4 business considerations you must take into account before building an Agentic AI system
Agentic AI systems indeed help businesses boost basically any aspects of business and operational activity efficiency, and they have been penetrating every detail and field. Personally, I’ve witnessed their performance and results, which indeed can replace some manual work, in particular which are repetitive tasks
However, the hype of using AI or even overemphasizing AI's importance, which ignores the purpose and reason of applying AI in business scenes, catches my attention. This momentum and trend might cause some new problems that most people might miss out on now.
So in this article, I want to share 4 business considerations you must take into account before building an Agentic AI system, for the purpose of helping you determine whether you need AI, or what AI you need, and how to leverage it.
And in the premium section, I share a comprehensive Agentic AI system building playbook for people to judge if they need an Agentic AI system, what factors they need to take into consideration such as model, framework, orchestration, workflow, database and so on to avoid issues occurring in finance, business model conflict, bad user experience, and decrease the legal and security risk from the built Agentic AI system
If you want to get the complete Agentic AI system playbook, please visit the YouTube video in the opening of the article above, and leave a comment “playbook code” in the video content comment. I will reply to you in the comment with the redemption code
We write the agentic AI system playbook without any AI input and AI hints, which is based on our Agentic AI system development project’s experience from different sectors. Thus, checking the playbook out is not only to explore the Agentic AI system guideline, but also it can be a datasource for you to train or finetune your AI model with our easy2digital human input writing which also includes our style of speaking and expression in both business and technology
Table of Contents: 4 business considerations you must take into account before building an Agentic system
- Purpose of Agentic AI System
- Finance Considerations
- Business Model Alignment Matters
- Marketing Cashflow and User Experience
- Legal and Security Risk
- Comprehensive Agentic AI System Playbook
- FAQ
Purpose of Agentic AI System
A classic workspace scenario nowadays is basically that AI becomes the focus in every meeting, with the term AI repeatedly occurring ” AI…AI….AI ... .etc”. Gradually, most people start just focusing on the new AI tools, new AI platforms, new data engineering strategies, and unthinkingly just apply the new features to the Agentic AI system, where people often react without cautiously taking the business situation into consideration.
Even AI model companies themselves are just testing the market and AI potential, so please do not feel lagging or lose cutting-edge competition, compared to peers or rivals in the market, when not adapting to the newest and latest AI technology instantly.
Instead, thinking about the purpose of building and leveraging an Agentic AI system for your business is indispensable. Any adaptation without thoughtful business consideration has huge potential to affect your current business health.
Thus, here I consolidate the 4 main business considerations you can not miss out as follows when you are building an Agentic AI system, which are Finance, Business Model, User experience, Legal and Security Risk.
Finance Considerations
First things first, AI looks like being free, and also it does not seem expensive. In fact, it’s not free, and somehow it can be very expensive if you are not using it properly. There are mainly two types of costs you need to take into consideration before it potentially impacts your profit and loss calculator.
1. Cost per thousand token
AI token consumption and cost occur from every single action activity (database CRUD, MCP tool calling, input, output and so on). Meanwhile, the uncertain scaling cost occurs on and off as traffic is up.
Every type of AI, including input and output from each brand AI model, has a relative cost per thousand token rate and quote, no matter whether the type of media is text, code, image, audio, video, or so on. Based on the AI features and output quality, some models might be more expensive, such as Claude. But it doesn’t imply better quality is the best, because it very much depends on what type of AI output that the specific field of task or business division needs, and where you use the AI output in the actual application scene
Except for closed AI models, open models are also not free, which most people thought that they were free. For deploying and running open models, you need to be equipped with the relevant specification hardware to install the models and let it have sufficient computing capability and capacity to perform, such as RAM, CPU, GPU and so on. Or if you host in Cloud platform, computing resources can be very expensive no matter which payment model you select - prepay/upfront payment/fixed monthly cost, or pay as you go. And based on the actual models you are using, you need to upgrade the hardware along with the model upgrade. Furthermore, electricity bills are the cost you can not ignore, particularly when your Agentic AI systems run 24/7/365. The total cost of the mentioned items can cost you a lot so that you need to take it into consideration and help you calculate the cost per thousand tokens and create a profit and loss module that can fit in your business operation, even though they are the open models
2. AI Talent cost in different business special field
One myth here and there is that AI can replace humans. The fact is Yes or No. AI indeed causes lots of jobs which have gone away. However, the fact is it is becoming more and more difficult to hire people. The core reason is AI is a must-have skill set for each specialist field in a company, such as engineering, marketing, sales, operation, finance. That implies you are not only hiring talents who specialise in one of that area, but also you are hiring those guys who are able to engineer AI and handle AI in their own field. AI does not kill and finish the required functions for a business, on the other hand it requires employees are capable to work with AI, strategise the purpose, direct it
Please keep in mind that Agentic AI systems at the end serve humans and business so that it requires humans to strategize the purpose based on what humans need. Also it needs humans to monitor, fine-tune, and maintain for the purpose of ensuring things are going in the right direction, after all AI underlying logic is related to probability rate, which means it eventually has an error rate.
