Oftentimes, when vendors come in to talk about how AI can help a business, they focus on areas that look exciting in a demonstration — things like sales conversations, design work, or areas that need expert judgement. Rarely will you see a demo for a “boring” thing like responding to RFQs. However, these kinds of repetitive, rule-bound, time-consuming tasks are exactly the areas where AI-enablement and integration can bring the highest ROI.
I’ve just finished a five-post series covering AI as applied to a variety of different businesses, ending with the five patterns that showed up in all of them. Those posts stayed at a high level, so this time I want to pick one process and drill into the specifics of turning it from a manual one into an AI-enabled one. I’m going to use the response to RFQs at a steel service centre. Your quoting process will not look like this one, but I hope you’ll see how the same approach could fit your workflow.
Overview
In the steel service centre world, the two most common RFQs that salespeople see are for one-offs (spot pricing) or annual (program pricing). For the purposes of this example, I’m going to focus on spot quotes, where a customer says “I need this grade in this gauge on this date”. Spot quotes are typically for near-term delivery at current market pricing. Quality, pricing, and ability to deliver quickly are table stakes — but when a customer is looking to fill an inventory gap that may halt their production, a quick response is paramount. Program pricing is a different beast altogether — if there’s interest, I’m happy to dive into that.
Accuracy and response speed are important when responding to RFQs, but there’s another factor to consider. In my experience, a salesperson will often pick the “easiest” material to quote when the business priorities might actually point somewhere else. One of the huge benefits of an AI-enabled quoting system like this is decision quality. The quotes go out faster — that’s great — but they also go out presenting options that match the business’s priorities.
First Things First
The first step in any AI-enablement project is confirming access to the data. The system needs to be able to see into the ERP (e.g., inventory, pricing tables, customer history, scheduling), and the data there has to be timely and accurate. Off-the-shelf and custom ERP systems are different routes to access, but both are workable.
Additionally, if there are rules or logic that aren’t embedded in the system (e.g., an unwritten rule of “anything in inventory over 270 days can be sold off at cost” or “if possible, we prefer to use galvanized material from mill XXX”), that logic needs to be captured first. That’s simpler than it sounds — a text document often works just fine.
If any of these pieces are missing, we need to resolve those gaps before proceeding.
The AI-enabled Workflow
So now let’s look at the actual flow. I’m going to note throughout what is actually doing the work — AI, encoded business rules, or a human. It’s also worth noting that the salesperson is alerted whenever a decision or exception needs them — this could be an MS Teams message, an email, or a text.
Step 1 — Read the RFQ and determine if you even supply this material. If not, draft a courteous decline that also sets out what you do supply. This simple guardrail allows your salespeople to focus on real opportunities without slamming the door. [AI: Interpretation and drafting. Human: Reviewing and sending.]
Step 2 — Is there enough detail to quote? If there isn’t enough detail in the RFQ, identify precisely what is missing and draft a clarification message in response. [AI: Interpretation of requirements and drafting. Human: Reviewing and sending.]
Step 3 — Is this a new or existing customer? For new customers, sometimes the credit check can hold up the process, so it’s worthwhile to kick off an inquiry to the internal credit team to get their initial feedback. For existing customers, query the ERP for credit flags so that any issues are known by the salesperson right away. [Code: Run the checks and send the inquiry to the credit team.]
Step 4 — What inventory can satisfy this? The customer’s spec sets the breadth of the search. A narrow automotive spec constrains the search, while general manufacturing specs open it up. The search is going to look for material in stock, already ordered/incoming, or neither in stock nor on order. [AI: Interpret spec for alternative options. Code: Execute search.]
Step 5 — Where multiple materials could fit the customer’s needs, which one is actually best from the company’s perspective? This is the decision quality point I raised earlier, and it’s the step where the system earns its keep. This includes looking at factors such as conversion cost to finished goods, production lead time, aged stock, raw material cost, and what’s worked for similar parts before. [AI: Apply management priorities. Code: Build costs, times, and other factors.]
Step 6 — If it has to be ordered, what’s the best source? The system now looks at deterministic factors such as cost, lead time, and purchase commitments, but also applies “fuzzy” priorities such as preferred vendors or other management/quality department considerations. [Code: Identify suitable sources, build cost. AI: Apply management/quality priorities.]
Step 7 — Apply pricing policies. The system then applies the organization’s pricing rules (e.g., replacement cost, straight markup, matrix pricing) and (if necessary) applies adjustments (e.g., discounts to move stale inventory, win a new customer, or fill a slow line). [Code: Apply pricing rules and adjustments.]
Step 8 — Present a prioritized list of options to the salesperson to confirm and choose. This step is pure human judgement, albeit based on options that have been built and ranked using a mix of AI and deterministic code. [Human: Apply judgement to pick preferred option.]
Step 9 — Draft the response for final review. [AI: Drafting. Human: Approving and sending.]
What This Takes
A reasonable first version of this — scoped to one grade family or one processing type — is a matter of months, not years. The integration work is usually the most time consuming part of the build, not the AI. Start narrow, producing usable quotes for one slice of the business, then widen it as trust builds.
When I’d Tell You Not to Build This
Not every quoting desk should be automated, and there are a few situations where I’d tell you to leave it alone.
If most of your RFQs are genuinely one-off engineering rather than variations on work you’ve done before, the economics don’t hold up. You’d be building a system for the exception instead of the rule.
If the pricing and material logic lives in one estimator’s head and nowhere else, that’s a knowledge-capture problem before it’s an AI problem. Get it written down first. You may well find that exercise is worth doing whether or not you ever build anything on top of it.
And if your quote volume is low, the payback isn’t there. A better intake form and a good template might get you most of the way.
There’s one more thing worth saying: the workflow I’ve described never orders material, and it never reserves it for a customer. It presents options with the costs and the timing attached, and people commit to a path.
Conclusion
Of the nine steps in this workflow, three involve no AI at all. But the more useful pattern is what AI actually does across the whole flow. It interprets unstructured input, it applies fuzzy priorities, and it drafts text. Everything else is deterministic code. And the decision itself stays with a person.
That’s the honest shape of most AI-enablement work, and I think it looks quite different from most demos.
I hope this gives you a better sense of what’s possible when building custom AI-enabled workflows. The question is no longer “can we use AI”, it’s “which process is going to benefit the most from AI-enablement”. If you’d like help answering that for your own operation, I’d love to talk to you.
This post is part of a series on the current state of AI, focused on how it can be applied in practical ways to deliver measurable improvements in productivity, cost savings, and response times. If you’d like to explore more, all previous posts are available here under “Insights”; please read them and reach out with any questions or comments you have.