We continue the series of articles about digital advisors and artificial intelligence. The first article is available here , the second here .
Content.
Using specialized standards and methodologies for “additional training”
Combination of systems with ready-made recommendation structures + chatbots as an additional option
Entering and gaining popularity through small and medium-sized businesses
Personalization of recommendations based on the psychotype and competencies of the manager and team
Introduction
As we already understand, digital advisors are a promising but complex area.
On the one hand, they are able to prepare recommendations taking into account your personal qualities. And the combination of the Internet of things with artificial intelligence is generally the key and most promising direction of development.
But on the other hand, all digital advisors have a huge number of shortcomings, which lead to the fact that this technology is in its infancy and does not reveal its potential.
So let's think about what to do and where to run? What awaits us in the next 5-20 years.
Specialization of advisors with artificial intelligence
We communicate a lot with different specialists and always agree on one thing: there will no longer be all-knowing chatbots that are trained all over the Internet. There will be strong and complex models, but their scope of recommendations will be narrowed to specific topics. Interestingly, researchers from MIT came to a similar opinion.
That is, the same chatbot and language models will go into the area of narrow and medium specialization. Everything is like people. You can't be an expert in everything, you need to understand your specialty and develop in it.
If we talk about AI systems, then in this scheme we get several key advantages:
the demands on computing equipment are reduced;
the issue of organizing security is simplified;
When the model does not have information about how to create explosives, then it will not be able to answer such a question.
the risks of AI hallucinations are reduced and the quality of recommendations increases
When a model is based on a specific theme, it does not try to combine everything possible in this world. This means her recommendations are getting better. An example of this is the Eliza bot, which was developed in the 1960s as a robot psychologist and passes the Turing test better than ChatGPT-3.5, second only to models based on ChatGPT-4. The study is available here .
And these results coincide with our observations. We conducted tests ourselves, and the model, which responded based on our books, gave orders of magnitude more relevant and specific recommendations than open chatbots. In addition, there were also an order of magnitude fewer hallucinations there.
AI designers
One of the key tasks for popularizing the technology is lowering the entry threshold. The same thing happened with websites. Once upon a time it was necessary to hire specialists, but now you can assemble it on a construction set. Yes, highly loaded and optimized websites are made by developers, but a completely working solution can be made using a website builder.
The same will happen with AI advisors. Yes, complete solutions for complex problems will remain. But there will also be the development of boxed construction kits that can be configured according to the No-Code principle. It will be possible to upload regulations or a book into it, and it will provide consultations.
At the same time, these designers should be relatively lightweight and undemanding to the IT infrastructure. That is, we are talking about convenient UX|UI and optimization of solutions (limiting capabilities and simplifying models). This is necessary in order to have a viable on-premise deployment model or an acceptable cost of renting infrastructure using IaaS or PaaS models for startups.
Using specialized standards and methodologies for “additional training”
The next direction we see is the creation of DSS (digital advisors) based on international or national standards. Although it is possible to use proprietary programs and methods, this approach has every chance, taking into account the paragraph below about implementation in small and medium-sized businesses.
Our advisor and I are following this path. We took PMBOK 7 and created the structure of recommendations based on it with our modifications. Yes, of course, any business wants everything to be done for it, but let's be honest - these are mostly wishes, and not a reasoned need.
For 90-95% of companies, “basic” methodologies, which contain either international standards or proprietary methods, are sufficient. And from our experience, which is based on work in oil production, its refining, construction, furniture production, energy, and logistics, we can confidently say that discipline and implementation of standard recommendations are more important, rather than trying to finish everything for ourselves.
It is also appropriate here to recall the Japanese approach to learning - SHU HA RI . According to this approach, when learning and developing skills, one must go through 3 stages:
Stage Shu (守)—“follow the rule.”
We do everything according to the letter of the rules. The goal is to understand the base and gain experience. In the case of teaching theory, we form a system of principles on which this theory is built.
The next stage of Ha (破) is “break the rules”
We begin to get rid of the excess. You should move on to this stage only after the “base” has been completely mastered. Ideally with the support of a “teacher”.
The final stage of Ri (離) is “separation from rules.”
You have already absorbed the essence of the methodology, you have realized the very essence of the teaching. Rules are no longer needed and the time has come to “separate” from the teacher/standard. We create our own “style”, our own theories and practices.
The same is true, for example, with project management. If the company does not have a project culture and experience, then it is better to first be based on what people with experience have come up with in the world and follow the rules. Yes, it will be painful and uncomfortable, but useful. And only then come up with your own standards. Otherwise, as experience shows, there is an attempt to pile everything up in order to end up with a methodology that is not applicable at all.
