AI research team can concentrate human resources and overcome technical problems.
Finally, the “AI landing” can be better operated. Because AI landing is not only a matter of the development level, but also a matter of operation and management, it is necessary to have an organization that can control the team’s “application of AI” from the overall perspective, and do a good job in the team’s AI+.
AI landing around co-creation ecology
Next, we will start to promote the “AI landing”.
Xiao believes that in the internal B-side, AI technology can truly empower the business, not first study a very good technical ability, and then go to the business to see “what is the effect?” ”
It’s like taking a hammer and going everywhere looking for nails. This is just a self-satisfied pseudo-need.
And what we should do is find the nail and bring a suitable hammer. What can AI do? But what do our users need?
Therefore, we want to discover valuable demand scenarios in the business, and then study relevant AI capabilities based on this demand, and finally combine them with the business to play a role.
Therefore, I personally believe that we need to build a “co-creation ecosystem” to assist us in tapping real needs and implementing AI. For this purpose, I have summarized three methodologies.
- [Building an AI-based business co-creation mechanism and accumulating best practices]
- [Build system functions around best practices and accumulate business results]
- [Deepen the co-creation ecology around the achievements of the system and form a virtuous circle]
Let’s talk about it in detail.
[Building an AI-based business co-creation mechanism and accumulating best practices]
Our first step is to “build an AI-based business co-creation mechanism and accumulate best practices.”
The implementation of an AI capability does not directly start with function development, but needs to first judge its adaptability to business scenarios.
Because an AI model is blown as “powerful”, it does not mean that it is “powerful” in the customized scenario of the enterprise, and it is very likely that the effect of AI is not good.
Therefore, to implement an AI technology, it is necessary to first conduct “feasibility study”, “business testing”, and then “function development” and other steps.
Therefore, there is still a long way to go from discovering an AI technology to the actual implementation, and we cannot let the business wait for us to complete the process before using AI, which will waste a lot of opportunity costs for the business and accumulate certain complaints internally.
Therefore, we need to build an “AI-based business co-creation mechanism” that can:
- Through a set of standardized and fast AI testing and verification methods, it ensures that the adaptation of AI technology to the business is quickly verified. And in the subsequent AI iteration, it can tell the business the best choice.
- Through some rapid deployment mechanisms, the business can quickly use the latest AI capabilities, and in the process, further judge the degree of adaptation between AI and the business, and accumulate AI best practices for the business. Let the business assist us to filter out some directions that are “originally determined to be achievable and valuable, but there are stuck points in actual implementation”, which may be because the current AI capabilities do not meet business demands, or it may be found that “the value is not so high” through practice.
- If the team has the ability to provide systematic solutions in the future, this process can cultivate a group of core seed users and facilitate the subsequent promotion of the “co-creation” ecosystem.
The “AI-based business co-creation mechanism” is mainly composed of the following processes.
- AI exploration: Experiment and explore AI tools based on business directions, and conduct initial screening of AI capabilities on the market.
- Capability introduction: Build a form that allows the business to use external tools, generally in the following three forms. Third-party account: Apply for a budget to purchase an account, do a good job of distribution and management, and let the business experience AI by distributing the account. Open source deployment: Deploy internally with the help of open source tools and third-party platforms for internal use. API: By accessing the API, it can be used by raising requirements or simply accessing the system.
- Business trial: A group of core business members give us a business perspective to evaluate AI and assist us in judging whether AI can really bring value.
- Best practice precipitation: Through business use, record and precipitate our best practice case library to prepare for subsequent systematization.
The overall process is that AI explorers first conduct research and initial screening, and then introduce it into internal business use, and finally precipitate into best practices.
In this process, we need to establish the following:
- AI intelligence information sources: The development of AI is changing rapidly, and we need to establish high-quality information access channels so that we can obtain the latest AI intelligence in a timely manner, and then conduct experiments and explorations to find AI capabilities that are valuable to the business.
- Internal evaluation system: It is necessary to establish a unified and standardized AI evaluation standard internally to conduct horizontal comparison of AI capabilities on the market and vertical comparison of AI capabilities during iteration. Find the most suitable AI tool for your business through quantitative methods.
- Collaborative SOP: If an AI capability fails to provide a systematic solution, it requires human resources to ensure business operation. For example, to access the API for video generation, the business party needs to submit materials to the technical generation. If there are too many demand sides, it will lead to inefficiencies due to management chaos. Therefore, it is necessary to form a set of SOPs to guide cooperation to standardize the docking process and the materials submitted by each process. First, to ensure the efficiency of the team, and secondly, to reduce communication costs.
- Best practice knowledge base: This AI-based business collaboration process is bound to accumulate valuable AI usage best practices. We need to precipitate this part of the content and reuse it for other business parties.
Therefore, we need to build a knowledge base with a unified entrance.
The point here is the “unified entrance”. Why do you say that? Because the essence of the internal middle office is reuse and precipitation, if it is divided very scattered, it will definitely increase the cost of business understanding, and it is not convenient for the team to “accumulate less” to form a scale effect (the development trend of AI is also a kind of accumulation of less). Moreover, assuming that we want to systematize in the future, our system will also develop based on the content contained in this “knowledge base”.
[Build system functions around best practices and accumulate business results]
When our “AI-based business collaboration” model runs and continues to accumulate “valuable” best practices, we need to consider systematic construction and give full play to the role of AI in improving quality and efficiency. Let’s talk about my personal thoughts on this piece.
1. Build the core: Systematic construction with Allinone as the core.
Personally, I believe that the core of system construction should follow the “All in one” idea.
