LevelSetter: Building An AI-Augmented B2B Pricing Platform

TL;DR: LevelSetter started in 2014 as an internal tool with one job, which was to get pricing work out of spreadsheets before a rounding error cost a client something nobody could undo. It took until 2021 to work properly. The turning point was a bake-off I lost to it, and what it settled was not that software prices better than practitioners do. It settled which half of the work is mechanical search and which half is judgment that has to stay with someone who has priced things before. That division is what an AI-augmented B2B pricing platform is for. The search runs at machine scale, the frame and the final call stay human. Everything we do now, including the discipline we named Continuous Monetization that same year, comes out of that one result.

In 2014 I was coming off my own startup, driving down to Charlotte to interview for a job running another software company. On the way I reached out to SPP instead, and ended up taking the CEO seat here.

At the software company I ran before this one, we priced made-to-order interior and exterior products. Our margin was calculated inside our customers’ price to their own customers, so an error in our engine did not stay in our engine. It walked out into somebody else’s market and repriced their catalog. We tested that engine harder than the rest of the product combined, and I still checked it.

Then I arrived here and found spreadsheets. Not a spreadsheet problem, a spreadsheet civilization. Spreadsheets referencing spreadsheets, tabs that existed only to correct other tabs, and Excel formulas written by people no longer employed anywhere in the building.

The Rounding Error That Changed Everything

Early on we found a rounding error. We caught it in the same cycle it appeared, so the damage was small and correctable. A few more cycles and it would not have been either.

That is the part that stayed with me. Not the error, which was ordinary, but the catch. We caught it because somebody happened to look. There was no structural reason we would have looked, and no reason to believe the next one would surface the same way.

We had templates. We had checks. What we did not have was traceability. If a number was wrong three engagements back, I was not confident we could walk it backward through the nest of spreadsheets and legacy systems to find out where it had gone wrong, or which clients had already acted on it.

Why pricing errors haunt every spreadsheet-based engagement

Right then it occurred to me that this was not our problem. It was everyone’s problem. Every software leader running pricing off spreadsheets is carrying the same unlit risk, and most of them know it.

It is the feeling of skating on ice you have not tested. There is a hard, tangled problem in front of you, and every quarter you find a reason not to open it, until a renewal or a board meeting opens it for you.

First Swings At A Really Hard Pricing Problem

That was where LevelSetter started. My first stated goal as CEO was to eliminate Excel from our own work entirely. I had no idea what I was asking for.

The original 2014 vision: one pricing data format for every client

We began the way everyone begins, by trying to get customers to hand us data in a single format. Simple enough on paper. It was not simple at all. Every new client arrived with oddities that broke the format: product structures nobody outside their company had seen, pricing logic that lived in a quoting system rather than a price list, and history that had been merged in from an acquisition without being reconciled.

We came close to concluding it was too variable to solve. Too many shapes, not enough pattern.

We did not back off. The answers were in the detail, so a format that could not carry the detail was never going to produce a result worth having, and a result that is not worth having is worse than no result, because somebody acts on it.

Why discount disguises make pricing data hard to normalize

Isolating discounts alone took years. Free months. Bundles carrying a discount inside the bundle rather than on the line. Caps on modules. Free editions that were discounts with a marketing name. Global pricing that reads as a discount and is not. If a deal in APAC closes at what reads as a ninety-nine percent reduction, is that a discount, or a rep correcting for a currency gap nobody built into the pricebook?

Discounts actively disguise themselves inside software companies. They are recorded by whoever needed them approved, which is rarely the person who will later need to find them.

The dirty data underneath was its own problem. Acquisitions merged without reconciliation, list prices overwritten rather than versioned, product dependencies that existed only as convention, refresh rates that varied by system. All of it had to be handled before any of it could be modeled.

Moving the pricing logic out of spreadsheets

Then the harder question. A pricing practice runs on logic that lives in the judgment of people who have done it many times. How much of that could move into software, and which part had to stay with a person?

That was the vision, and it took until 2021 to get there. It required back-end workflows, purpose-built tooling, and a genuinely deep understanding of how a client’s revenue model, downstream systems, and SKU structures interact, because a change in one spawns changes in the other two.

