Bot Traffic IndexThe bot problem, quantified

Sep 15, 2026 - Bot Traffic Index

The Economics of Inventory Bots: Who Pays When Bots Buy First

On paper, a sellout is a sellout. If bots clear your limited release in ninety seconds, the revenue looks identical to a human sellout. The damage shows up everywhere except the top line.

The real costs

Customer trust. Genuine shoppers who repeatedly lose drops to bots learn that your store is not worth trying. That is a churn event that never appears in churn reporting. Data quality. Bot purchases pollute customer records, sizing curves, and demand forecasts - you start planning inventory for a buyer mix that does not exist. Support load. Drops followed by waves of 'order canceled' and 'why did this sell out' tickets cost real hours. Marketplace leakage. Your product immediately appears on resale platforms at a markup, training your best customers to shop there instead.

Why simple defenses fail

Rate limits and CAPTCHAs catch crude bots and frustrate real customers at exactly the wrong moment - during the drop. Modern purchasing bots route through residential proxies, rotate device fingerprints, and solve challenges faster than humans. Queue systems help fairness but do not stop automated buyers from joining the queue.

What works

The effective stack is behavioral: detection at the network edge that scores sessions on signals bots cannot cheaply fake, applied before the storefront ever sees the request. Blocking scraper and hoarder traffic at the edge also restores analytics accuracy as a side effect - fake sessions never reach your pixels. Post-purchase order screening is the second line, not the first.

The takeaway

Inventory bots are not a security footnote; they are a customer-experience and data-integrity problem with a revenue-shaped hole in the middle. Measure them, block them at the edge where possible, and stop counting bot demand in your forecasts.

Someone Is Scraping Your Prices Every Hour. Here Is How to Tell.

Price scrapers rotate through residential proxies, copy real browser strings, and request product pages in catalog order. Rate limiting alone will not stop them, and the signs are in the request pattern, not the user agent.

A competitor's pricing tool hits your product pages every hour, on the hour, in catalog order, from a thousand different residential IP addresses. Your analytics show the traffic as real shoppers, because the requests carry real Chrome user agents and residential IPs that look like households. Your server logs show the truth: nobody browses a catalog in alphabetical order at 3am, and nobody reads a product page for two seconds flat, a thousand times a day.

Price scraping is the most common bot problem on ecommerce, and the hardest to see, because the scrapers have learned to look human. The old signals, a datacenter IP, a python-requests user agent, a thousand requests a minute, catch the lazy scrapers. The serious ones rotate through residential proxies and behave just enough like shoppers to pass a casual look.

The pattern that gives them away

Scrapers are regular. They request pages in catalog order, or by product ID sequence, with consistent timing between requests, and they visit pages a human would never visit in that order, like every variant of every product, including the ones that have been out of stock for months. A human browses in bursts, with pauses, backtracking, and dead ends. A scraper browses like a list.

The second signal is what they do not do. They never add to cart, never reach checkout, never trigger the interactions a shopper triggers, and they request the price element on every page while ignoring the reviews, the images, and the related products. A session with a hundred product views and zero interactions is not a shopper. It is a harvest.

Why rate limiting alone fails

Rate limiting per IP assumes the scraper has a small number of addresses. A scraper running through residential proxies has thousands, and each one stays under the limit while the total volume stays high. DataDome, whose whole business is stopping this, puts it plainly: simple IP-based rate limiting fails against scrapers using residential proxies and rotation.

User-agent filtering fails the same way. The scraper that announces itself as python-requests is the one you catch. The one that copies a real Chrome string, which is most of them, sails through. The filter that blocks the honest bot and misses the dishonest one is worse than no filter, because it reports success while the scraping continues.

The layered defense

Stack the layers cheapest first. IP intelligence scores the source address and kills the bulk datacenter scrapers for almost nothing. Rate limiting keyed on something better than a bare IP, like a session or a device fingerprint, catches the ones that rotate. Behavioral signals, request cadence, header consistency, interaction patterns, catch the residential-proxy scrapers that survived the first two layers.

Then decide what to serve them. A hard block is one option, but a degraded response is often better: a cached price, a delayed response, or a page with the price element removed. The scraper gets stale data, the competitor's pricing tool reports a price that is wrong, and your server spends less effort than it would fighting a block the scraper routes around in an hour.

Related reading: Bot Traffic Index, Answers, Glossary.

Frequently asked questions

How do I know if bots are scraping my prices?

Look for regularity: product pages requested in catalog order, at consistent intervals, including out-of-stock variants, with hundreds of views and zero interactions. A human browses in bursts with pauses and backtracking; a scraper browses like a list.

Why does rate limiting not stop price scrapers?

Because serious scrapers rotate through thousands of residential proxy addresses, and each one stays under the per-IP limit while the total volume stays high. Rate limiting per IP assumes a small number of addresses, which is exactly what the scraper does not have.

Should I block scrapers or serve them bad data?

A degraded response is often better than a hard block. Serve a cached price, a delayed response, or a page with the price element removed, so the scraper's data goes stale. A block gets routed around in an hour; stale data poisons the competitor's pricing tool.

What are the layers of a bot defense?

IP intelligence first, to kill bulk datacenter scrapers cheaply. Rate limiting keyed on session or device fingerprint second, to catch rotation. Behavioral signals, cadence, headers, interaction patterns, third, to catch the residential-proxy scrapers that survived the first two layers.

Frequently asked questions

Who pays when bots buy limited stock first?

Almost everyone except the bot operator: genuine customers pay resale prices or miss out, the store pays in support, returns, and reputation, and the market pays as launch data and pricing signals fill with automated demand.

Why do stores tolerate inventory bots for so long?

Because bot purchases still record as sales, and the costs land in other budgets: support time, chargebacks, customer anger, and distorted demand data. The revenue is visible immediately. The costs are diffuse and delayed.

What is the real cost beyond the sold out product?

Corrupted information. Launches read as more popular than they are, waitlists inflate, and stock and pricing decisions get made on demand that was never human. The store optimises for an audience of scripts.

What actually reduces inventory bot economics?

Raising the cost per attempt: verified accounts for limited drops, queues that reward patience over speed, per customer limits enforced across identities, and detection aimed at behaviour. Defences work when they make botting unprofitable, not merely harder.