All in one is divided into four levels.
- All AI capabilities are integrated, and internal and external AI capabilities are concentrated.
- All business experience is integrated, and all high-quality business practices and technology exploration are concentrated.
- All business processes are integrated, centralizing all business processes, such as customer service and designing processes from start to finish.
- All application scenarios are integrated, and all application scenarios call our system, such as the company’s data analysis backend, CRM backend, delivery backend, etc., AI capabilities can be called.
To achieve all in one at the level of internal and external AI capabilities, we need to maintain stable AI exploration and accumulate AI capabilities that are beneficial to the interior.
At the same time, the best practices accumulated in the process of “co-creation mechanism” are essentially a kind of high-quality experience, which also constitutes the all-in-one of “business experience”.
To achieve all-in-one at the business process level, we need to accumulate best practices in our “co-creation mechanism”, dig deep into relevant business processes, and accumulate the “point-line-surface” of AI capabilities.
The “point-line-surface” of AI capabilities actually corresponds to the multiple stages of “AI landing on the B-side” mentioned above, such as single-point access empowerment, workflow integration empowerment, and workflow intelligent scheduling stages.
The trial mechanism we built in the process of “co-creation” meets the demand of “single point access empowerment” to a certain extent. But this only serves a small number of people. From this, we can build a feature that provides these single capabilities so that non-co-founders can also use them.
Next, we need to dig deep into its business processes based on existing “business practices” and provide embedded solutions for the entire link to provide automated or intelligent solutions. This is also the key to our entry into the “second phase of AI landing on the B-side”. At this stage, the role of AI in “improving efficiency” and “improving quality” will be further amplified to provide one-stop services to the business.
For example, in stage 1, we only provide a “customer service Q&A assistant capability”, and in stage 2, we need to consider building a one-stop service in the customer service scenario, for example, if a user asks a question, we can record the type of question, register it as a work order, and then assign it to the appropriate member for follow-up, and after the follow-up is completed, the AI will conduct quality inspection scoring.
Finally, we can explore the “workflow intelligent scheduling stage” to try to bring higher efficiency possibilities to business processes.
Next, there is the All in One at the application scenario level, which we can achieve in the following two ways:
- Build an AI capability gathering place: We can aggregate existing AI capabilities into a system, so as to cover various AI usage scenarios, and thereby cultivate the perception that “this platform has all AI capabilities”, so that more people can develop usage habits.
- Package and distribute to various business systems: Most manufacturers already have a certain number of business system construction, and users have also developed relevant usage habits. Therefore, we can encapsulate AI capabilities in the form of APIs, browser plug-ins, etc., so that other business systems can call them on demand.
Of course, these two ways can coexist and do not affect each other.
2. Build order: MVP principle, verify first, then amplify
After determining the core idea of system construction, it is the order of system construction.
The main thing here is to follow the MVP principle, that is, the principle of least feasibility. In this way, the value is verified at the lowest cost, and then it is built at scale.
It can also be understood as “online first, then optimized”. This process is also the key to “accumulating results”, from which we can accumulate sufficient systematic proof of value.
Of course, you are all making products, and everyone has probably heard the cocoon in their ears about this principle, so I won’t go into detail here.
3. Bottom-up strategy: Neptune thinking, two-handed preparation.
Due to the rapid development of AI capabilities, the AI solutions we select through the “co-creation mechanism” may only be suitable for the present, but there is a possibility that they will be subverted by other AI capabilities at any time in the future.
For example, at the beginning, AI video generation was more powerful than runwaypikaluma, but now it is basically a dream.
Therefore, when implementing related functions, it is necessary to have a “sea king” thinking:
1) Be prepared to “break up” at any time:
Ensure that the implementation scheme based on AI-related functions is not too customized and coupled, and clarify the boundaries between your system and AI capabilities. Treat AI capabilities as a “plug-in” that can be replaced at any time, and be prepared to replace them at any time, and when the AI capabilities currently used need to be replaced due to cost or capability issues, you can withdraw at any time.
2) Regularly evaluate the best implementation scenario:
Due to the rapid development of AI, in order not to let enterprises fall behind, the product team needs to regularly access the latest models for testing and verification, and use the “internal evaluation system” built to judge the difference between the latest model and the original model, so as to judge “whether to replace” or “keep using”.
When the system is being built stably, we should pay attention to the accumulation of system results and prepare for the next step.
[Deepen the co-creation ecology around the results of the war and form a virtuous circle]
Based on the best practices and systematic results accumulated in the previous two links, we need to regularly publicize the results, and we can synchronize our results information through internal journals and seminars.
To achieve the following effects:
- Attract more users through successful cases, lower the threshold for use through user training and education, attract more AI users, and further exert the value of the system.
- The internal credibility of the construction team has become one of the leaders of the ecosystem, which is convenient to borrow from all parties to further promote the implementation of AI.
- Through the role of throwing bricks and attracting jade, we will tap more “ecological co-builders”, expand the scale of co-creation, assist us in finding better AI directions and AI needs, and ensure that our AI applications can indeed be carried out around the “real needs of the team”.
Focusing on the core of “co-creation ecology”, we first promote business co-creation exploration, accumulate best practices, implement the system around best practices, and then stimulate more co-creation needs through the promotion of results, and then enable exploration.
In this way, a virtuous circle is formed.
brief summary
The above is my personal practical experience sharing on the implementation of AI in internal B-end products. Of course, these experiences are only extracted from my personal experience and may not be applicable to all scenarios, but I hope to inspire everyone.
