AI quoting tools apply customer-specific pricing in 2 ways. A CPQ suite applies prices, discounts and margin rules once every rule has been rebuilt inside it. An AI agent connected to your ERP reads the price agreements, discount groups and validity dates already stored there, and sends any quote below your margin floor to a person.
At Lleverage we think the pricing logic belongs in the ERP, as the one place the price is decided. The AI should read and apply it and keep no copy of its own. A copy drifts the first time someone updates a contract price in the ERP and forgets the other system.
This is the approach behind our AI agents for quotes and orders: the agent drafts the quote in your ERP, the ERP's own price logic fills the lines, and a person signs off what falls outside your rules. If you want to see that against your own price lists, book a demo.
Which AI quoting tools can apply customer-specific pricing?
Three kinds of tool claim to. CPQ suites hold their own pricing engine and apply customer-specific pricing once you load it. ERP-connected AI agents read the pricing that sits in your ERP and apply it per line. Generic AI assistants that draft a quote from an email apply no customer pricing at all unless someone feeds it to them.
The difference is where the rules live. In a CPQ suite, someone moves the price agreements, discount matrices and approval rules into the CPQ and keeps them in step with the ERP after go-live. With an agent connected to the ERP, the rules stay where finance already maintains them, and the agent looks them up when it drafts the quote.
AI in this kind of work is still new for most companies. Eurostat's 2025 survey found that 20.0% of EU enterprises with 10 or more employees used AI, up from 13.5% in 2024. Eurostat also found that analysing written language was the most common use, at 11.8% of enterprises. Reading a customer's request is that kind of task.
For a distributor or manufacturer with thousands of items and hundreds of customer agreements, a second rule set is a second job. Our CPQ for manufacturing post covers where a configurator still earns its place.
Where does customer pricing actually live?
At most distributors and manufacturers, customer pricing lives in 4 places at once: the ERP's price lists and contract prices, discount groups per customer and item group, spreadsheets kept by sales, and the memory of the 2 people who have handled the big accounts for years. Only the first 2 are visible to a system.
Most of these companies have an ERP to read from. According to Eurostat's 2025 survey of ICT use in enterprises, 41% of small EU enterprises and 89% of large ones use ERP software, the most widely used business application in the EU.
The ERP part is usually well structured. Microsoft's documentation for Business Central describes special prices set by any combination of customer, item, unit of measure, minimum quantity and the dates the price is valid, plus line discounts set the same way and invoice discounts above a minimum order value. Customer price groups and discount groups let one agreement cover many customers or many items.
The rest is less tidy:
- A spreadsheet of project prices that never made it into the ERP because the project was "temporary".
- A volume deal agreed by phone and honoured from memory.
- A rule such as "never below 18% margin on cable, unless the account manager says so", written down nowhere.
Any quoting system, AI or not, can only apply what it can read. The first job is to find which of these prices matter and get them into the ERP, or into a rule the team can see.
What goes wrong when a quote ignores the price agreement?
A quote that ignores the agreement goes out at the wrong price, and either the customer or your margin pays for it. Too high and the customer pushes back or buys elsewhere. Too low and the order arrives at a price finance did not plan for, which shows only when the invoice is posted.
The damage is usually a steady leak: a 5% discount on an item group the contract excludes, a volume break given on 80 pieces when the agreement says 100. Each quote looks fine alone. The pattern shows months later, in a margin report.
When the quote and the ERP disagree, someone also has to fix the sales order once the customer says yes: override the ERP, or call the customer to explain.
How does an AI agent apply customer-specific pricing per line?
An agent applies customer-specific pricing by building the quote inside the ERP, line by line, for the right customer. It matches the request to your customer record and your item numbers, adds each line with its quantity and unit, and lets the ERP calculate the price and discount from the agreements it holds. Then it checks the result against your rules.
Take Business Central. Microsoft's documentation explains that it calculates the best price when a unit price and line discount are added to a sales document line. It checks whether the bill-to customer, or a group the customer belongs to, has a price or discount agreement for the item or its discount group, whether the document date falls inside the agreement's starting and ending dates, and whether the unit of measure, minimum quantity and currency fit. The best price is the lowest price with the highest line discount allowed on that date.
A second copy of the rules tends to get that logic wrong. When the agent creates the quote in Business Central, the ERP runs its own calculation, so 400 pieces for a customer in a price group get that group's 400-piece price today.
The agent does the work around that calculation:
- It reads the request. Customers write "same cable as last month, 400 metres", or quote a competitor's code, and the agent matches that to your item and unit first.
- It picks the right customer. A forwarded email from a branch office has to land on the right bill-to customer, or the wrong agreement applies.
- It checks each line against your rules: a price below the floor, a discount above a salesperson's authority, a line with no agreement.
- Anything that waits for a person comes with the reason attached, so the reviewer sees why the line stopped.
The connection to your ERP decides what the agent can see and change: in Business Central, the connected account's permissions, with any action set to wait for approval.
How do margin floors and approval thresholds work?
A margin floor is the lowest margin a line or a quote may carry before a named person has to approve it. An approval threshold sets who that person is: a salesperson up to one discount level, the sales manager above it, the commercial director for anything below cost. The agent applies both, and holds any quote that crosses them.
The floor often lives in a policy document or a sales manager's head. In a setup with us, it is written down once as a rule your team can read and change.
When a draft quote falls below the floor, it stops. The agent hands the decision to the person who owns it, in Microsoft Teams, Slack or our request inbox, with the line and the reason it stopped. That person approves, changes the price or rejects, and the agent carries on with the answer. A request such as "have Mila approve the 8% discount" works the same way when a salesperson asks for it.