I do not know how many design sessions I sat through. I know I was waking up having dreamt about one problem or another. Small wins, then a larger loss when an approach did not survive contact with a real client’s data. We would take data in under the confidence of a new format and conclude, weeks later, that we had to customize it anyway. Cost overruns made a steady argument for abandoning the whole thing.

2018, And The Pricing Platform We Could Not Buy

By 2018 we had enough of the puzzle to commit. We decided to build a pricing platform properly, and to become, at least in part, a software company in order to do it. That was not a side bet. It changed what the firm was.

We looked hard at buying instead of building. The pricing software available then, and much of it now, is built to execute a pricing decision rather than to make one. Billing systems rate and invoice. Quoting systems assemble and approve. Neither is where you decide what the price attaches to, and a tool that assumes the decision is already made cannot help you make it.

So we built, and we rebuilt what is now LevelSetter several times over that stretch. Each rebuild settled something the previous one left ambiguous, and the design grew quieter each time, which is usually the signal you are converging on the actual problem rather than a description of it.

Setting LevelSetter to beat us at finding the best price

The goal we set was specific and slightly uncomfortable: LevelSetter had to beat us. Not match us. Beat the people in this firm at finding the strongest pricing answer, and analyze detail faster and more accurately than any formula we could write in Excel.

Excel was no way to live and we were still living it. Every analysis wanted a different spreadsheet. Our methods had already outgrown Excel, and we were modeling database structures inside a spreadsheet, forcing it into shapes it was never built to hold. By 2015 I had bought a workstation for the sole purpose of being able to open client spreadsheets at all, and a single engagement would generate dozens of versions of the same workbook as data was cleaned and re-cut.

When you build a pricing model, the answer space is enormous. When you are working it by hand, you know with certainty that you are not exploring it. You are exploring the part of it you have reason to trust, and then you stop, because the engagement has a date on it. Practitioners will concede this if you ask them directly. You get close enough, you run out of calendar, and you ship. What nobody can tell you is how good the answer you shipped was, or how far away the better one sat.

Nobody is doing that badly. The calendar is doing it. Once we treated it as a search problem rather than a judgment problem, it became something software could take a real bite out of.

The Bake-Off, And What It Settled

LevelSetter began as a system for us, built to take the labor out of model development and retire manual analysis.

The bake-off: LevelSetter against our own practitioners

When LevelSetter ran its first polished optimization routines, we pointed it at our own practitioners. I competed against it directly, on an engagement I knew well, with the best model I could build. Up to that point I had been winning these.

I lost, and not narrowly. LevelSetter returned a structure I had not considered, and when I checked it against the client’s real deal history it held up better than mine did. We then pointed LevelSetter at past engagements, and it kept finding stronger answers there too, including patterns none of us had known to look for.

I want to be careful about what that proved, because for a while we told the story wrong.

What the result divided

What I lost was a search. I had explored the region of the option space I had reason to trust, because that is what a practitioner does with a deadline in front of them. LevelSetter explored the rest of it. Those are two different capabilities, and discovering they were separable was the most useful thing that ever happened to this firm.

The other half did not move at all. LevelSetter did not decide which outcomes to optimize for. It did not know that the second-largest account was mid-renewal, or that the packaging structure it favored would strand a channel partner, or that the value metric it priced against was one the client’s own finance team could not yet count. Somebody had to set the frame before the search meant anything, and somebody had to read the result against a business afterward and decide whether it could survive contact with real buyers.

That somebody has always been a person, and after the bake-off we stopped pretending otherwise. The difference between AI-driven and AI-augmented pricing is exactly this line. The engine is built to scale judgment, not to replace it. A confident wrong answer in pricing does not get unwound, and only someone who has priced things before can tell a confident wrong answer from a right one.

What it changed about the work

It was not until Excel was permanently retired from our own work that we understood what we had built. The practical effect was that the thorniest mechanical part of our process stopped consuming the engagement. The hours we had spent on model construction moved to what clients always needed more of, which is interpretation, sequencing, and the argument inside their own executive team about what to do.

