Six startup mistakes that killed PricingBot and built ScrapingBee

Executive overview

Pierre de Wulf and his co-founder Kevin spent a year building PricingBot, a price-monitoring SaaS for e-commerce owners, reaching only $450 MRR before quitting. They made six compounding mistakes — chasing vanity metrics, skipping real customer conversations, spreading effort across too many channels, and staying consistent at the wrong things. The lessons directly shaped how they built ScrapingBee, a web scraping API that grew to $1.5M ARR. Every decision at ScrapingBee was a deliberate inversion of a PricingBot failure.

The only true validation is a paying customer — everything else is noise.

Confusing users with customers

• PricingBot launched on Product Hunt, got hundreds of upvotes and press coverage, and treated it as product validation.
• Signups and trial accounts felt like momentum but produced zero revenue.
• The team gathered feedback from hundreds of non-paying users and optimised for their opinions.
• ScrapingBee ignored Product Hunt entirely in early days and instead targeted narrow communities: growth-hacker forums, web scraping groups, developer Slack channels.
• Fewer trials but far earlier paying customers — that signal changed everything downstream.
• Rule: until one or two people use the product and pay for it, nothing has been validated.

Knowing the industry but not the customer

• The team studied e-commerce deeply: Amazon FBA, Shopify, dropshipping, pricing dynamics — and built a detailed knowledge map.
• This created confident assumptions about what e-commerce owners wanted that turned out to be wrong.
• When asked "what else have you tried?", PricingBot prospects said things like "we've never really managed to do price monitoring" or "it's too complicated, we just don't bother."
• That answer revealed price monitoring was a nice-to-solve problem, not a must-solve one.
• ScrapingBee prospects said things like "we built an in-house tool that's a hell to maintain" or "we're spending $5,000 a month on web scraping."
• Those responses showed a problem people were already spending real resources to fix — a far stronger signal.
• To generate more customer conversations early, ScrapingBee offered 10,000 free API credits (vs. the standard 1,000) in exchange for a 15-minute call, generating ~150 conversations in the early months.
• Key interviewing principle from The Mom Test: ask about their life and existing behaviour, not about your idea. "How do you solve this now?" beats "What do you think of my product?"

Waiting for users to ask for features

• With few customers early on, it was hard to know what to build next through direct conversation alone.
• ScrapingBee instrumented the documentation page with Hotjar (later Microsoft Clarity) heatmaps.
• A clear table of contents listed every feature being considered; unbuilt features linked to a "coming soon" paragraph.
• Clicks on those sections acted as a low-cost proxy for feature interest — for example, a "screenshot" entry in the table of contents attracted heavy clicks, validating that feature before a line was written.
• The method is imperfect (clicks can reflect curiosity rather than intent) but useful as an early signal at near-zero cost.
• Pierre still runs this test at scale: he adds speculative sections to the docs without telling his co-founder, discards users who spent fewer than two minutes on the page, and watches the heat map.

Treating a slow sales cycle as an email problem

• PricingBot had months-long sales threads with prospects who kept deferring: "I'll talk to my co-founder," "bad time right now," "let's revisit in a few weeks."
• These prospects were avoiding saying the product wasn't good enough — and the team kept optimising emails instead of recognising the signal.
• ScrapingBee adopted a different test: if the product solves a real problem people want solved now, they should convert within a short conversation.
• Within 10 minutes on the phone or in chat, ScrapingBee was closing several-hundred-dollar-per-month subscriptions.
• Fast closes are a health indicator, not just a sales outcome — they confirm the problem is urgent and the solution is credible.

Spreading across too many acquisition channels

• PricingBot tried SEO, affiliate marketing, cold outreach, social content, Reddit ads, and Google ads — often in quick succession.
• The team consistently abandoned channels before reaching the compounding phase, mistaking early low results for proof of failure.
• ScrapingBee committed to SEO only from day one and did nothing else while the team was two people.
• Kevin had prior experience running a Java web scraping blog, so SEO played to an existing strength.
• Early articles got only a few hundred views, but the team kept publishing and watched domain authority compound over months.
• Methodology: started with the Skyscraper technique (write the best possible educational content on the industry topic), then graduated to data-driven keyword selection using Ahrefs — specifically the free "Blogging for Business" course — prioritising keyword difficulty to avoid unwinnable battles against high-authority competitors.
• Target cadence: four to six quality articles per month; at that rate they identified six months to a year of remaining SEO opportunity even after significant growth.

Believing in silver bullets

• When PricingBot's growth stalled, the team repeatedly convinced itself a single tactic would fix everything: affiliate partnerships, a new pricing page, a different email sequence.
• Each silver bullet was a way to avoid the harder underlying questions: Is the product actually useful? Do users choose it repeatedly? Do any of them pay?
• At PricingBot the answer to all three was no — but the team never stopped to ask them directly.
• The three diagnostic questions Pierre now recommends: (1) Do I know what I need to do to make this work? (2) Am I actually able to do it? (3) Do I want to build the company this requires?
• PricingBot's path to success would have required competing with enterprise tools serving thousands of SKUs — technically harder, required funding, and not what the founders wanted to build.
• Answering those questions honestly early would have saved several months.

Confusing consistency with progress

• Startup culture celebrates showing up every day, but consistency at the wrong thing compounds failure rather than success.
• PricingBot's most vivid example: a customer named Amit sold adult dolls and needed competitor price matching. The team spent a full week, ten hours a day, manually matching product listings to keep him as a customer.
• Amit paid $600 out of pity, then cancelled two weeks later saying the product wasn't useful.
• The team repeated the same manual-onboarding effort for dog food, mattresses, and light bulbs — and it never converted into retained customers.
• The hard part is distinguishing "not working yet" from "working just a bit" — zooming out to months-level trends, not weeks, is the only reliable way to tell.
• If the needle never moves at all regardless of effort, that is a clear signal to stop.

Knowing when to quit

• PricingBot was shut down after nine months; Pierre believes quitting earlier would have been better.
• The risk is the opposite trap: founders who keep iterating on a failing product for years because there is always one more thing to try.
• The three questions work as a quit framework too: if you do not know what would make it work, cannot execute it even if you did, and do not want the company it would require — the rational move is to stop.
• Building in public (Twitter) contributed zero ScrapingBee customers directly but opened speaking opportunities, press mentions (including a Microsoft case study), and relationships — useful, but not an acquisition strategy unless your audience perfectly matches your product.
• Retrospective customer interviews become redundant after a few months: ScrapingBee stopped the 10k-credits-for-a-call programme once Kevin was fielding 30 calls a week with diminishing new information, replacing it with a short signup form that routes high-intent prospects to a sales call.