Opens in a new tab

Blog / Sales & CRM

Sales & CRM

Quoting on the buyer’s cycle: an AI sales agent plan for a B2B timber wholesaler

Nov 20, 2026·Updated ·13 min read

A salesperson checks a customer conversation beside strapped timber packs in a wholesale timber warehouse

A wholesale timber importer and distributor runs on a sales rhythm that looks nothing like retail. Only a handful of orders move each day, but each order is a full pack of timber worth a great deal, and the buyers are manufacturers who come back on a production cycle. The problem this business brought to the table was not a lack of customers. The problem was that quoting ran on individual memory: whichever salesperson happened to remember which customer was about to run out of stock, and which lot was still in the warehouse, was the one who quoted at the right moment.

This article records the problem and the AI agent plan proposed for the business, at solution-pattern level. It is anonymized under the rule for proposal-stage customers, and it states no results, because there is no live system yet to measure. The value of the case is in how the problem was dissected and how the automation scope was chosen: start with quoting, because quoting is where the system touches revenue.

Key points

  • B2B timber buyers reorder on a production cycle, so the timing of a quote matters almost as much as its price.
  • The proposed AI agent reads purchase history in the ERP, checks real stock, then drafts and sends personalized quotes over Zalo and email twice a week.
  • Salespeople step in only when a customer replies. Negotiation and long-term relationships stay human.
  • A companion internal assistant lets sales pull lot photos, specifications and documents in seconds instead of chasing colleagues across chat groups.
  • The whole scope connects to the existing self-built ERP through its API. Nothing already running gets replaced.

The setting: wholesale timber is a game of cycles and stock

The business in this case imports full packs of timber from North America, Europe, South America and Africa, and sells wholesale to manufacturers and sub-distributors. It does not sell retail and it does not manufacture. The catalogue is the familiar commercial species, oak, walnut, ash, teak and pine among them, but every lot differs in dimension, grade and landed cost.

Three operating facts shape the entire sales problem.

First, orders are few and heavy. A handful of orders a day means every missed opportunity is expensive, and there is no large flow of small retail buyers to make up the volume.

Second, demand repeats on a cycle. A workshop buys timber against its own production schedule. When a lot of oak runs out, a few weeks later the buyer returns for the same species or an equivalent. Whoever quotes in the week the customer is about to run out has a real advantage over whoever quotes at random.

Third, stock turns over by shipment. Timber arrives in large shipments every few months. A species can be plentiful this month and gone the next. Quoting without looking at live stock is the fastest way to lose credibility with B2B buyers: the quote goes out, then the warehouse turns out to be empty.

The business already runs a self-built ERP covering warehouse, customers, orders and lots. Everything the sales process needs is inside it. What was missing was a layer that reads that data every day and turns it into well-timed quoting actions.

The problem: quoting by memory, information scattered across chats

Working through the sales routine with the team surfaced three bottlenecks.

The first is passive quoting. Quotes were usually drafted only when a customer asked first. Meanwhile, the signal that a customer is about to reorder was already sitting in the ERP purchase history. Take a buyer who takes walnut roughly every six weeks, with the last purchase five weeks ago: that is exactly the kind of signal a system should be catching on its own. Nobody, and no system, scanned the full base of past customers every week looking for signals like that.

The second is manual stock checking. Before quoting a lot, a salesperson had to open the ERP and check what was actually available, in what quantity and dimension. Repeat that for every customer and every species, and the checking itself eats the time quoting was supposed to save. Many quote opportunities were simply skipped because there was no time to verify stock.

The third is scattered sales information. Customers write in over Zalo, email and phone. Lot photos, price lists and the technical specifications of each species live in the ERP, in Drive folders and on individual phones. To send a customer one photo of a lot, a salesperson often had to message a colleague and wait. This is the classic multi-channel sales problem in B2B: the channels are not missing, the layer that gathers and processes them in one place is.

The proposed scope: start with an automated quoting agent

The full proposal for this business covered six workstreams. The one chosen as the center of the sales side is the automated quoting agent, because it connects directly to revenue and runs on ERP data that already exists. Its scope has four linked steps.