And if we talk about a digital advisor, we get:
an expensive solution in which millions have been invested;
people who do not understand the processes, what and how it works, and, at best, simply fill out forms without thinking about it in order to spend minimal time on it.
As a result, we have an expensive toy.
Creating advisors based on industry statistics
The third direction is the creation of industry DSS. Regulators and the state have a key resource - they receive any necessary information. That is, they do not face the fact that there is not enough data for AI. Regulators and the state have the same Big Data . Thus, they have the opportunity to train AI on anonymized data and build DSS with high quality recommendations.
Manufacturers of industrial equipment from Europe and the USA are following a similar path. As designers, they know what the equipment is made of, making digital models even before production starts. And then they collect data on the operation of their equipment from customers from the moment of launch to the moment of disposal. As a result, they have digital twins that combine both a mathematical model and operational data. They are developing predictive analytics and can offer clients a service model for equipment sales (when you buy equipment, but delivery, installation, commissioning, and maintenance are carried out by the manufacturer and distributor). This adds value to customers and improves business margins.
Unfortunately, Russian manufacturers have not yet come to such a model and simply sell hardware.
Combination of systems with ready-made recommendation structures and chatbots as an additional option
Businesses want systems, not chatbots. We discussed this thesis in the previous article . To communicate with a chatbot, you need to ask the right questions, and this is the most difficult thing.
But this does not mean that chatbots have no future. They will show themselves perfectly when, in addition to ready-made recommendations, you can ask questions to the assistant, already having analytics with conclusions. In this case, the user can formulate the question that interests him in detail. And a chatbot trained on certain data will be able to give not abstract recommendations, but relevant and substantive ones. That is, there is a rapprochement between the two sides - the user no longer invents something incomprehensible, but asks a question in a certain context, and the chatbot does not try to guess from the coffee grounds from a database all over the Internet.
Entering and gaining popularity through small and medium-sized businesses
As we discussed earlier, introducing innovations through corporations is, to put it mildly, a difficult undertaking.
First, let's remember the Moore curve for product innovation.
Let's refresh its decoding a little.
Innovators - always think in new ways and want to change everything
Early adopters (visionaries) - they like new ideas and perspectives; they are ready to try them, they buy not a “product”, but a “promise”, trying to be in the forefront
Early Majority - trying to identify new opportunities, participating in discussions
Late majority - trying to identify errors and threats, require preliminary verification
Those who lag behind are afraid of making a mistake and demand proof. They don't want to change anything
And an excerpt from the book, for those who like to understand in more detail:
Innovators actively follow new technologies. Sometimes they try to gain access to them even before launching a formal marketing program. This is because technology is central to their lives, regardless of what function it performs. They are interested in any significant new product and often make a purchase simply for the pleasure of exploring the capabilities of a new device. There aren't many innovators in any given market segment, but getting their attention early in the marketing campaign is still key because their endorsement reassures others in the market that the product really works.
Early adopters, like innovators, buy into new concepts early in a product's life cycle but, unlike innovators, do not understand the technical intricacies. Rather, they are people who can easily imagine the benefits of a new technology, understand and appreciate them, and relate those potential benefits to their interests. And their purchasing decision will be based on the degree of this fit. Since early adopters do not rely on authority when making purchasing decisions, but rather rely on their own intuition and foresight, they play a key role in opening up any new segment of the high-tech market.
The Early Majority shares some of its passion for technology with the Early Adopters, but is ultimately driven by a well-developed practicality. They know that many new-fangled inventions turn out to be passing fads, so they prefer to wait and see how others handle the product before buying it. They want to consult reputable sources of information before making significant investments. Since this segment includes a large group of people (approximately a third of the entire technology adoption lifecycle), winning their sympathy is the main requirement for generating significant profits and ensuring sustainable growth.
The later majority shares the position of the early majority, but there is one very important difference. Members of the early majority are quite capable of handling a technological product, but members of the late majority are not. As a result, they wait until the product becomes an accepted standard, and even then they want to receive enormous support, so they gravitate toward purchasing products from large, well-known companies. Like the early majority, this group makes up about a third of all buyers in any market segment. Gaining recognition for them is actually very profitable, because as the product ages, the profit margin decreases, so do the selling costs, and virtually all R&D costs are already amortized.
The last group is the bumpkins. They simply don't want anything to do with the new technology for a variety of reasons, both purely personal and economic. The only time the louts buy a technological product is when it is buried so deep in the depths of another product (like, say, a microprocessor built into the braking system of a new car) that they do not even suspect its existence. It is generally believed that louts should not be taken into account at all.
