What is price intelligence, and how do teams use it?

Price intelligence is what you have when observed market prices have been made trustworthy enough to act on. Anyone can collect a number from a page. Price intelligence is the whole chain that makes the number mean something: knowing it is the right product, knowing when it was read, knowing whether it is plausible, and turning a pile of observations into a position you can describe in one sentence to a merchandising lead.

What is price intelligence, in operational terms

Strip away the category language and price intelligence is four questions answered continuously. What are comparable products selling for right now. Are those really comparable products. Can the numbers be trusted. What does the pattern say about our position. A team that can answer all four has price intelligence. A team that can only answer the first has a spreadsheet.

The distinction matters because the failure modes are quiet. A feed that silently returns a list price instead of a sale price throws no error; it moves your index a point and a half and nobody investigates. A match that drifts from a 4.0Ah drill kit to the 2.0Ah version reports a $40 undercut that never existed. Both look like data. Neither is intelligence.

What price intelligence includes, and what it excludes

Being clear about the right-hand column saves a lot of disappointment. Price intelligence is an evidence layer. It will tell you three competitors moved a hero SKU down between Thursday and Sunday and that your index slipped from 99 to 96. Whether to follow depends on margin, stock cover, elasticity and what you are trying to do this quarter.

Price intelligence vs price monitoring, price optimization and competitive intelligence

These four terms are used interchangeably in vendor material and they describe different jobs with different inputs and different owners. The clearest way to tell them apart is to ask what each one stops short of.

The sequence matters. Optimization built on unvalidated observations produces recommendations with the same errors as the input, dressed in the authority of a model. Fix matching and validation first. The product matching guide covers why that is harder than it sounds, and price monitoring covers the collection layer underneath.

The pipeline behind price intelligence

Five stages, each with its own failure mode. Knowing the stages is how you diagnose a number you do not believe, the most common request a pricing analyst gets.

1. Discovery and matching

Before anything is collected, each of your products has to be linked to the competitor listing that represents the same thing. A GTIN, UPC or EAN exact join is the strongest link and is tried first. Where identifiers are missing, which is most of apparel and much of private label, title and attribute embeddings narrow the field and image comparison breaks ties. Every match carries a confidence score, because a match asserted without one is a guess with good posture.

2. Collection and extraction

Where a site publishes schema.org Product structured data, read that first: the retailer maintains it because rich results depend on it, so it is the field least likely to rot. Where it is absent, a per-domain extraction recipe picks the price out of the page. Relying on either path alone breaks predictably.

The defences are structural. Read structured data first, so a layout change does not touch the primary path. Regenerate the extraction recipe automatically when a page shape changes, rather than waiting for someone to notice. Re-discover a product by identifier when its URL moves, instead of following a redirect into a category page. Alert on a source that degrades before the customer sees a gap. That is what data quality monitoring is for.

3. Validation

Every observation should pass a plausibility test before it is stored. A price 82 percent below the trailing median for that listing is more likely a decimal error, a per-unit price on a multipack, or an accessory from a related-products carousel than a real move. Validation quarantines such readings rather than deleting them, because the ones that survive scrutiny are the moves you most want to know about.

• Range checks against the product's own history, not against a global threshold, since a $19 accessory and a $1,900 appliance need different tolerances.
• Variant checks, because an aggregate offer with a low price and a high price will happily hand you the cheapest size in the range, and a per-100ml price and a per-bottle price are both real numbers on the same page.
• Cross-source agreement, comparing the structured data price with the rendered page price and flagging disagreement past tolerance.
• Stock coherence, since a price attached to an out-of-stock listing is a historical artefact rather than a competing offer.

4. Normalisation and storage

Observations become a series. Each reading carries its timestamp, its source, the seller where relevant, stock status, promotional context and the confidence of the match it belongs to. Storing the timestamp on the observation rather than the report is what lets you answer the question that matters in a dispute: what did this page say, and when.

5. Measurement

The series turn into measures. This is where a pile of numbers becomes a sentence somebody can act on. Be precise about how each measure is calculated: two teams using the same word for different arithmetic is a recurring source of pointless meetings.

The metrics price intelligence produces

Worked example, one product. Your price is $189.00. Four matched competitors are at $199.00, $204.50, $189.99 and $194.00.

