Integration Debt, Part 2: The Number That Doesn't Exist

A $847-per-month Zapier bill from 15,000 tasks shows up constantly in articles about integration costs, usually paired with a worked example: an eight-step workflow times 100 leads a day producing 24,000 tasks a month. It's a specific, citable-sounding number. It's also, as far as we can tell, made up.

That figure and that exact worked example appear nearly verbatim across a long list of SEO-driven comparison sites. None of them link to an original post, an invoice, or a named source. It reads like an illustrative "here's the math" example that got copied from article to article until repetition made it feel like a documented fact. We looked for where it actually came from and couldn't find it. We're not citing it here, and you should be skeptical of any article that cites it without a source attached.

That single untraceable number is worth dwelling on, because Part 1 made the case that integration debt is a real, structural mechanism: the combinatorics of connecting systems, the ownership vacuum between them, the way failures propagate silently instead of throwing errors. All of that holds up. The moment the conversation shifts from "how does this happen" to "how much does this cost," the quality of the evidence drops off a cliff, and most articles on this topic don't tell you that.

What actually holds up

The billing mechanism that would produce a bill like $847 is real. Zapier's own documentation confirms that each action in a multi-step Zap counts as a separate billable task, so a five-step Zap consumes four tasks every run. The best aggregate evidence that this surprises people isn't a set of matching dollar-jump anecdotes, it's the rating gap: Zapier sits at 1.4 out of 5 on Trustpilot, against 4.5 on G2 and 4.7 on Capterra, driven heavily by billing-surprise complaints. One verifiable Reddit thread, describing a bill jumping from $30 to $160 after adding a single Zap, illustrates the mechanism well. Treat it as one example, not evidence that stories like it are common; the rating gap is what actually shows that.

Two individual practitioners, posting under their own accounts rather than as SEO content, documented real cost comparisons after migrating to self-hosted n8n. One reported paying $19.99 to $49 a month on Zapier for roughly 2,000 tasks across eight workflows, versus a $6-a-month server with unlimited executions after switching. Another itemized roughly 2,400 euros spent on Zapier over three years, against a 5-euro-a-month n8n server now, a figure worth trusting more than that same post's headline claim of 2,000 euros a year saved, a rate its own spend total doesn't cleanly support. Both are individuals, not companies, and both have an obvious motive to favor the tool they switched to. But they're primary accounts with real numbers attached, a different evidentiary class than a recycled blog statistic nobody can source.

The consolidation side is thinner, not stronger

If integration debt is real, consolidating onto fewer systems should reduce it. Fewer systems means fewer point-to-point connections, which follows directly from the combinatorics in Part 1. The evidence for this side of the argument, though, is if anything weaker than the cost side. Search for a company that documented "we had N integrations, we consolidated, we now have M," and we couldn't find one. What exists instead is vendor case studies using phrases like "single source of truth" and "eliminates data silos" that gesture at the benefit without measuring it.

A couple of cases come closer than most, caveats attached. rtCamp, an ERPNext customer profiled on Frappe's own site, replaced seven named tools (Recruitee, Keka, Pipedrive, Xero, Greytip, ActiveCollab, and Everhour) plus "dozens of spreadsheets" with one ERPNext instance, reporting "tens of thousands of dollars" saved, no exact figure given, and 30 to 40% time saved per week in operations. It's the most specific case we found: a real company, a real named tool list. But even this doesn't quantify integrations eliminated or sync failures avoided. The efficiency figure is a general operations claim, not an integration-specific one.

Liquidity Services, in a HubSpot-published case study, consolidated eight tools onto HubSpot and reports a 50% overall cost reduction, a specific number attached to a named company. But only two of the eight consolidated tools are actually named, so the number can't be checked against a full before-and-after list.

Past those two, the evidence gets vague fast: case studies describing unnamed "disconnected systems" replaced with "centralized data," an anonymized manufacturer reporting savings that can't be checked against a public company, and a widely repeated claim that "Gartner says vendor consolidation cuts SaaS spend 15 to 30%" that, on inspection, traces to no locatable Gartner report at all. It's the same shape of unverifiable floating statistic as the $847 Zapier figure, just pointed in the opposite direction.

Why this gap matters more than it seems

This is the part most content skips, and it's the part that actually matters for how you should read everything else on the topic. The mechanism-level case for integration debt is strong. The combinatorics are real, the silent-failure pattern is corroborated across independent sources, and the billing-model incentives are confirmed by vendors themselves. What's missing is the case-study literature proving it costs a specific amount, or that fixing it saves one, on either side of the argument.

That doesn't make the concept less real. It makes it a mechanism you can trust and a set of dollar figures you mostly can't, which is an unusual place for a business topic to sit and an important one to be honest about before acting on it. The practical question is what a business should actually do while that evidence gap exists, budgeting for a cost it can't fully quantify and betting on a fix that hasn't been rigorously measured. That's where Part 3 goes.