The three-month project that never ended. Anand is a senior engineer at a B2B data enrichment company. His company paid a managed provider $1,600 a month for proxy rotation. In April, his CTO asked one question: "We're spending too much on proxy infrastructure. Can we build it in-house?"
Anand said three months. He understood SMTP, he'd worked with proxies, and he was right that it wasn't beyond him. Over three months he put in 140 hours. The system worked. He handed it over and moved on. Then the maintenance started — and he kept a log.
| Month | What happened | Hours |
|---|---|---|
| 4 | 40 of his 200 IPs appeared on a secondary reputation list. He found them, filtered them, sourced replacements and retested rotation. | 16 |
| 5 | Google-hosted mail servers changed how they replied. Unknown rate on Google-hosted addresses rose 6 points before anyone noticed. Diagnose, patch, retest. | 24 |
| 6 | 2am incident: the rotation worker got stuck. | 6 |
| 7 | Two suppliers raised prices and one dropped the residential tier he used. New evaluation, contract, integration. | 16 |
| 8 | Routine replacement of 10 IPs. | 5 |
| 9 | 2am incident: the blacklist check job had been failing silently. | 6 |
| 10 | Another mail host change, another patch. | 8 |
| Total, months 4 to 10 | 81 | |
In month nine, after the second 2am incident, Anand called us. He didn't want to replace what he'd built — he wanted to know if this was normal. He kept running the pool for two more months, which is why the log goes to month ten, and moved back to managed in month eleven.
"Build or buy" is the wrong first question. The better question is who owns the system at 2am, and what that person would otherwise be building. Since I sell managed proxy infrastructure, don't just trust my conclusion — the rest of this post uses Anand's own log and simple arithmetic you can redo with your numbers.
01Months 1 to 3: What the build cost
A working rotation pool for email verification needs five pieces. Here are typical hours for a senior engineer:
| Component | Hours |
|---|---|
| Evaluating and contracting 3 to 5 IP suppliers | 12 – 20 |
| Rotation logic (assign IPs, track usage, rotate on thresholds) | 40 – 60 |
| Health monitoring (quarantine slow or failing IPs) | 20 – 30 |
| Blacklist checking (remove listed IPs, flag for replacement) | 15 – 25 |
| Dashboard and alerts | 20 – 35 |
| Total | 107 – 170 |
Anand's 140 hours sit inside that range. At a fully loaded $150 an hour, his build cost $21,000. At $120 to $180 an hour, the range for the whole build is about $12,800 to $30,600. This is the number most teams estimate. It's also the smaller one.
02Months 4 to 10: What the maintenance cost
Two cautions: seven months is a small sample, so your number will differ, and this is engineering time only — it doesn't include what he paid suppliers for IPs. Look at what each line in the log really is:
- Month 4 is pool churn. IPs get listed and allocations change. Some replacement work never ends.
- Months 5 and 10 are mail hosts changing behaviour. Your rotation rules were tuned to how servers replied last month; when they change, results drift until someone notices and patches.
- Month 7 is supplier management. Prices, tiers and contracts change without asking you.
- Months 6 and 9 are the 2am incidents. Production pipelines fail at any hour, and the person who built it is the person who gets called.
None of these could be in a three-month estimate. They're the cost of owning the system after it works.
03What the CTO was actually saving
The CTO's idea was to cut a $1,600 monthly bill. Anand's in-house pool still needed IPs from suppliers, at about $900 a month. Managed: $1,600 × 12 = $19,200 a year. In-house IP supply: $900 × 12 = $10,800 a year. Gross saving from going in-house: $8,400 a year. Now put the engineering time against that saving.
| In-house | Managed | |
|---|---|---|
| Build (once) | $21,000 | $0 |
| Maintenance per year (139 hours) | $20,850 | $0 |
| IPs or subscription per year | $10,800 | $19,200 |
| Year 1 | $52,650 | $19,200 |
| 24 months | $84,300 | $38,400 |
Even if you ignore the build completely, in-house costs $31,650 a year to run against $19,200 for managed — $12,450 more every year, with the system running fine.
The formula you can use with your own numbersOn cost alone, in-house is cheaper only when:
is bigger than
(yearly maintenance + build cost spread over 3 years)
For Anand: $8,400 on the left, and $20,850 + $7,000 = $27,850 on the right. His saving would have needed to be about 3.3 times bigger before in-house made sense on cost. The formula doesn't count the IP history gap or the engineer's lost focus, which come next — and both make in-house look worse than the formula alone shows.
04The cost that is on no invoice: IP history
An IP's record with mail servers is built by the traffic that has passed through it over time. You can't speed that up — you can wait for it, or you can use IPs from someone who has already run clean traffic for years. Supplier IPs are also not blank: they carry whatever earlier users did with them, and you don't control that.
