I've noticed that many companies still set prices as if the market would be kind enough to wait until the next monthly meeting.
Costs change. A competitor launches a promotion. Inventory piles up in one category while another starts to run low. Some customers accept the price increase, while others cut back on volume. The information exists, but it’s scattered across the ERP, the CRM, sales reports, and various spreadsheets that only the person who created them understands—and sometimes that person isn’t even with the company anymore.
By the time a decision is finally made, the context has already shifted.
That delay explains the appeal of data science and artificial intelligence applied to pricing. It also explains many failed projects. The expectation is often to find an algorithm capable of revealing “the right price” and solving a problem that involves strategy, information, customer behavior, sales architecture, and execution discipline.
Technology can handle a level of complexity that no team should have to manage manually. But it can also automate a bad strategy with remarkable speed.
A disruptive pricing strategy doesn’t start with artificial intelligence. It begins when the company accepts that a single rule no longer works for its entire portfolio and that continuing to operate with averages, standard margins, and sporadic reviews comes at a real cost.
Competitive pricing without falling into the race-to-the-bottom trap
Raising prices by 4% across the entire catalog is not a strategy. It’s an instruction.
Applying the same target margin to products with different turnover rates, substitutability, competitive exposure, and relevance to the customer isn’t a strategy either. It’s simple to explain and easy to execute. The problem arises later: products capable of capturing higher margins end up undervalued, while others receive price increases that harm volume or competitiveness.
And this is where data science breaks that uniformity. It allows for analyzing products, segments, and transactions in greater detail; identifying relationships that disappear in an aggregated report; and estimating what might happen if a variable changes. AI adds the ability to learn from new results and generate recommendations at a scale that would be unfeasible manually.
The crucial question remains a basic one: What on earth does the company want to optimize?
Gross margin, revenue, market share, inventory turnover, segment penetration, or cost recovery all yield different recommendations. The price that maximizes revenue may not be the one that best protects the margin (in fact, it isn’t). The price that helps clear inventory may be counterproductive for a product that defines the price perception of the entire category.
One of the most costly mistakes is treating the catalog as a flat table: SKU, description, cost, price, and margin. That structure is useful for tracking products. It’s far less useful for understanding how prices should be set.
A pricing model needs dimensions. Product family, category, brand, channel, region, customer type, volume, purchase frequency, inventory, seasonality, product lifecycle, competitive position, and strategic function are just a few. Not all of them apply to every business.
The challenge lies in separating the variables that explain behavior from those that merely add noise.
Two products with similar costs can have completely different sensitivities. One might be easy to compare and have visible substitutes. The other may be protected by availability, technical specifications, brand trust, or switching costs. If both receive the same markup, the company is ignoring information that the market has already provided.
Some companies try to apply advanced models to every SKU from day one. The result is often insufficient data, unreliable recommendations, and endless discussions about exceptions.
You’ve likely already discovered that not all products require the same level of intelligence.
High-volume items with sufficient historical variation may be candidates for elasticity models. Products with few transactions, negotiated sales, or complex B2B conditions may require sensitivity models, business rules, signals from comparable products, or expert knowledge. Some must follow a competitive strategy. Others need to protect a minimum margin. Still others serve as key value items and must maintain an attractive perception, even if their individual margin isn’t the highest.
This is where a composite strategy comes into play.
Instead of choosing a single approach—cost-plus, competition, elasticity, or rules—for everything, the company assigns different methodologies by segment and establishes how they interact. The model may recommend a price, but a rule prevents it from falling below the minimum margin. The competitive signal guides positioning without forcing the company to match the competitor. Inventory can accelerate an adjustment within defined limits.
SYMSON, the pricing tool we use at ICX, incorporates differentiated strategies and business rules such as profit protection, minimum margin, price change caps, rounding, inventory review, and campaigns.
What’s disruptive isn’t changing prices all the time. It’s deciding precisely when to change them, by how much, and when to leave them alone.
Competitor information often triggers an automatic reaction. If someone lowers their price, there’s pressure to lower yours. If someone raises their price, an opportunity to raise yours arises.
A competitor might be clearing out inventory, using a product as a loss leader, or simply making a mistake. Copying them doesn’t make the decision scientific.
Competitive data works best as a signal within a broader model. The company needs to know which competitors are comparable, which products are sufficiently equivalent, what its desired position is, and how much weight to give to brand, service, or availability.
Product matching also requires caution. Comparing similar descriptions without validating format, presentation, or attributes can lead to absurd recommendations. AI helps scale matching and monitoring, but business judgment defines which comparisons make sense.
Elasticity attempts to estimate how much demand changes when the price changes. It seems like a direct relationship until promotions, seasonality, stockouts, competitor actions, and channel shifts come into play.
A historical correlation does not always represent an actual customer response.
If the price has hardly ever changed, there will be little to learn. If all price reductions occurred during high-traffic campaigns, attributing the increase in volume solely to price would be naive. If the product was out of stock for weeks, the observed demand does not represent available demand either.
Price sensitivity allows you to work with a broader set of drivers, especially when historical elasticity is insufficient. It can incorporate product attributes, competitive signals, customer behavior, positioning, and expert knowledge to estimate how vulnerable an offer is to a price adjustment.
SYMSON distinguishes between these two approaches and proposes using elasticity algorithms for products with sufficient volume, while sensitivity analysis allows for defining drivers and calculating scores to guide pricing. Its approach combines economic models, machine learning, and business rules, rather than assuming that a single formula solves every case.
