How to choose price monitoring software without a bad year

Every price monitoring tool demos well. They all show a clean grid of competitor prices on a catalog the vendor has already curated. The differences that decide whether you are happy in month six are invisible in a demo: how accurately the tool matches your products, whether it covers the competitors you named rather than the ones it finds easy, whether it tells you when it stops working, and what the bill looks like once your catalog and refresh rate are real. Score those explicitly, with weights, before you watch a single screen share.

Four questions that eliminate most vendors before a demo

Send these by email. The replies cost a vendor an hour and save you a week.

1. 1Here are 50 of my SKUs and 5 competitor domains. Match them and send the results before our call. This is the whole evaluation compressed into one task. A vendor who will not do it before a contract is telling you something.
2. 2What is my all-in price for year one, including onboarding, and what triggers an overage? A written number, not a range. If scoping happens after signature, you have found the pricing model.
3. 3Show me the screen that tells me a source stopped returning data. If there is no such screen, the tool cannot fail loudly, which means it will fail quietly.
4. 4Can I export my full price history and match map today, myself? Data portability is cheap to promise and expensive to discover you do not have.

How to choose price monitoring software: a weighted rubric

Copy this table. Score each vendor 0 to 5 on every criterion, multiply by the weight, and sum. The maximum is 500. The weights are the argument: most of the value sits on the two things hardest to fix after you sign, very little on what a vendor can add in a sprint.

A worked scoring example

Two hypothetical vendors, scored on the same rubric. Vendor A scores 5 on match accuracy, 4 on coverage, 3 on refresh, 5 on reliability, 2 on MAP evidence, 4 on integrations, 3 on cost, 4 on support and 4 on exit. Weighted: 100 + 80 + 30 + 75 + 20 + 32 + 30 + 16 + 12 = 395. Vendor B scores 3, 5, 5, 2, 5, 3, 4, 3, 2, which weights to 60 + 100 + 50 + 30 + 50 + 24 + 40 + 12 + 6 = 372.

Twenty-three points apart, and the shape of the gap is the useful part. Vendor B wins on everything visible in a demo: more competitors on screen, faster refresh, better MAP screens. Vendor A wins on the criteria you only feel in month four, when a competitor redesigns their site and one tool tells you while the other keeps drawing a chart from stale data.

How to test match accuracy before you sign

Match accuracy carries the highest weight because it is the input to everything else. An index built on wrong matches is a confident wrong number, and a repricing rule fed by them moves real prices.

1. 1

   1. Pick 100 SKUs that represent the hard cases

   Not your 100 easiest. Include pack-size variants, model-year successors, items where the competitor sells a bundle, private label near-equivalents, and at least ten SKUs with no GTIN published anywhere. A test set of clean electronics with clean UPCs tells you nothing.

2. 2

   2. Name five competitor domains, including the awkward one

   Include the competitor whose site loads price into a JavaScript component, or hides price behind an add-to-cart step. Every vendor covers the easy domains. You are buying the hard ones.

3. 3

   3. Ask for the matches with confidence scores attached

   You want the pair, the confidence, and the signal that drove it. A vendor returning matches with no score cannot tell you which ones to distrust, which means all of them need manual review.

4. 4

   4. Review every returned match by hand

   Open both pages. Count three numbers: correct matches, wrong matches, and SKUs where a match clearly existed but was not found. It takes about an hour for 100 SKUs, and it is the most valuable hour in the evaluation.

5. 5

   5. Compute precision and recall separately

   Precision is correct matches divided by matches returned. Recall is correct matches divided by SKUs that had a findable match. They fail in opposite directions, and a single accuracy number hides both.

Worked example. A vendor returns 84 matches across your 100 test SKUs. You review them and find 9 are wrong, so 75 are correct. Precision is 75 / 84 = 89.3 percent. You also determine that 92 of the 100 SKUs genuinely had a competitor offer to find, so recall is 75 / 92 = 81.5 percent. Now compare a second vendor that returns only 40 matches and gets all 40 right. Precision is 100 percent, which sounds better, but recall is 40 / 92 = 43.5 percent, so more than half your catalog has no competitive data at all. Neither number alone tells you which tool to buy.

Demo questions, and the answers that should worry you

Pricing models and how each one bites at scale

Four models dominate this category. Each is defensible, and each has a point where the bill stops matching the value. Knowing which model you are buying tells you which growth path is expensive.