It is quite logical that large companies are clearly not the categories through which innovation comes. The very culture of such companies contradicts this. They want security and incremental growth, not risky breakthroughs. What is the price of failure? Remember when and what truly powerful innovation came from the corporate segment?
At the same time, small and medium-sized businesses are a completely different culture and environment. He wants to grow and become big, but he does not have such resources and opportunities, he cannot “vacuum clean” the market of people and form huge states. He wants to grow, but is limited in resources, which means he is more willing to take risks and more flexible.
This is a field in which you can introduce innovations in order to later gain a foothold in people’s minds and earn billions in the corporate segment. It is also useful to recall a tool such as the Overton window .
That is, you must first enter a small and medium-sized business that is interested in innovation and is willing to take risks. And for this, media exposure and popularity, the exclusion of academicism and simple human language are important. At the same time, you need to understand that at this stage we will not make money from the technology and we need patience. At this stage, the technology is being popularized and errors are being debugged. The first stages of the Overton window are going through due to innovators and early adopters.
First you need to become the norm in society and only then gradually create corporate products and enter corporations.
From our own experience, we can say that even if there are guys inside the corporation who are ready to do everything for a conditional “thank you” and bring enormous value, they are unlikely to succeed. If this happens, it will not be thanks to it, but in spite of it. There are many examples of this, and the most striking is the history of the creation of Linux.
Personalization of recommendations based on the psychotype and competencies of the manager and team
The latest trend in DSS is personalized recommendations. And this is quite logical. Every person has their own strengths and weaknesses. Strengths need to be made even brighter, and weaknesses brought to the required minimum, and not try to make a rhinoceros out of an elephant.
In our work, we actively use Adizes’ PAEI theory, which we have already written about here.This is an easy to use tool. So, it makes no sense for a bright E-type to give recommendations on detailed elaboration of plans and risks. This will contradict his psychology and such recommendations will be ignored by default.
Therefore, one of the directions in the development of DSS is taking into account both the psychological type of the leader and the composition of the team with a description of the competencies and psychological type of each key participant. In this case, the DSS will prepare its own recommendations for everyone.
Here you can use various tools, both Adizes and DISC or any others. The key is that the tools should not be overly complex. We put the same logic into the target model of our digital advisor for project management - the final recommendations should be based on the strengths of the manager and suggest the required minimum in other areas.
And here we gradually approach the key factor - a systematic approach.
Systematic approach to implementation
Yes, and again we return to the systems approach. Until you bring order to your business and systematize it, remove the chaos, no AI advisor will help you.
If a business does not have an organizational structure, then how will it understand who the key team members are and what competencies people have?
If there are no business processes or they are not described, there is no quality data, projects are implemented anyhow, by people without competencies, then AI will again be powerless. And it doesn’t matter whether they are speech models or analytical systems.
For example, if in your ERP there is unreliable data that is entered by people and contains tons of errors, then what can AI advise, even if it understands your psychotype?
You also need to think about lean manufacturing, both in terms of processes and data flows. This will reduce the load on the IT infrastructure and advisors, which means costs will be reduced. And if not, prepare for huge IT budgets.
If you have weak horizontal communication, then it is better not to think about introducing innovations at all.
In such a situation, you will not be able to create a DSS for yourself or take a ready-made solution from the market. That’s why we believe that the implementation of AI in business should be based on a systematic approach and team training.
Yes, this is difficult to do in a small business. It has few people and competencies, and does not have huge budgets to create a high-quality IT infrastructure with various solutions. Therefore, mechanisms for testing psychotypes, descriptions of teams and processes need to be embedded in the AI solutions themselves in the format of ready-made modules or No-Code, as well as integration mechanisms (API) with other services must be incorporated and tested.
Personnel training and competencies development
If your people are absolutely not ready to interact and accept technology, then whatever you implement will be sabotaged. Or they will wait until someone else works out everything for them, comes up with it and implements it. But this will not happen, especially in small and medium-sized businesses.
That is, people need to be trained in digital competencies and move away from a culture of force or bureaucracy. And here another question arises: is this possible in corporations, or is everyone there waiting for the IT department to come and do everything perfectly, giving ready-made instructions for every day?
Therefore, you need to create sandboxes where you can conduct experiments and reward staff for implementing AI solutions, for example, advisors for procurement.