One, read the purchase history of every past customer in the ERP on a schedule, and flag customers approaching their reorder cycle, based on the species they usually buy and the spacing between their orders.

Two, check stock before quoting. The agent only shortlists items that are actually in stock and match the recorded needs of that customer. Out of stock means no quote, even if the customer is due.

Three, draft a personalized quote from an agreed template: item, dimension, price, with real lot photos pulled from the ERP. Each customer receives a quote about the items they care about, not a generic price list blasted to everyone.

Four, send the quotes over email and Zalo twice a week, and stop there. When a customer replies, the conversation hands over to the salesperson who owns that account. The agent does not negotiate.

Alongside the quoting agent sits an internal assistant for the sales team. A salesperson types a question into one chat window and gets lot photos and product images from the ERP, document links, price lists and the technical specifications of each species. Frequent questions about products, dimensions and company policy are answered the same way. The operating goal is concrete: a salesperson standing in front of a customer should never wait on a manager or a colleague for a lot photo.

The same assistant carries a second, quieter job: onboarding. New sales staff learn the catalogue by asking it, species by species and dimension by dimension, alongside the company policies and process documents it can fetch. In a trade where product knowledge is mostly oral tradition held by senior staff, a searchable internal memory shortens the time a new hire needs before they can answer a customer alone. It also means the knowledge stays with the business when experienced people move on, instead of leaving with them.

In a later phase, the same agent layer extends to the website chat, so new customers asking about species, dimensions and availability get answers grounded in the business’s real data instead of a generic chatbot script. This is the same division of labor described on the AI sales agent page: the agent takes the repetitive work with clear data behind it, people take the work that needs judgment.

One quoting cycle, step by step

Step What happens Who does it
1 Scan ERP purchase history, list customers nearing their reorder cycle AI agent
2 Cross-check live stock per item of interest, drop items that are out of stock AI agent
3 Draft a personalized quote: item, dimension, price, lot photos AI agent
4 Send over email and Zalo on a twice-weekly schedule AI agent
5 Customer replies, asks more, or negotiates Customer
6 Take over the conversation, advise in depth, close the order Account salesperson

This table doubles as an operating checklist for one quoting cycle: a sales team can audit a real sending round against these six steps to find the step its current process is skipping.

The detail worth noticing is that step 2 comes before step 3. Many automated quoting setups fail because they generate quotes from a static price list. Here the quote is produced only after stock has been checked at the moment of quoting, so what the customer receives is an offer the business can actually fulfill. With lot-based goods like timber, that detail decides whether the automated channel earns trust or burns it.

A second, internal cycle runs in parallel: sales asks the assistant, the assistant fetches from the ERP and Drive and answers inside the same chat. A lot photo no longer travels through three people before it reaches the customer.

Why not automate everything at once

The complete proposal also included agents for purchasing and inventory, document handling with OCR, email processing and an automatic end-of-day report. The roadmap, though, was split into three phases over roughly six months, and the customer-facing sales work was placed in the final phase, once the data foundation and the API connection had settled through the internal agents.

That order is a deliberate choice. The quoting agent touches customers, so its mistakes are the most expensive kind: a quote built on wrong stock, sent straight to a buyer, destroys trust faster than any internal error. Putting it last, when the ERP data has already been exercised and checked by the internal agents in daily operation, is how the riskiest surface gets de-risked.

There is also a data reason for the order. A quoting agent is only as good as the stock and purchase records underneath it. Running the internal agents first pushes those records into daily use, which exposes their gaps early: duplicate customer entries, lots logged under inconsistent names, missing purchase dates. While the internal agents are the only readers, an error costs an internal correction. By the time the quoting agent reads the same data, it has already survived months of real questions from the sales team itself. The lesson generalizes to any business bringing AI into sales: automate the internal work first so the data is clean, then open the customer channel. The AI sales chatbot page walks through the same staging logic from the chat side.

Boundaries that were agreed up front

An honest note on this case: this is a plan at proposal stage. Everything described is a recorded problem and a proposed scope, not an operating result. The article states no improvement percentages, because there is no running system to measure yet.

Three boundaries were fixed in the proposal itself.