The agent also scores how sure it is of each piece of work. Below the confidence line a person decides, above it the work goes through on its own, and only you move the line. For quotes we advise keeping a person on the send for a long time, because a quote commits you to a price and a date.
"It’s a matter of building trust in the organisation with these kinds of initiatives. You can’t just throw something like this over the fence." Cees Maaskant, General Manager, Xpol
How should exceptions like one-off deals and expired agreements be handled?
Exceptions should stop the quote and go to a person with the reason stated. The common ones are a one-off deal outside any agreement, an agreement that has expired or overlaps another, and an item with no price for this customer. Each needs a different answer, and the agent should say which one it hit.
| Exception | What the ERP does on its own | What the agent should do |
|---|---|---|
| One-off deal agreed by a salesperson | Applies the normal agreement | Draft at the agreement price, flag the requested price, ask the owner to approve |
| Expired agreement | Stops applying it after the ending date | Draft at the price that applies today, tell the salesperson the contract has lapsed |
| Overlapping agreements | Picks the best price for the customer | Flag the overlap so someone cleans up the data |
| No special price for this customer | Falls back to the item's standard price | Flag that no agreement covers the line before the quote goes out |
| Item not found | Cannot add the line | Leave the line blank and ask the salesperson which item is meant |
The missing price is the one nobody notices. When no special price can be calculated, Business Central inserts the unit price from the item card. That is correct for the ERP, and it is how a key account receives a list price nobody meant to send.
Xpol, a flower wholesaler running Business Central, applies the same principle on order intake. Its agent works with 25 customer rulesets and leaves ambiguous fields blank for a person to fill in. A blank field often points at source data that needs fixing, such as a framework contract with overlapping validity dates, so each exception makes the next order cleaner.
What should you ask a vendor about ERP pricing?
Ask 4 things: does the tool read your pricing from the ERP or keep its own copy, does it show why it chose each price, can a person override a price and is the override logged, and what happens when no agreement applies. Test each on your own data, with a customer whose agreement you know by heart.
| Question | A good answer | A warning sign |
|---|---|---|
| Where does the price come from? | The ERP calculates it when the quote line is created | An imported price file refreshed each night or each week |
| Can you see why a line has this price? | Each line names the agreement, price group or discount it used | One total with no trace |
| Can a person override, and is it logged? | Overrides are allowed by role and recorded with who, when and why | Prices edited in a document with no history |
| What happens with no agreement? | The line is flagged before the quote goes out | The list price is used without comment |
| Where is the quote created? | As a sales quote in the ERP, ready to become an order | A PDF outside the ERP that someone retypes |
Bring 20 real requests, including a customer in a price group, an order just under a volume break and an item with an expired agreement. Count the lines priced right, the lines flagged and the lines wrong without a flag, because a wrong price nobody flagged is the one that reaches the customer.
CPQ or an ERP-connected agent for a large catalogue?
For a large catalogue with many customer agreements, an ERP-connected agent is usually the better fit, because it reuses pricing you already maintain. CPQ fits better where products are configured from many options and the price is built up from the configuration, often for a small number of high-value quotes.
| CPQ suite | ERP-connected AI agent | |
|---|---|---|
| Where pricing rules live | Inside the CPQ, copied from the ERP | In the ERP, read when the quote is drafted |
| Upkeep after go-live | Two rule sets kept in step | One, in the ERP |
| Input it expects | A salesperson selecting products and options | Requests as customers send them: email, PDF, Excel |
| Strength | Complex configuration and guided selling | High volume, large catalogues, many customer agreements |
| Approvals | Built into the CPQ workflow | Held by the agent and sent to the owner |
| Where the quote ends up | In the CPQ, often synced to the ERP | As a sales quote in the ERP |
Some companies need both: a configurator for engineered products and an agent for the catalogue lines. Our post on turning emailed requests for quotation into draft quotes covers the matching side in more depth, and sales order automation covers what happens once the customer says yes.
Frequently asked questions
Can AI apply customer-specific pricing from my ERP?
Yes, when it is connected to the ERP and creates the quote there. The ERP then calculates the price and discount from the customer's agreements, price groups, quantities and validity dates, as it does for a quote typed by hand. The AI reads the request, matches customer and items, and checks the result against your margin and approval rules.
What is a margin floor in quoting?
A margin floor is the lowest margin a quote line may carry without approval. Below it, a named person such as the sales manager decides whether the price stands. Floors are often set per item group or customer type. An AI agent holds any quote that crosses the floor and sends it to that person with the numbers.
Do I need CPQ software to apply customer-specific prices?
Not when your prices already sit in the ERP. Most ERPs used by distributors and manufacturers hold customer prices, discount groups, volume breaks and validity dates. CPQ software helps most with products configured from many options. For catalogue items with customer agreements, an agent that uses the ERP's own pricing avoids a second rule set.
What happens when a customer has no price agreement for an item?
The ERP usually falls back to the standard price. Business Central, for example, inserts the unit price from the item card when no special price applies. That may be correct, or it may be a gap in your data. A good quoting agent flags the line so a salesperson confirms the price before the quote goes out.
Where to start with customer-specific pricing
Start with 1 customer group and 1 mailbox. Pick the accounts with the most agreements and quote requests, check that their prices live in the ERP, and write down the margin floor and who approves below it. Then run a few weeks of real requests through a supervised agent and compare its drafts with what your team would have sent.
We set this up with your team in the ERP you already run, and you see every line the agent flagged and why. To see it on your own price agreements, book a demo.