Simplifying interfaces and eliminating unnecessary features
Unfortunately, IT developers tend to create solutions based on functionality. And this leads to two problems:
The interface of solutions becomes more complicated, which immediately becomes a barrier and leads to technical resistance . Especially if we are talking about small and medium-sized businesses.
we get a huge number of functions that no one needs, and these are useless costs (increasing the price of the product) and staff resistance. Therefore, it is optimal that any function comes from a user request. Below are statistics on the use of the functionality of IT solutions from Standish Group .
We discussed this issue in detail here .
Let's also look at this issue from the perspective of those who will use the AI advisor, that is, managers. Now let's remember which two operating systems for phones dominate the world? IOS and Android. What do people with high incomes use more often? Relatively simple and limited to tamper IOS or functional Android?
It’s the same here. It is necessary to create AI advisors with a minimum of functions and interface elements for users, but with a deep study of scenarios for interaction with the solution and optimization of UX|UI design .
Using top-level AI systems (AI orchestrators)
Another promising area is the creation of high-level AI. Let's call them AI orchestrators.
This approach has long become a mandatory element for IT infrastructure. Ask an experienced CIO if it's possible to build a productive IT infrastructure without VMware or its equivalent.
Orchestrators control the start, stop and restart of IT services. They are responsible for distributing the load between services and monitoring their availability. The orchestrator provides the ability to define dependencies between services and set the order in which they are launched.
Also here, AI orchestrators are needed that will decompose user requests between different AIs, eliminate conflicts between the recommendations of different AI systems, and determine the order of launches and data exchange between AI advisors.
Implementation of feedback mechanics
Let's remember how AI learns?
That is, feedback is important for AI. How to evaluate the quality of recommendations? In our opinion, it is optimal based on feedback from the user + evaluation of results from accounting systems over time. The same logic applies to our digital advisor.
The target system provides a module for assessing project results, which will work based on feedback from project participants and collecting data from the system.
Project participants will answer three questions:
What project product created value (high, medium, low, none)?
Are you satisfied with the implementation of the project?
Any suggestions for the future?
Information will be taken from the system:
Did the project close on time (estimate based on plan and actual)?
Did the project come within the planned budget (based on plan and actual)?
Have the requirements for the technical specifications been met (based on closing documents and acts, availability of claim work)?
And yes, each organization will have its own characteristics; it is impossible to create an ideal system for everyone. Therefore, we again move away from the need to create efficient learning algorithms (which need less data) and to local models.
Another example, but just for chatbots and language models, is the introduction of a system of likes and dislikes. Based on them, the system will be able to understand how satisfied the user is with its answers.
But with all this, there is a pitfall - the introduction of feedback mechanics can lead to the AI coming to very ambiguous conclusions. After all, he has one criterion - success or failure, nothing else is important to him.
Popularization and marketing
So, the potential of AI needs to be unlocked through small and medium-sized businesses. That is, you need not to target individuals of large corporations, but to build mass marketing, as in B2C, using digital channels. That is, you need to build a marketing strategy similar to B2C, with aggressive advertising, simple words, meanings and a marketing budget of 50-70% of the total.
A strategy through conferences here is unlikely to yield high returns, but through social media. networks with viral advertising, provocations and quality content - yes.
Another key point in such promotion is thinking through CJM and convenient entry into the product: registration with a minimum number of actions, a subscription model with a low cost of entry and a complete history, useful materials for users, etc. And if we look at the current leader (Chat GPT from Open AI), this is exactly the path they are taking, making their product the standard for everyone and giving access to everyone at a minimal cost, rather than focusing on corporate sales with billions in checks.
AI regulation
Based on the pace of AI development and an analysis of what is happening in the world in the field of AI regulation, we clearly understand that a risk-based approach to AI regulation will develop in the world.
And the development of a risk-oriented approach will in any case lead to the fact that strong and super-strong A.I. will be considered the most dangerous. Accordingly, there will be the most restrictions for large and powerful models. Every step of the developers will be controlled. This will cause the costs and challenges for development and implementation to increase exponentially. As a result, we have problems for both developers and users, which reduces the economic potential of such solutions.
At the same time, specialized models based on local and cut-down AI models that can do little will be in the area with the least regulation. Well, if these AIs are still based on international/national/industry methodologies and standards , then there will be not restrictions, but subsidies.
As a result, the combination of such “weak” and limited solutions based on AI designers in combination with an AI orchestrator will make it possible to bypass limitations and solve business problems. Perhaps the bottleneck here will be AI orchestrators. They will fall under the medium risk category and will most likely have to be registered.
Case Study
Here we want to make a small example of the practical application of several trends, namely AI orchestrators, specialized advisors and AI designers.
