Lenss to decide build vs buy

Price the build honestly before you cancel the vendor.

Build versus buy memos, checked for the total cost of owning the thing you build: people, upkeep, compute and egress.

Works on: Build vs buy memos · Vendor replacement proposals · Platform investment cases

How the audit runs

  1. 01

    Pin the decision criterion

    Cost, control or capability: what the memo says decides it.

  2. 02

    Quote the cost the memo leaves out

    Maintenance, on-call, security patching and egress, where the memo mentions them or implies they are zero.

  3. 03

    Rebuild total cost of ownership

    Both options on the same basis, from the memo's own figures.

  4. 04

    Compute the volume threshold

    The usage at which building starts to pay.

Typical findings

Engineers priced at zero

Upkeep the team already pays for on similar systems, left out of the build case.

Today's volume, tomorrow's price

Vendor costs projected at a growth rate the build case ignores.

Hidden infrastructure

Egress, storage and redundancy the vendor was absorbing.

Worked example

The same draft, read by a standard AI assistant and by Lenss. Every figure is computed from the draft.

Source draft

Memo: replace the hosted search vendor

Proposal: replace our hosted search vendor with a self-hosted cluster. Vendor cost. $24,000 a year platform fee plus $0.50 per 1,000 queries. We run 40 million queries a month, so the vendor costs $264,000 a year. Build cost. Six nodes at $1,100 a month each: $79,200 a year. Building saves $184,800 a year. Maintenance. Upkeep should be minimal once it is running. Context. Our self-hosted logging cluster took 1.5 engineers of upkeep last year (upgrades, security patches, on-call), at $200,000 per engineer fully loaded. Egress for search would be about 2 TB a month at $0.09 per GB.

Pinned objective

Cut search cost without losing reliability.

Hidden assumption

That upkeep is minimal. The team's own comparable cluster took 1.5 engineers last year.

“Upkeep should be minimal once it is running.”

If it’s wrong: Priced like the logging cluster, building costs $381,360 a year: $117,360 more than the vendor, not $184,800 less.

Flip threshold

Monthly query volume at which building pays

≥ 59.6M queries/mo draft states 40M

Vendor at 40M queries/mo
$264,000/yr
Build: compute
$79,200/yr
Build: 1.5 engineers of upkeep
$300,000/yr
Build: egress
$2,160/yr
Build: total
$381,360/yr

At 40M queries a month, keep the vendor. Revisit once volume passes about 60M, or if upkeep can be shown to be under 0.9 engineers.

Cheapest test: Ask the logging cluster's owners for last year's on-call and upgrade hours, and price the build with them.

Used most by Engineers

Audit your own draft now.

Paste up to 24,000 characters. You get the objective, every quoted assumption, the tensions, the flip threshold and a hardened rewrite.

Audit your draft