Mail providers don't publish how they score IPs, so this comes from our experience running verification traffic rather than from a published rule — but it held up in Anand's case:
That's a gap of roughly 5 points. At his volume, here's what it means:
- 600,000 verifications a day. Enterprise-hosted domains are 40% of his list — 240,000 a day.
- 5 points of 240,000 = 12,000 more unknown results every day, about 360,000 a month. The gap applies to the enterprise share only, not the whole volume.
Each of those unknowns is a contact his customers can't confidently send to or drop. They don't see IP history — they see a result quality that got worse, and ask why. You can't fix this by writing better rotation code. You fix it by running clean traffic for a long time.
05The cost of the engineer's attention
Add Anand's hours together: 140 to build and 81 to maintain. That's 221 hours before he migrated back — about 5.5 full working weeks of a senior engineer, spent on proxies. His company sells enrichment, not proxy infrastructure. Those weeks didn't go into the product his customers pay for.
And he was the only person who understood the system. Both 2am incidents were his. That's the real support cost: the system works only as long as the one person who built it is available and awake.
06When in-house does make sense — and when it doesn't
In-house makes sense when- The gap in the formula is large — your managed bill clearly beats maintenance plus build.
- You have more than one engineer who understands the system, with documentation and an on-call plan.
- Proxy infrastructure is part of your product, or you need control over the IPs for reasons the formula doesn't capture.
- One engineer owns it — that's a single point of failure.
- The saving is smaller than the maintenance. That was Anand's case.
- You're in the first year. In his log, the first months brought the most surprises.
- Your customers are sensitive to result quality on enterprise domains, because of the IP history gap.
07What happened next
In month 11, Anand moved back to managed infrastructure. The migration took three weeks. He told me afterwards that those three weeks were the last proxy-related time he spent that wasn't on the product.
The CTO's original question was fair: are we spending too much? The honest version is different. It isn't "build or buy." It's whether $8,400 a year in savings is worth 139 hours a year of a senior engineer, plus a $21,000 build, plus a worse result on enterprise domains. For Anand's company the answer was no — and it took eleven months and 221 hours to learn it.
Start with 500 free credits
Residential proxy infrastructure for email verification, running on IP history that takes years to build in-house. Run the formula with your own numbers — and if in-house comes out ahead, build it.
Get started →Frequently asked questions
Is it cheaper to build proxy infrastructure in-house?
Usually not once engineering time is counted. On cost alone, in-house is cheaper only when the gap between the managed bill and the in-house IP bill is larger than yearly maintenance plus the build cost spread over three years. In the case described here, the saving was $8,400 a year against about $27,850 a year of maintenance and amortized build — roughly 3.3 times too small to justify building.
How much does proxy infrastructure cost to build?
A working rotation pool for email verification typically takes 107 to 170 hours of senior engineering across supplier evaluation, rotation logic, health monitoring, blacklist checking, and a dashboard. At $120 to $180 an hour fully loaded, that's roughly $12,800 to $30,600 — and the engineer in this example spent 140 hours, or about $21,000 at $150 an hour.
What does it cost to maintain a proxy rotation pool?
In the logged example, maintenance took 81 hours over seven months — about 11.6 hours a month, or roughly 139 hours a year, which is around $20,850 at $150 an hour. That's engineering time only, and it excludes what's paid to IP suppliers. Seven months is a small sample, so actual maintenance will vary by team and pool size.
What are the hidden costs of in-house proxy infrastructure?
Three costs rarely appear in the initial estimate: ongoing maintenance (pool churn, mail host behavior changes, supplier management, and off-hours incidents), IP history, which can't be built quickly and showed up as a roughly 5-point rise in unknown results on enterprise domains, and the opportunity cost of a senior engineer's attention — 221 hours in this case, about 5.5 working weeks not spent on the core product.
Should you build or buy proxy infrastructure?
Run the cost formula with your own numbers first. Building tends to make sense when the savings gap is large, more than one engineer understands the system with documentation and an on-call plan, or proxy infrastructure is part of what you sell. It usually doesn't when one engineer owns it, the saving is smaller than maintenance, or your customers are sensitive to result quality on enterprise domains.
How does managed proxy infrastructure compare with in-house infrastructure?
In the worked example, managed cost $19,200 a year with no build or maintenance burden, while in-house cost $52,650 in year one and $84,300 over 24 months against $38,400 for managed. Managed also came with an established IP history that the in-house pool lacked, and it moved the on-call responsibility off a single engineer. Note that the author's company sells managed infrastructure, so the arithmetic is worth rechecking with your own figures.
Why does IP history matter for email verification?
An IP's record with mail servers is built by the traffic that has passed through it over time, and it can't be sped up. In the example, the unknown rate on enterprise domains rose from about 12% on a managed setup to about 17% on a new in-house pool, then returned to about 12.5% after moving back — roughly 12,000 additional unknown results a day at that volume. Mail providers don't publish how they score IPs, so this reflects observed patterns rather than a published rule.