Per product

You pay per monitored SKU. Adding competitors is free, adding products is linear. Example: 5,000 products at $0.10 per product per month is $500. It bites on a long tail you would like to watch cheaply, because the thousandth SKU costs the same as the first while earning far less. It does reward choosing SKUs deliberately, which is the right habit anyway.

Per URL, or per product-competitor pair

You pay per tracked link. Example: 2,000 products across 6 competitors is 12,000 URLs. It bites when you add a competitor: a seventh adds 2,000 URLs, a 16.7 percent increase in your bill for one domain, because 2,000 / 12,000 = 0.167. This model quietly discourages the thing that makes the data useful, watching enough competitors to have a market rather than a rival.

Per check, or per request

You pay per collection event. Example: 12,000 URLs at four checks a day for 30 days is 12,000 x 4 x 30 = 1,440,000 checks a month, and moving from twice daily to four times daily doubles that from 720,000. It bites hardest in volatile categories, exactly where you need frequency most. Ask whether failed requests and retries are billed, because on a flaky source they can be a large share of the total.

Flat tiers

You pay for a band of products and a refresh rate. Predictability is its whole value. It bites at the boundary: find out before signing whether exceeding the tier triggers an overage rate or a forced upgrade, and how much notice you get. Also check what the tier contains, because a flat tier with the MAP module, the API and extra users priced separately is a per-feature model in a flat-tier costume.

The costs that are not on the pricing page

• Onboarding or implementation fees. Ask what the fee buys and what happens if you do the setup yourself.
• Per-domain or per-marketplace charges. Check whether each marketplace counts as one source or one source per region.
• Module pricing. MAP monitoring, repricing and API access are frequently separate lines. Price the bundle you will use in year two, not the one you need in month one.
• Seat costs. A tool nobody outside the pricing team can open does not change behaviour. Ask what unlimited users costs.
• Overage. The rate per extra product, URL or check, and whether it charges automatically or triggers a conversation.
• Historical data. Some contracts bill for backfill and some delete history at term end.
• The annual lock. Annual billing is cheaper, and it removes your only real recourse if the data quietly degrades in month five.

Where PriceIntelligence.io fits, and where it does not

It would be a strange buyer's guide that pretended we answered every requirement, so here is the honest version. We fit a mid-market US retailer or brand with a defined competitor set that wants monitoring, MAP evidence, rule-based repricing and analytics in one place without paying extra per capability. Our strongest arguments are the ones weighted highest above: a confidence score you can inspect on every match, a coverage and freshness view you read yourself, extraction that regenerates when a layout changes, and re-discovery of a moved product by identifier rather than URL.

We are the wrong fit in several situations, and it is cheaper for both of us to find out now.

• You want demand forecasting or econometric elasticity modelling from your transaction history. We do rule-based repricing with hard guardrails, and competitive analytics. We do not build elasticity models for you. Weight that criterion and look at dedicated price optimization platforms.
• You want a managed service. We are self-serve software. There is no analyst on our side running your program, writing enforcement letters or maintaining your competitor list.
• You need store-level pricing across thousands of physical locations. Our monitoring is built for online storefronts and marketplaces, not local shelf-price collection.
• You want a raw data feed with no interface. Our API and webhooks will feed it, but you would be paying for an application you do not intend to open. Weigh price scraping infrastructure instead.

A two-week evaluation plan

Trials fail when they are used as extended demos. Use the two weeks to generate evidence for the rubric.

1. 1Days 1 to 2. Load your 100 hard test SKUs and your five competitor domains, including the awkward one. Do not use the vendor sample catalog.
2. 2Days 3 to 4. Review every match by hand and compute precision and recall. Score the match accuracy criterion from your own numbers.
3. 3Days 5 to 7. Let it run untouched. On day 7, open the coverage and freshness view and check whether every source has been read on schedule. Score data reliability from what you see, not what you were told.
4. 4Days 8 to 10. Configure one alert rule and one repricing rule with a margin floor. Confirm the alert reaches the right channel, and that the rule refuses to breach its floor when you deliberately feed it a deep undercut.
5. 5Days 11 to 12. Export price history and the match map, self-serve, and time it. This is your exit rehearsal.
6. 6Days 13 to 14. Fill in the rubric from your notes, calculate the weighted totals, and write one paragraph on where the leader loses points. If you cannot write that paragraph, you have not tested hard enough.

If you are running this evaluation before you have a monitoring process at all, build the process first. The competitor price monitoring guide covers competitor selection, SKU selection, refresh frequency and alert thresholds, and it will tell you what plan size you actually need.