Let's simulate the following situation. You have a large organization with various independent divisions. Let's say a holding company with subsidiaries in different regions. Your key processes are identical: procurement, repairs, sales, etc. Naturally, most regulations do not work, since they are overloaded and incomprehensible to non-specialists.
As a result, you decide to implement digital advisors for employees to simplify their work and ensure compliance with rules and regulations. If we take a combination of an AI orchestrator and an AI designer for local models, we end up with the following picture.
The ability to implement everything in one IT system.
You don't have to make 50 different interfaces and systems. You can do everything in the window of one chatbot or personal account. The user does not have to switch. This means we are improving UX|UI and reducing staff resistance (including for technical reasons )
Improving the quality of recommendations and safety
As we covered earlier, AI advisors with specialization will have higher quality recommendations and less hallucinations. In addition, it is easier to ensure security in such advisors.
Speed and low cost of launching solutions
If we deploy an AI designer into which you can simply load a document and immediately communicate with it, then this will allow departments to quickly launch new advisors. Test them, and either refuse (no money is invested, it’s not scary to admit a mistake), or connect them to the orchestrator and put them to work.
Low price of the designer and load on the IT infrastructure, high speed of work
Purchasing pre-trained, but optimized and local models at the heart of the AI designer will be economically feasible. The example is the same as with websites. Developing a website by developers can cost 1 million, and launching it on a website builder can cost 20 thousand. Here the economic model will be similar.
Again, the on-premise model is less demanding on IT infrastructure, which simplifies on-premise implementation and reduces the cost of IaaS or PaaS rental.
As a result, we get an affordable price and low IT infrastructure requirements for both on-premise and cloud deployment. An additional bonus is higher speed of work on high-quality and modern infrastructure.
Ease of maintaining and updating data
Regulations and processes change regularly. There is no business that does not change. If we have a set of local models that can be updated without special training, then we get a huge advantage. Firstly, departments will be able to update the database themselves, which means the AI advisor will give up-to-date recommendations. Secondly, you will relieve the IT service. And considering the prices for IT specialists, this is a huge savings in resources. Trying to retrain a centralized, large model every month will be both difficult and expensive.
As a result, using an approach that in IT is called microservice, we get:
reduction in cost and implementation time
the ability to implement everything through one interface
simplifying and reducing the cost of maintaining and updating the “knowledge” of the advisor.
And the more regulated the industry is, the stronger the effect of such implementation will be. Firstly, the requirements for employees and their verbatim knowledge of all regulatory documents are reduced. Secondly, the speed and quality of management decisions increases. Instead of waiting for responses from employees and loading them with ideas, a manager can quickly get a response from an advisor. At the same time, as experience shows, digital advisors have learned to give a simpler and more humane answer than specialized experts with professional deformation. And thirdly, our efficiency and safety increases. We don't need to hire employees with intimate knowledge of documents. We save them time by reducing the importance of constantly monitoring new rules. And this means independence and cost reduction.
However, even with this approach it is impossible to manage with IT implementation alone. A systematic approach is still needed:
described processes and regulations are needed;
we need to simplify them and eliminate losses (gaining knowledge of the process and following it does not make the company more efficient);
the team needs digital and project competencies (someone must prepare initiatives, launch them, update them);
it is necessary to build communication between departments , implement changes step by step and with PR, etc.
Summary
AI advisors will undoubtedly be the trigger that will take people to the next stage of evolution. And this is a huge market that will take someone’s business to a new level, and for others it will become the main product.
But AI expansion will come through small and medium-sized businesses, not corporations. SMEs are more open to experiments and willing to take risks.
Developers need to provide:
specialization of advisers in certain areas;
create AI constructors with optimized models inside;
the use of specialized standards and methodologies for “additional training” so that off-the-shelf solutions can be implemented;
creating advisors based on industry statistics;
combine ready-made recommendations with language models (chatbots) on the topic of use;
combine mathematical modeling and analysis with psychology and organizational management, that is, prepare recommendations taking into account the psychotypes and competencies of the team, coordinating the participants;
implement feedback mechanics;
simplify interfaces and eliminate unnecessary functions;
a huge amount of work will lie in the area of orchestration of AI advisors;
it is necessary to allocate 50-70% of the budget for marketing and build it as for the B2C segment.
For business you need:
introduce system management in business based on classical tools (strategy, organizational structure, processes, project management, lean manufacturing);
train staff and develop competencies;
prepare for the implementation of packaged solutions based on international / proprietary / internal methodologies.