Finding Value in Your Own Data: Pricing, Customers, and Assortment

Nikolay deliberately chooses accessible cases: ordinary businesses, information already held in CRM, ERP, accounting systems, or spreadsheets, and a bounded project with a tangible financial outcome. These are not projects to build a system that continuously makes decisions in real time. The first example is Motor City Industrial, a B2B distributor with a large parts catalog. Branches set prices independently, using their own understanding of the market. The company brought sales and pricing data together, established a more consistent approach, and adjusted prices incrementally while tracking results. This is different from increasing every price by the same amount: different products may need different changes. Alexander points out that the story comes from an ERP and pricing-module vendor. It illustrates the mechanism but does not fully disclose the effort required at each stage. The hosts’ estimates of possible costs therefore cannot be treated as a verified budget for another company.

The discussion of a one-off project reveals two distinct needs. Nikolay advocates putting prices in order once and demonstrating value; trying to build a perfect automated system immediately can prevent a company from starting. Alexander emphasizes durable rules because purchasing costs change, new products appear, and branches keep making decisions. A bounded analytical project can establish an ongoing management process. Much of the work may involve reconciling sources, checking transaction completeness, resolving contradictions, and getting employees to follow the agreed process. Andrey asks how enabling an existing module compares with a project that must first assemble data from several systems. The hosts also discuss language models as assistants for writing calculation code, while leaving people responsible for validating inputs and decisions. An easy calculation does not mean an organization is ready to use its output safely.

The second case concerns a logistics company comparing customer revenue with cost to serve. Support, transaction handling, returns, and employee time all enter the calculation. A customer who pays a lot may consume even more resources, a difference hidden by total revenue. Nikolay describes a customer base split into profitable, roughly break-even, and loss-making groups. The management question becomes which customers to attract and on what terms to continue serving them. A small company may be able to perform the calculation in Excel if it can estimate time spent honestly. Establishing that record and accepting an uncomfortable finding may be harder than the arithmetic. Andrey stresses observation and domain knowledge: someone must notice the problem and formulate a useful hypothesis before the spreadsheet exists. Having data does not, by itself, identify the most valuable question for the business.

The result does not justify automatically dropping every loss-making customer. Nikolay describes network effects: losing one group can also drive profitable users away. Alexander adds the complexity of an ecosystem where the same person uses several products. A loss on one product may be accepted in anticipation of future gains elsewhere, but that expectation can fail when external conditions change. The more elaborate the model, the more important it becomes to explain what a metric actually represents. Andrey returns to verification: an increase after a change does not prove the change caused it. Purchasing costs may have fallen, or other improvements may have occurred in parallel. Nikolay resists importing the full complexity of large technology companies into a small B2B business. Their disagreement remains useful: a simple first step reduces uncertainty, but watching revenue is not automatically a controlled experiment.

The third case applies the same reasoning to products. Besides purchase and selling prices, the analysis must consider storage, delivery, and spoilage risk. Removing every item with weak standalone profitability can still be a mistake: some products bring customers in and help sell the rest of the basket. The hosts discuss items bought together, products that attract visitors, and store placement. Familiar retail stories illustrate the reasoning rather than establish that any particular adjacency will increase sales. Andrey raises the question of substitutes: if a specific brand is removed and replaced by a more profitable alternative, will customers behave the same way? Receipt data reveals observed relationships but does not explain every motive. That sets a boundary for assortment optimization: a decision needs to account for the shopping journey as a whole, or a locally attractive calculation may hurt the store’s overall result.

Moving to online businesses broadens what counts as data. Andrey describes research into hotel search at trivago: observing users showed that they checked selected filters again in hotel details, photos, and reviews. The response he describes was to show a relevant review beside the search result, confirming the feature the user wanted. Alexander interprets this as building trust and reducing unnecessary steps toward a booking. He also explicitly distinguishes correlation from causation: stronger claims about an intervention require a deliberate verification design. Qualitative observation helps uncover why someone acts a certain way, while quantitative analysis helps establish how common that behavior is. Combining them can be useful, with the complexity of the research chosen to suit the decision. The discussion ultimately returns to action: existing information creates value when it helps people choose a change, explain it to those who will carry it out, and observe the consequences for the business and its customers.