It also meant we could put LevelSetter in our clients’ hands rather than describing its output to them. Once you understand how the software supports pricing work, you know what happens when the people living with the decision operate it themselves.

LevelSetter Evolves, And The Slides Go Away

With the analysis solved, we pointed LevelSetter at delivery next. What stood between the work and the outcome was presentation labor. So. Many. Slides. The building, the polishing, the reformatting. Everybody likes a well-made slide and nobody likes what it costs, and the moment it is perfect the client asks for a different cut of the data and the cycle restarts at your expense.

The deeper problem was that slides served executives and abandoned operators. A deck explains where things are heading. It does not help the person who has to quote a deal on Monday. We needed outputs that were usable on day one rather than read on day one, which meant the output could not be a document at all. It had to be live state the client could open, question, and act on, with our people in the room while they did it.

Building a quoting engine for sales teams

Going after slideware meant building quoting. Not a generic quoting engine. Those exist, and most of them become dumping grounds of custom fields holding data nobody trusts.

We wanted something a salesperson would actually use, in the one part of the sales process that makes many strong reps uneasy, which is pricing itself.

Seeing the Sales Cycle More Clearly

Most reps genuinely like solving problems for customers and building relationships that last. The same reps hate it when a deal reaches the finish line and pricing turns it sideways. It puts them on defense without the pricing knowledge to hold the line, so they concede to save the deal, and the discount spiral starts.

We thought that was fixable. How do you make the economics of a deal a natural conversation rather than a confrontation? Could a rep walk a customer through pricing options as transparently as they walk them through a demo?

Helping reps see deal economics in real time

LevelSetter’s quoting capability started as a way for clients to feel their new pricing model rather than read about it. Structure the deal this way and here is the outcome, for the rep, for the customer, and for the company, visible while the conversation is still happening.

We wanted each deal legible as it evolved toward close, and all deals legible together, so patterns surfaced early enough to matter. A sales manager can interrupt a discount spiral in week three. Nobody can interrupt one in the end-of-quarter report.

From internal tool to something clients operate

Clients asked to put their hands on the wheel, which we had not planned for. That request produced the first LevelSetter API, and the surface kept expanding as customers embedded the capability into their own systems.

Today you can build applications on top of those APIs and run LevelSetter behind your CRM, gathering analytics on how customers and salespeople interact with packaging and pricing before a deal is won, across every channel in the business. That feeds directly back into packaging and pricing improvements. For companies boxed in by their existing systems, LevelSetter also extends what their pricebooks, SKU structures, and product dependencies can express, which is often the real reason a good pricing decision never reaches a customer.

The newer request is not for an application built on top of the API. It is for the decision layer itself to sit inside a workflow the customer already runs. A quoting assistant that drafts a configuration still has to know which combinations are coherent and what each one is worth. An agent that answers a pricing question in a sales conversation still has to answer it from the company’s own architecture rather than from whatever it can infer. A renewal workflow that proposes terms still has to respect the grant that was actually sold. In every one of those, the model supplies the language and the speed, and something else has to supply the pricing truth. That is the seat LevelSetter takes: it holds the current architecture and answers against it, so the workflow around it can move as fast as it likes without inventing the part that has to be right.

One thing does not change when a company does this. The judgment stays with people. An AI workflow wired to a live pricing architecture is faster and more consistent than one guessing, and it is still executing decisions somebody made deliberately. The instrument makes those decisions reachable at the moment of the deal. It does not make them.

The Pattern Library Underneath It

Every engagement that ran through LevelSetter left something behind. Not a client’s data, which stays theirs and stays put, but the shape of what we saw: how a value metric behaved once volume arrived, where a discount structure started leaking, which packaging boundaries held under pressure from a sales team and which ones collapsed the first time a large deal tested them. Conformed, stripped of anything identifying, and set beside every other engagement that ever produced the same shape.

That accumulation is the pattern library, and it is the part of this that could not be bought or rebuilt quickly. Software can be written in a year. A corpus spanning decades of real B2B software pricing decisions, with the outcomes attached, only exists if somebody was in the room for those decisions and kept the record.

