From Excel Pricing Models to AI Assistants: A Practical Lesson from AWS

Many businesses in Saudi Arabia and the wider Gulf still depend on large Excel files to decide prices, prepare quotations, approve discounts, evaluate feasibility, or compare procurement options. These spreadsheets often become difficult to maintain: one person understands the logic, several versions circulate by email, approvals happen outside the file, and management has limited visibility into how a final price or recommendation was produced.

What the AWS example actually shows

Fortune reported that Amazon Web Services CFO John Felton said the AWS pricing team turned an 18-tab Excel model into an AI chatbot used to evaluate customer deals. Fortune also noted Felton’s view that AI value is not only about productivity gains, but about rethinking how work gets done.

For business leaders, the important lesson is not “Excel is bad” or “every spreadsheet needs a chatbot.” The useful lesson is that when a spreadsheet becomes a decision engine, it may be time to convert it into a more controlled business system.

An 18-tab workbook usually means there are many dependencies: pricing rules, eligibility conditions, cost assumptions, margin checks, exception cases, and approval thresholds. In this situation, the spreadsheet is no longer just a calculation tool. It is part of the company’s operating process.

That is exactly where AI can help, if implemented carefully. Instead of asking staff to search through tabs, formulas, old emails, and policy documents, an AI assistant can guide users through the right inputs, explain what information is missing, highlight exceptions, and route the case for approval. But the assistant must sit on top of clean rules, secure data, and a clear workflow.

When a spreadsheet is ready to become an AI assistant

Not every spreadsheet should be converted into an AI tool. A simple budget tracker or one-time analysis may be better left as Excel. The opportunity appears when the workbook is repeatedly used by many people to make important decisions.

For Saudi and MENA companies, common examples include pricing calculators, quotation templates, project feasibility models, procurement comparison sheets, commission calculations, financing assessments, and discount approval files. These tools often grow over time because each new business case adds another tab, formula, or manual workaround.

A spreadsheet-based AI assistant may be useful when:

• Sales teams rely on complex rules to prepare offers or discounts.
• Finance teams need to check margin, payment terms, tax treatment, or approval levels.
• Managers ask the same questions repeatedly before approving a quote.
• Different branches or teams use different versions of the same model.
• The company cannot easily see who changed what, when, and why.
• New employees need a long time to learn the workbook.

The goal is not to make the process look more modern. The goal is to reduce errors, shorten cycle time, improve governance, and make business logic easier to manage.

A good AI assistant should help a non-technical user ask questions such as: “Can I offer this discount?” “What approval is needed?” “Which cost assumptions are missing?” or “Why is this quotation below the required margin?” The answer should be based on approved company rules, not on a generic AI guess.

What must be standardized before adding AI

The biggest mistake is to wrap a chatbot around a messy spreadsheet and expect reliable results. If the underlying model is inconsistent, undocumented, or full of hidden exceptions, AI will only make the confusion faster.

Before building an assistant, the business needs to standardize several elements.

First, the decision rules must be documented. This includes pricing formulas, discount ranges, margin thresholds, approval levels, customer categories, product rules, payment terms, and exception handling. If the rule is “ask Ahmed in finance,” that is not yet a rule. It needs to be converted into a clear policy.

Second, the required inputs must be defined. For a quotation, this may include customer type, product or service scope, quantity, contract duration, delivery location, payment schedule, implementation cost, and any special terms. The assistant should not continue if required information is missing.

Third, the output format must be consistent. A quote, approval request, or feasibility summary should follow a standard structure. This helps management compare cases and reduces back-and-forth between teams.

Fourth, ownership must be clear. Someone in finance, sales operations, procurement, or management must own the business rules. The technology team can build the system, but the business must decide what is correct.

Fifth, the company should identify which parts of the model are calculations and which parts are judgment. AI can support both, but they should not be treated the same. A formula can be automated. A special discount for a strategic account may still require a manager’s decision.

This preparation may feel slower than launching a chatbot quickly, but it is what makes the final system trustworthy.

Controls matter more than the chatbot interface

For a business-critical pricing or approval process, the chat screen is the least important part. The real value is in permissions, audit trails, approvals, and integration.

Permissions decide who can view, calculate, change, approve, or export information. A salesperson may be allowed to simulate pricing but not approve a margin exception. A finance manager may see cost assumptions that should not be visible to all users. Senior management may need dashboard access without editing rights.

Audit trails are equally important. The system should record who submitted a case, what inputs were used, what the assistant recommended, which rules were triggered, who approved the exception, and what changed between versions. This protects the business and helps resolve disputes later.

Human approval must remain part of the design. AI can prepare the case, check policies, summarize risks, and recommend the next step. But for sensitive decisions such as unusual discounts, non-standard payment terms, or high-value commitments, approval should stay with authorized people.

Integration with CRM and ERP systems is also essential. If the AI assistant is disconnected, users may need to copy customer data from the CRM, product data from another system, and invoice or cost data from the ERP. This creates manual work and new errors. A better design connects the assistant to trusted systems so that customer records, product lists, price books, inventory, tax settings, and approval statuses are aligned.

For example, a quotation assistant should ideally pull customer details from the CRM, apply approved pricing logic, check finance rules, create an approval workflow, and then send the final approved quotation back to the relevant system. The user experiences a simple conversation, but the business gets a controlled process behind it.

How to start without overcomplicating the project

A practical first step is to choose one spreadsheet that is important, repeated, and painful. It should be valuable enough to justify improvement, but not so broad that the first project becomes too large.

Start by mapping the current process: who uses the file, what decisions it supports, where data comes from, who approves exceptions, and where the final output goes. Then separate the spreadsheet into three layers: data inputs, business rules, and outputs. This makes it easier to decide what should be automated, what should be reviewed, and what should stay manual.

Next, build a controlled prototype. The prototype does not need to replace every feature at once. It can focus on a narrow use case, such as discount eligibility, margin checking, or quotation preparation for one product line. Let real users test it with real scenarios, then adjust the rules and workflow.

Finally, measure success in operational terms. Are fewer versions of the spreadsheet being used? Are approvals clearer? Are exceptions easier to review? Are managers getting better visibility? Are users spending less time searching through tabs and more time serving customers?

Key takeaways

• Complex spreadsheets often contain important business logic that deserves stronger control.
• AI assistants are useful when they guide decisions, not when they simply add chat to an unclear file.
• Standardized rules, inputs, outputs, and ownership must come before automation.
• Permissions, audit trails, human approval, and CRM/ERP integration are critical for trust.
• Start with one high-value workflow, prove it, then expand carefully.

If your company has a pricing, quotation, finance, or approval spreadsheet that has become difficult to manage, Pioneers.dev can help you assess whether it should remain in Excel, become a custom business system, or evolve into a controlled AI assistant. You are welcome to request a free WhatsApp consultation to discuss the best next step.