
When businesses want to pay AI fees based on results
An AI system that consumes millions of tokens may still not create any value for the business. Costs only make sense when tied to a measurable operational outcome.
A survey by Andreessen Horowitz of 50 AI technology buyers, cited by TechCrunch , revealed that more than half want AI costs tied to the work or results produced rather than the amount of tokens consumed. While the survey's size isn't representative of the entire market, it reflects a growing question: are businesses paying for the technology itself or for the value it creates?
DUDI Software's Perspective: Tokens are Not Business KPIs
From a system development perspective, the noteworthy aspect isn't the emergence of an additional pricing method. This trend indicates that many businesses have begun requiring AI to demonstrate its effectiveness through operational data.
Tokens are a technical unit used to measure the amount of data a model processes. They are necessary for providers to calculate infrastructure costs, but they don't indicate what the business has received in return.
A million tokens could be used to process thousands of useful requests. However, they could also be wasted on incorrect responses, poor design processes, or tasks that don't create value.
The CEO doesn't need to know how many tokens the chatbot has processed. They need to know how many requests the system has resolved, how much processing time has been reduced, how many manual operations have been eliminated, and whether it has impacted revenue.
Pricing based on results is not simple.
Paying based on results sounds reasonable, but first, businesses need to define what "results" mean.
In customer service, the result might be a request resolved without needing to be forwarded to a staff member. In accounting, it might be an invoice checked and entered correctly. In business, it might be a prospect categorized accurately enough for staff to follow up.
Every outcome requires a quality standard. If AI fulfills a support request but the customer still needs to contact you again, the job isn't truly complete. If AI generates many leads but the majority are unsuitable, the quantity doesn't reflect effectiveness.
Businesses also need to establish a baseline for comparison. Before using AI, how much time did an employee need to complete a task? What was the error rate? What was the average processing cost? Without data before deployment, it's difficult for businesses to know whether AI has created a real improvement or simply added another layer of technology.
The website must be able to record the customer journey.
If AI is used for advising or generating leads on a website, its effectiveness cannot be measured solely by the number of conversations.
Websites need to record where users come from, what content they are interested in, whether they have provided information, and whether they continue with the desired action. This data should be transferred to a management system instead of remaining separate within the chatbot tool.
Even if AI creates a good exchange, if customer data isn't integrated into the CRM, the business can still miss opportunities. In that case, the problem isn't with the model's capabilities, but with how the systems are connected.
CRM transforms AI performance into sales results.
CRM is where businesses can track where a customer, identified by AI, has progressed in their buying journey.
Were the customers contacted by staff? Did they receive a quote? Did the opportunity translate into a contract? If unsuccessful, was the cause due to customer quality, response speed, or the way the consultation was conducted?
When websites, AI, and CRM are connected, businesses can compare the effectiveness of different customer sources, consultation scenarios, and individual customer service processes. AI is no longer judged by the number of responses, but by its impact on conversion rates and processing time.
ERP measures value at the operational level.
For tasks related to orders, inventory, production, or accounting, the resulting data is typically stored within the ERP system.
An AI-powered demand forecast is only valuable if that forecast helps a business adjust its purchasing or production plans. An automated order processing system is only effective when errors are reduced, delivery times are shorter, and financial data is accurately updated.
If AI operates outside of ERP and internal processes, businesses only see the work the AI has done without measuring its impact on overall operations.
Therefore, if a business wants to charge based on results, it must have a system capable of verifying those results.
The relationship between businesses and suppliers will change.
When costs are tied to results, AI vendors must have a deeper understanding of customer processes. They shouldn't just provide APIs or models, but must be involved in defining input data, completion standards, and quality control mechanisms.
Future pricing models could incorporate platform fees, usage fees, and performance-based costs. However, regardless of the approach, businesses still need to clarify data ownership, how results are calculated, and accountability when AI errors occur.
Where should a business begin?
Before choosing an AI model or negotiating pricing, businesses need to prepare for measurement capabilities:
- Choose a process with clearly defined inputs, outputs, and responsible parties.
- Connect AI with your website, CRM, or ERP system to capture real-world results.
- Evaluate simultaneously the speed, cost, accuracy, and impact on the overall process.
Businesses don't need AI to create more operations. Businesses need AI to create better results, with controllable costs and verifiable data. To achieve that, the technology platform and operational processes are still the parts that need to be invested in first.