It changes what the instrument can tell you. A tool with only your data can describe your history. It can show you what you charged and what closed. It cannot tell you whether the structure you are considering has been tried, what it did to renewal behavior two years later, or which version of it survived contact with a real sales organization. Patterns across a corpus answer questions a single company’s history cannot, because a single company has only run its own experiment once.

It also changes what the AI in an AI-augmented platform is doing. A model reasoning over a company’s own numbers is confined to what that company already did. A model reasoning against patterns across decades of comparable decisions is doing something different, and the frontier one is the second. That is also the fence: the library informs a recommendation, it never exposes another company’s specifics, and nothing a client contributes leaves as anything but an anonymized pattern.

What An AI-Augmented B2B Pricing Platform Is For

An AI-augmented B2B pricing platform is software that runs the mechanical half of pricing analysis, the search across an option space too large to work by hand, while every decision stays with a practitioner. Say it affirmatively, because the category is crowded with software that is not this.

It does not set your prices. It makes the search survivable so judgment gets spent where it earns something. LevelSetter runs the option space you cannot run by hand, holds the detail a spreadsheet drops, and keeps the answer current after the engagement that produced it has ended. The decisions stay with people, yours and ours, working from the same live state in LevelSetter rather than a document that started aging the day it was made.

Generative AI put a metered cost line underneath software products, so LLM inference now sits inside the margin of the thing you are pricing. Credits have appeared across the category as surrogate units standing in front of several value metrics at once, which obscures what a customer is buying. AI agents consume on their own schedule, which breaks the old assumption that a seat is a reasonable proxy for consumption. Each of those changes the arithmetic under a pricing model, and none of them wait for your annual review.

That is the case for operating pricing continuously rather than studying it occasionally, and if you want a read on where your own architecture is exposed, talk to a pricing expert and describe how you price today.

Looking Backward, To Keep Moving Forward

What I have enjoyed most is how the questions changed. “How should we view discounts” became “how should our clients be viewing, managing, and improving their discounts.” “How should we show the impact of the new pricing” became “how should a CFO review that impact and isolate what needs validating.” Each version pushed us further into how executive teams absorb a pricing decision. That is where most pricing work dies.

So the executive team works with the analysis directly now, inside LevelSetter, in sessions with our people rather than in a handoff. They see what the recommendation does to their own customer base before anyone commits to it.

Every one of those questions widened the product in the same direction, which is toward clients doing more of this inside their own teams, at their own pace. Pricing was always meant to be a standing capability, not a study commissioned every few years by a firm that leaves when the invoice clears, and not something reconstructed from a fresh set of spreadsheets each time somebody asks.

LevelSetter is a force multiplier for the people here. It strips the mechanical labor out of an engagement without costing quality, and I would argue the outcomes are better, because the practitioners are spending their time on the part that requires them. Deployed inside a client, LevelSetter becomes the place a team argues productively about pricing decisions and deal structures, with the same numbers in front of everyone.

A product is only an idea until you attach a pricing model to it. Then it becomes the vehicle that lets a software company grow profitably. And I will be unfashionable about profit. It stabilizes the business, reduces the need for outside capital, and builds an enterprise worth more than the sum of its bookings.

Why continuous monetization replaces project pricing engagements

Which means the model of expensive one-time pricing studies is finished. Pricing becomes a process a company runs, developed as the internal capability it should have been from the start, and the trust between seller and buyer improves because the price stops being an annual surprise. A company operating that way adjusts packaging and pricing before a competitor has finished writing theirs into the January kickoff deck.

A client recently called this agile pricing. We call it Continuous Monetization, we named it in 2014, the same year this instrument started, and we deliver it through LevelSetter. The two were always the same project.

We are working on harder problems now, mostly around isolating where customers actually perceive value in a product, in this release and the next one, so product managers and marketers invest in the right places and keep packaging and pricing moving with the value delivered rather than trailing behind it.

When pricing is treated as a core capability rather than a periodic panic, CEOs keep more control of their own company while still satisfying investors. That is not a story. We have done it alongside many software leaders, and the conversation is better had before your next pricing decision gets made under deadline. Talk to an expert and describe what your pricing has to survive this year.

FAQs