Data strategy and security management is one of key areas most people miss out and underestimate the hiring cost and difficulty. In the future, differentiating the performance of AI is data as it directly bundles to the whole process and journey of any AI functioning and output delivery. Business needs employees to have strong knowledge and experience in handling data, securing business confidential, managing data sensitivity and risk
Thus, rather than saying jobs are gone, new function and purpose jobs are coming up. In terms of the new jobs, the hiring cost and labor cost might be higher than ever before, and it’s more difficult to find talent in different special fields particularly.
Business Model Alignment Matters
One of the biggest mistakes you know companies make when building and adopting Agentic AI systems is focusing too much on the technology itself whether actually the AI fits the business model
Different businesses operate under completely different financial models, profit margin models, revenue structure, ideal customer profile, customer expectations, operational model and so on. Because of this, the definition of valuable Agentic AI system can vary significantly between industries and companies.
a SaaS business with thousands of repetitive customer support requests may benefit greatly from an Agentic AI system that is able to automate the onboarding, document instruction, frequent FAQ and entry level troubleshooting. In this case, the Agentic AI system directly supports scalability while reducing the labor cost to deal with high frequent repetitive workload.
On the other hand, people oriented-driven businesses may gain less value from the Agentic AI system. For example, a luxury consulting company replies heavily on personalized human relationships and high-touch communication. So an Agentic AI system might even weaken customer trust and reduce perceived service quality. Also, in eCommerce customer service, when dealing with customers who are requiring to refund or the customer is pissed off or unsatisfied with the product and service, too AI interaction might somehow escalate the conflict as people very often can know who they are talking to, even though nowadays the AI human voice simulation quality is high but lack of human temperature and sentiment at the end.
Marketing Cashflow and User Experience
Another important consideration is return on investment in marketing. When talking about return on investment, I very often get used to drilling down into the quantitative aspect, which analyzes the cash inflow and cash outflow. I believe one of the most active AI adoption fields is marketing, such as the creative generation, ads copy writing, ads material production, social media post automation, email marketing and so on. However, does an Agentic AI system always help increase the revenue and cash inflow and reduce the production cost? The answer is obvious, which does depend.
Take AI commercial video production for example. When building an Agentic AI system with commercial video production capability, generally we need to equip the system with AI image and video generation capability APIs, such as Google gemni, VEO 3.1, Runway and so on. In common, AI video generation costs count by second and AI image generation costs count by token. If we produce a 30 second video commercial published in social media, most people think we just need to prepare a prompt with image reference to generate the video. But the fact is not.
Regardless of the AI material generation cost, please keep in mind the trial and error cost, because AI is not able to judge if we are testing or not. AI charges us every time we use it to generate a video even though the result does not fit in our qualified criteria and our expectation. If every second costs 1 USD dollar, every 30 seconds video commercial costs 30 USD dollars. The final total cost of one video commercial seems like just costing 30 USD dollars, but in fact the final version of the AI video commercial might need 100 times revision, without counting the other materials cost such as image generation for reference, music generation, seed character IP cost (including celebrity), seed character design cost
Then one of the biggest misunderstandings people very often have when using AI is the expectation. Using AI is not equivalent to increasing customer acquisition, customer retention, revenue and cash inflow, because at the end it’s all about the business branding, product pricing, distribution and advertising strategy. Potential customers will not pay for the bill because of using AI. In these mentioned fields, although AI can facilitate you to make decisions, implement A/B testing, increase the turnaround time, AI does not guarantee your marketing success. If you expect an Agentic AI system that can better the marketing outcome without your effort on the marketing strategy and quantitative model, it’s not realistic
As mentioned earlier in the business model, AI can not guarantee enhancing user experience. On the other hand, some cases have already shown AI can further damage the relationship with customers and affect the customer retention rate or repurchase pattern or increase refund rate. eCommerce customer service is a typical case. It not only might cost you no ceiling AI token, but also due to robotic or conversation without sentiment and human temperature, it can escalate the conflict
Legal and Security Risk
Unlike traditional software systems that follow fixed rules and predictable workflows, Agentic AI systems can make decisions, execute actions, retrieve external information, and interact with multiple tools dynamically. While this flexibility increases efficiency, it also introduces new layers of operational and legal uncertainty. So one major concern is data privacy and compliance. And if this field is not handled cautiously and carefully, it might cause legal issues.
As AI agents become more autonomous, it becomes increasingly difficult to determine who is responsible for incorrect decisions. There is one popular saying: “Human in the loop is for no human in the loop next time. As your Agentic AI system can do more things and make decisions and become more autonomous, you might become overreliant on autonomous systems without sufficient human oversight. This pattern can create operational vulnerabilities and uncertainties.
Thus, you need to have a strategy and flow on how to alleviate or reduce the risk of security and legal issues when designing the Agentic AI system in your business model and finance goal.