Impact Pricing · 21 Sep 2026 · From the week of 21 September
Why AI Gets Pricing Wrong and How to Train It to Get Pricing Right with Roberto Rivera
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Roberto Rivera, founder of Pricing Nerd and author of AI for B2B Pricing: Turning Value into Margin at Scale, joins host Mark Stiving to discuss why asking AI what price to charge gives poor answers. The conversation covers pricing skills that teach AI to interpret business context, deterministic and triangulated price recommendations, learning from human overrides, and where human judgment still matters. Rivera's central argument is that AI needs both business context and pricing skills, and that companies should start with one painful problem rather than waiting for perfect data.
For founders
- Structure AI pricing around deterministic inputs and formulas rather than letting the AI guess or search the web.
- Start with one painful, repetitive pricing problem, such as reconciling invoices to shipment records and contract clauses, rather than attempting a full overhaul.
- Audit signed contracts for clauses such as fuel-linked freight surcharges that allow repricing when costs change.
- Document the value your investments deliver to customers, since AI will not account for value that has not been captured and shared.
- Keep the final pricing call with people, because relationships, politics, sensitivities and accountability sit with the humans in the situation.
For revenue leaders
- Build pricing recommendations from several angles, such as formula results, the ten most recent closed deals, how the customer buys and how much value you have produced for that customer.
- Feed price overrides back into the system so future recommendations for that customer account for the adjustment.
- Let less experienced sellers use AI guidance so they do not have to pull busy experienced sellers into every pricing question.
- Ask the AI to show its justification for a price, which saves gathering the supporting pieces yourself and gives the system something to learn from when you override it.
- Use AI to combine marketing, CRM, ERP and contract data into one defensible price argument.
What was said 21, most useful first
Escalation clauses in signed contracts, such as fuel-linked freight charges, are often forgotten, leaving margin unrecovered. Listen
Rivera's example is an escalation clause that lets a company charge for freight or delivery if fuel prices rise, which he says is filed away after signing and never checked. He says AI can now read contracts to identify which clauses affect pricing, and that these clauses should be monitored continuously because failing to act on them leaves money on the table.
“that if the price of gas goes up to $4 a gallon, you should be able to charge for freight charges or for delivery charges”
AI pricing becomes trustworthy when it uses deterministic inputs, formulas and math rather than guessing or searching the web. Listen
He says that once AI has been trained with skills, business context and information, a price should come from a deterministic method of inputs, formulas and math. He argues computation is cheap, so the system can produce several perspectives that a team can evaluate. Whether a team accepts the output is still up to its own judgment.
“you don't let it when it comes to pricing right you don't let it guess you don't let it go and search the web and give me the answer”
AI can draft a value model by asking how a buyer's procurement would value an offering, drawing on public sources such as forums, industry publications and competitor websites. Listen
Rivera says AI can create a value model whether a company likes it or not, for value-based pricing purposes rather than deterministic pricing. He describes it searching forums, industry publications, the company's website and competitors' websites to rank the offering from a procurement perspective and surface value drivers such as less rework, less downtime and less waste. If the AI shows the price-value relationship as off relative to competitors, the company has to decide whether to communicate its value better.
“AI has the skill to help you get started down that path. It can create a value model whether you like it or not, right?”
AI only reflects value a company has documented, so improvements such as ERP-driven error reduction are missed until they are shared with customers. Listen
Rivera's example is a company that invested millions in a new ERP system that reduces errors by a percentage he did not specify. He says the AI would not capture this unless the company puts that value on paper and shares it with customers as part of its value-based pricing process.
“And if the AI is not capturing that, it doesn't understand that.”
People trust AI more on topics they do not know, even though they can see its errors on topics they do know. Listen
Mark Stiving says AI is wrong on pricing questions he knows the answers to, yet he believes it on unfamiliar topics such as a skin rash, which he calls dangerous. He says he tries to use AI as a first-pass answer and does not necessarily trust it.
“I know that it's wrong because I happen to know the answers.”
AI struggles with pricing because pricing professionals do not agree on what pricing means or what the right process is. Listen
Mark Stiving puts forward a theory that AI gives poor pricing answers partly because it has no clear definition of pricing to work from. He points to pricing professionals who write and talk about pricing and disagree on definitions of terms and on the right process. He says AI can only read what has been written and try to make sense of it, and he doubts anyone could make sense of it.
“AI doesn't really know what pricing means. And I say that because as I look at all the different pricing professionals, all the different people who write about pricing and talk about pricing, we don't agree what pricing means.”
AI needs pricing skills, meaning instructions for interpreting business information, on top of business context. Listen
Rivera says context about the business has to be provided in parallel with skills, which teach the AI how to interpret the information it has access to. He says skills are what let the AI interpret the business well enough to produce solid price guidance, rather than just knowing facts about it.
“Skills is a massive component of giving the AI the power to not only know about your business, but interpret your business in a way that's going to produce good and solid price guidance.”
Rivera organises AI pricing skills along a value-to-pocket operating system, with a skill set needed at each stage from value to list price onward. Listen
Rivera says his book breaks the approach into what he calls the value to pocket operating system. At each stage, from value to list price, a skill or set of skills is needed to help the AI understand value positioning and quantify value, drawing on economic value estimation and customer value modeling, which he credits to Tom Nagle's work. He describes the full set as maybe 30 to 50 pricing skills.