For a CFO, this distinction determines how much to trust the recommendation, what level of risk to accept, and what evidence to require before scaling up.
Appian Updates Its Platform with Artificial Intelligence
An optimal price that no one can defend to Sales, Finance, or Management ends up being just an interesting number on a screen.
Interpretability matters because every recommendation affects a business relationship. A sales director needs to know whether the price increase is due to lower sensitivity, greater differentiation from the competitor, improved availability, or a change in the product mix. Finance needs to understand the expected impact on margin and revenue. The pricing team must identify which constraints were applied and how confident the model is.
There’s no need to turn every user into a data scientist. But we do need to show the economic and operational logic behind the recommendation.
The SYMSON algorithm detects patterns on a scale that would be impossible for a person to achieve. On the other hand, expert judgment recognizes events that do not yet exist in the data, such as the entry of a competitor, an exceptional negotiation, a campaign, or a branding decision that will sacrifice margin on certain items. That is why excluding that knowledge would be just as unwise as ignoring the data.
Many initiatives produce valuable analysis but fail when it comes to execution. The team calculates elasticities, designs segments, and defines rules. Then it tries to manage everything with spreadsheets, manual data entry, and email approvals.
The methodology exists. The process doesn’t scale.
A specialized platform must connect internal and external data, execute differentiated strategies, simulate scenarios, enforce constraints, log approvals, and monitor results. Traceability matters: who changed a rule, why a recommendation was accepted, and what happened next.
The SYMSON tool brings together features to build strategies, monitor competitor prices, gain insights, make predictions, and combine algorithms, machine learning, and business rules. Its approach incorporates variables as diverse and complex as geography, economic conditions, competition, inventory, and costs.
The difference between an analytical tool and an operational tool becomes apparent after the first model. The former answers a question. The latter allows you to repeat the decision, learn from the result, and maintain control over hundreds or thousands of prices.
Waiting for flawless data is a polite way of saying you’re not going to start. Loading everything available and trusting that the model will find something useful is a more modern way of losing control.
A reasonable implementation starts with a portion of the portfolio where there is a visible opportunity, sufficient data, and the ability to observe results. It must have a specific financial objective and clear boundaries: improving margin in a category without exceeding a certain volume risk, correcting undervalued products, or reducing unnecessary discounts.
Next comes the less glamorous but labor-intensive work: cleaning up transactions, separating list prices from actual prices, identifying promotions, reviewing costs, and validating hierarchies. There’s no need to overhaul the company’s entire data architecture. What matters is that the dataset used accurately reflects business reality.
Next, the portfolio is segmented, a strategy is assigned to each group, and guardrails are defined. The model is tested using historical data and against business criteria. A statistically elegant recommendation may be commercially unfeasible.
Before automating, it’s best to operate in recommendation mode. The team compares suggestions with current decisions, documents discrepancies, and learns where the model works well. Automation can be scaled up later, starting with low-risk, high-frequency changes. There’s no need to rush this.
Finally, measure margin, volume, revenue, product mix, acceptance rate, frequency of overrides, competitive reaction, and stability. A system whose recommendations are ignored by the team has a problem, even if its charts are flawless.
Pricing often gets caught between two approaches. Finance wants to protect margins and control exceptions. Sales wants to preserve volume, competitiveness, and relationships. When each department operates based on its own reports, the discussion boils down to conflicting intuitions.
Data science makes these “trade-offs” visible. How much additional margin is expected? What volume is at risk? Which segments show the least sensitivity? Which products are out of position? What portion of the increase can be captured without compromising a strategic account?
AI doesn’t eliminate disagreement. It makes it more specific.
The CFO should demand governance, traceability, and scenario analysis. The chief commercial officer should demand that the model recognize differences among customers, channels, and sales contexts. Both should be wary of any recommendation that promises to optimize everything at once.
There are decisions where it makes sense to accept lower margins to defend a position. Others where maintaining volume only prolongs weak profitability. The tool quantifies the trade-off. Management still makes the choice.
The partnership between ICX and SYMSON is based on a simple reality: a powerful platform alone cannot correct an inconsistent pricing architecture, and a good strategy loses its impact when there is no consistent way to execute it.
ICX provides the diagnosis, segmentation, goal setting, selection of methodologies, rule design, governance, and alignment with business and financial processes. SYMSON provides the technological capability to implement those decisions through elasticity and sensitivity models, competitive intelligence, forecasts, business rules, and monitoring.
This division of labor prevents the need to purchase technology only to later wonder what to do with it. It also prevents the creation of a model so ad hoc that it remains forever dependent on a few individuals and scattered files.
The implementation should result in an established capability: a process that learns, rules that can be reviewed, decisions that can be explained, and a team that understands when to follow the recommendation and when to intervene.
When an organization postpones its pricing decisions, the actual price does not remain static. It changes due to discounts, negotiations, inflation, promotions, payment terms, execution errors, and delayed responses to the competition.
The company may keep its list price intact, while the effective price deteriorates transaction by transaction.
That is the cost of continuing to operate as before. Visibility is lost regarding who captures value, under what conditions, and with what financial impact.
Disruptive pricing strategies using data science and AI allow for much greater precision, but they require sound judgment. You must set objectives, segment the market, select drivers, validate models, set limits, and learn from the results. Technology makes the system possible. The decision to abandon comfortable rules and misleading averages remains a human one.
And that’s probably the hardest part.