First, AI does not replace the salesperson in negotiation. Prices in automated quotes follow an approved frame. Bargaining, special payment terms and long-standing accounts that need personal care all go to the account owner.

Second, the agent only quotes what is in stock, and only items matching a customer’s recorded needs. Quoting the wrong item, or an item that has run out, counts as an operational failure, not a minor glitch.

Third, customer data and per-customer pricing are sensitive. A personalized quote travels to that customer only, over their own channel, and the whole flow runs on the business’s ERP data inside the system that already governs it.

Who has this same problem

This pattern is worth studying if your business has all three of these signs: customers reorder on a cycle that purchase history can predict; pricing is personalized per customer rather than one public price list; and the ability to quote depends on real stock at the moment of quoting. Wholesale distribution of materials and lot-based B2B goods usually has all three.

The reverse also holds. A retail business with heavy walk-in volume and public list prices barely has a quoting cycle to automate, and its AI sales agent should start at the consulting touchpoint instead of proactive quoting.

When a system like this does go live, the honest way to judge it is not the number of quotes sent. It is the reply rate per quoting cycle, the share of quotes built on stock that was actually available, and how often salespeople still had to rebuild a quote by hand. Those three measures separate an agent that sells from one that merely sends.

Frequently asked questions

How is an automated quoting agent different from blasting a price list to past customers?

The difference is the order of work: the agent reads each customer’s purchase history to flag who is nearing their reorder cycle, checks live stock on the items that customer cares about, and only then drafts a quote for that customer alone. A blasted price list sends the same content to everyone, with no view of buying cycles and no view of stock at the moment of sending.

Does the AI agent negotiate prices with customers?

No. Prices in automated quotes follow an approved frame, and once a customer replies to negotiate, the conversation hands over to the salesperson who owns that account. Special payment terms and long-standing accounts that need personal care stay with the account owner as well.

Why check stock before drafting a quote?

Because lot-based goods can sell out completely between two shipments. A quote built on stock that is actually available at the moment of quoting is an offer the business can fulfill; quoting first and discovering an empty warehouse afterwards is the fastest way to lose credibility with B2B buyers.

What data does a business need before it can quote on the buyer’s cycle?

It needs the purchase history of past customers and stock data that a system can read in a structured way. In this case both already lived in the business’s self-built ERP, and the whole scope connects to that ERP through its API instead of replacing the running system.

With no results yet, how should a quoting system be judged once it runs?

Judge it on three operating measures: the reply rate per quoting cycle, the share of quotes built on stock that was actually available, and how often salespeople still had to rebuild a quote by hand. Those three separate an agent that sells from one that merely sends.

Read next

If your business also quotes from memory and scattered spreadsheets, the practical next step is a demo of one quoting cycle on your own data: the purchase history of your past customers and your current stock. Book through the demo page and the InCard team will run one full scan-and-quote cycle so you can inspect the output before deciding on scope.

The quoting workflow above sits inside a wider sales system. The leaks it plugs are the same ones listed in why businesses lose leads: slow first replies, quotes that wait for one person, and follow-ups that depend on memory. Between quotes, lead nurturing automation keeps dormant buyers warm until their next cycle, and the record everything lands in is the subject of our guide to AI CRM software. At the front of the pipeline, AI lead generation covers where new buyers come from in the first place.

For the human side of the same desk, the AI sales assistant scripts show what the agent says at each deal stage, and personalizing sales messages with AI explains why the same quote lands differently when it references the buyer’s own history. The full pattern is one instance of agentic AI for sales and marketing, and for owners sequencing the change, digital transformation for SMEs suggests starting where the payoff shows up first. Related reading is grouped under Sales and CRM and Marketing.


Tam Ho is the Founder of InCard, the agentic AI platform for small and mid-sized businesses in Vietnam. He works directly with operations teams to bring each chain of work onto the system, with a human keeping the checkpoint on the decisions that matter.

Get advice

Want to apply this to your business?

Leave your details and the InCard team will contact you within 24 hours to look at which job is worth automating first. No cost and no commitment.

Apply it to your business

Want to see this running on your own work?

Book a 30-minute demo — we run it on the channels and processes you already have.