“And through every stage from value to list price, there's a skill or there's a set of skills that need to be there to really help the AI understand your value positioning in the marketplace, how you quantify value.”
Humans keep the final price decision because they weigh relationships, politics, sensitivities, risks and accountability that AI cannot replace. Listen
Rivera's example is a company that has been late on three shipments in five months, costing the customer millions, which affects how much value to capture in a particular negotiation. He says AI can compile information and tactics on positioning, but the human in the situation goes with their best judgment for the company and the customer.
“But at the end of the day, the human and the situation they're in, they're going to go with their best judgment”
A price recommendation can draw on several angles at once: formula-based prices, the most recent closed deals, and how the customer buys and the value produced for them. Listen
Rivera gives an example of one formula-based price, another formula-based price, support from the ten most recent deals closed, and support from how the customer buys and the value delivered to them over the last six months. He says the seller, commercial team, pricing team or deal desk should be able to trust this briefing.
“It's going to give you the price based on this math is X, the price based on this math is X, and it's also supported by the 10 most recent deals that we closed.”
When a human overrides an AI price, the system can learn from the adjustment so future recommendations for that customer account for it. Listen
Rivera says if a recommendation is overridden, the learning system built into the AI kicks in and the AI remembers that the recommendation did not fly for that customer. He hopes it will then account for some of that adjustment next time, which is a stated aim rather than a demonstrated result.
“So the AI is going to learn that this particular recommendation for this customer didn't fly last time.”
The reasoning behind an AI price is valuable even if you ignore the price, because it shows you why it was set. Listen
Mark Stiving says he liked that the AI gave the entire justification for the price, so he does not have to search for the relevant pieces himself. He imagines overriding a price for a personal reason, such as giving a brother-in-law a deal, and says the AI could then ask why and learn from the answer.
“It actually gave me the entire justification for why it created that price.”
AI can pull marketing, CRM, ERP and contract information together to build a defensible price argument. Listen
He describes AI as a very smart new hire who learns the business quickly and gathers information from the marketing, CRM, ERP and contract sides of the company. He says the result is good context and arguments for why a price is defensible, while the system stays open to learning.
“It can now gather information from the marketing side of things CRM side of things ERP side of things contract side of things Put it all together in one place”
Less experienced sellers can ask AI for pricing guidance without the reputational cost of bothering a busy experienced colleague. Listen
Rivera says new sellers may hesitate to take questions to a busy experienced seller down the desk, but can surface guidance from AI and ask it to explain its thinking. He says asking AI costs no reputation and involves no judgment.
“The other piece I love about AI in this context is that it doesn't judge you.”
Listen to the episode Sales team, hiring & comp Link to this
Start AI pricing work with one painful, repetitive problem, such as reconciling invoices to shipment records and contract clauses, rather than trying to boil the ocean. Listen
Rivera advises against trying to do everything at once and suggests taking one piece that is really painful. He gives reconciling invoices to shipment records and contract clauses, which produces reconciliation variances, as a candidate that can be automated with AI.
“So I would encourage people to go for it, you know, just take a piece.”
Getting contracts and invoicing right is a low-hanging target for AI, because errors force re-invoicing and rebilling customers. Listen
Rivera says the book includes a couple of chapters on this kind of problem and that fixing contracts and invoicing would save time on finance and pricing headaches. He describes it as low-hanging fruit with a lot of value.
“if we can just get contracts, we just get invoicing out the door better”
Experienced pricers can train AI better because they know what good looks like from trying and failing. Listen
He describes the scar tissue from implementing approaches that did not work, which lets experienced pricers recognise when an AI suggestion is unlikely to work. He says training AI takes time that not everyone has, and encourages the next generation to still learn what good looks like by making the mistakes experienced pricers made.
“And we know what good looks like, right?”
Rivera has published 48 pricing skills on his website that he says any good pricer should have. Listen
He describes these as skills any good pricer should keep in their back pocket and know how to apply, and says they can be built into AI as well. He puts the broader range of pricing skills at around 30 to 50.
“I have 48 in my website pricing skills that any good pricers should have in their back pocket”
Value-based pricing and willingness-to-pay methods are skills humans in pricing should learn, not just AI. Listen
Rivera says the best-practice frameworks for setting prices, such as value-based pricing and using big data to calculate pricing power and willingness to pay with model testing, are skills pricing people should know and apply. He points to PPS, forums, podcasts and articles as places to learn them.
“Whether it's value -based pricing, whether it's using big data to calculate pricing power and willingness to pay, and testing those models.”
Messy data, complex businesses and internal politics are excuses that keep companies from using AI for pricing. Listen
He lists the reasons companies give for holding back: our data is too messy, our business is too complex, our people are not on board and it is too political. He says these should not stop experimentation.
“companies say our data is too messy. Our business is too complex.”
Rivera expects that the context and structure AI needs about a business will stay useful however fast the underlying models change. Listen
Asked whether his book could become obsolete as AI changes, he says he worries more about how to use the changes well than about the technology itself. He expects context windows to grow and memory to improve, and says he will try to keep up with those changes.
“I don't think that changes. I don't think that changes regardless of how quickly that technology changes.”