• Price index. Your price divided by the market average, times 100. The four competitors sum to $787.49, so the average is $196.87. Your index is 189.00 / 196.87 x 100 = 96.0, meaning you are 4 percent under market on this product. An index of 100 is exactly at market. Choose mean or median and then never quietly switch, because the two diverge sharply when one competitor is clearing stock.
• Price position mix. The share of your catalogue that is lowest, matched and above. Across 240 monitored SKUs: 54 lowest is 22.5 percent, 96 matched within 1 percent is 40.0 percent, and 90 above is 37.5 percent. Mix tells you what kind of retailer you are being, which is often not what the strategy deck says.
• Volatility. Distinct price changes per SKU per week, by category. A category whose median SKU changes twice a week needs a different refresh cadence from one that changes twice a quarter. This number decides cadence, not a gut feel.
• Undercut frequency. The share of checks in which a given competitor is below you. One competitor below you 8 percent of the time is noise. One below you 61 percent of the time is a policy, and should be handled as one.
• Coverage and freshness. The percentage of monitored products with a successful reading in the last cycle, and the age of the oldest reading in the report. Every other metric here is conditional on these two.
• Margin impact. What a move costs before you make it. At a $132.00 cost, selling at $189.00 earns $57.00 a unit and selling at $179.00 earns $47.00. Holding gross profit constant needs 57 / 47 = 1.213, that is 21.3 percent more units. If you do not believe the volume lift will reach 21.3 percent, the price cut is a transfer from margin to shoppers.

Index and position mix are the two measures most often misread, usually because the comparison set is undisclosed. An index of 96 across a catalogue where only 40 percent of products have a confident match is not an index of 96. It is an index of 96 on the products that happened to match, a different claim. Price index and price positioning covers that in detail.

How a team actually uses it, week to week

Price intelligence used well is a short recurring routine, not a dashboard somebody opens during a crisis. The cadence below takes a pricing analyst a few hours a week in total and is the practical answer to what the software is for.

1. 1

   Monday: read the exceptions, not the dashboard

   Open the list of products whose position changed materially since the last review, and the list of competitor moves over your alert threshold. A dashboard shows you everything and therefore nothing. The exception list is the day's work, and on a normal week it is short.

2. 2

   Monday: clear the matching queue

   Confirm or correct the matches flagged as low confidence, and anything a colleague disputed. Fifteen minutes here protects every number produced downstream for the rest of the week, and each correction improves subsequent matching.

3. 3

   Midweek: investigate the moves that matter

   For each significant competitor move, check whether it is a promotion with an end date, a permanent reset, or a stock-clearing action on a line they are exiting. The response differs completely, and the price series plus stock status usually tells you which it is.

4. 4

   Midweek: work the review queue

   Repricing proposals from your rules land in a queue with the rule that fired, the guardrail that bounded it and the resulting margin. Approve, reject or adjust. Reviewing proposals is a far better use of an analyst than typing prices, and it keeps a human in the loop on the ones that matter.

5. 5

   Friday: read coverage and freshness

   Confirm the data behind the week was actually collected. A quiet week and a broken collector look identical in a report of price changes, and the only way to tell them apart is to look at source health directly.

6. 6

   Monthly: review index, mix and volatility trend

   Take the aggregate view to the commercial meeting: where the index moved by category, how position mix shifted, which competitors are moving most often. This is the conversation that changes strategy, and it needs a month of data rather than a week.

How to tell whether your price intelligence is broken

Four tests, none needing a vendor to run. Any one failing means the numbers in your pricing meeting are decorative.

1. 1Pick five products at random and check them by hand. If the number in the tool disagrees with the number on the page, the question is not whether the tool is wrong but how long it has been wrong and on how many products.
2. 2Ask what share of your catalogue has a confident match. If nobody knows, or the answer is a percentage with no confidence attached, every aggregate metric you are shown is computed over an unknown denominator.
3. 3Ask when each source last returned data successfully. A source that quietly stopped six weeks ago is still in your average, holding whatever value it last reported.
4. 4Check whether a promotion that ran last weekend is visible in the history. If it is not, your refresh cadence is slower than the market you are measuring, and short moves are invisible to you by construction.

If you are choosing a tool rather than fixing one, those questions are also the shortlist criteria, and how to choose price monitoring software turns them into a scoring sheet. To get the collection layer running this week, start with competitor price monitoring.