Course › Module 1 · Foundations of AI recruiting

The AI recruiting stack: seven layers

Module 1, Lesson 2  ·  4 min read ·  Updated 21 September 2026

Module 1 · Lesson 2

There are hundreds of AI recruiting tools and they all sound the same. Here is a simple way to tell them apart: picture seven layers, stacked on top of each other.

The seven-layer AI recruiting stackSeven stacked layers from system of record at the bottom to measurement and governance at the top. The bottom two are heavier and marked load-bearing.7Measurement and governanceFunnel metrics, audit trail, bias checks6OrchestrationMoves work between the layers5CommunicationOutreach, scheduling, rejection4Matching and rankingOrders candidates against a role3Search and enrichmentFinds and augments profiles2Ingestion and parsingTurns files into structured fieldsload-bearing1System of recordCandidates, roles, stages, historyload-bearing
The seven layers. The bottom two carry everything above them.

The seven layers

LayerWhat it doesWhat goes wrong without it
1. Your databaseStores candidates, jobs, and where each person isEverything lives in email and spreadsheets
2. Reading CVsTurns a document into data: name, employers, dates, skillsEvery number above this layer is built on mistakes
3. Finding peopleSearches for candidates who did not applyYou only ever see people who came to you
4. RankingPuts candidates in order against a jobYou still have a pile, just a bigger one
5. MessagingOutreach, booking interviews, updates, rejectionsFast at the start, silence at the end
6. Joining things upMoves work between layers without youYou are the glue
7. Reporting and checksNumbers, records, fairness checksYou cannot prove it works, or that it is fair

The layer nobody shows you

Sales demos almost always start at layer 4, the ranking. A neat ordered list looks clever.

But that list is only arithmetic done on layer 2 — the data pulled out of the CVs. If the tool read a CV wrongly, the ranking is wrong too. And here is the problem: a wrong ranking looks exactly like a right one. No error message. Nothing turns red.

That is why Module 4 spends most of its time on layer 2, which nobody finds exciting.

Most tools are one layer wearing a suit

A browser extension that finds candidates is layer 3. A tool that writes up interview notes is layer 5. A chatbot is layer 5. None of them is a complete system.

Buy five and you have five separate tools, candidate details in seven places, and no way to report on anything.

Layer 7 is where projects get cancelled

A team sets up layers 2 to 6. Everything works. Then someone in legal asks: "is this fairer or less fair than what we did before?" Nobody can answer, and the project stops.

Reporting and fairness checks are not paperwork you add at the end. They are a layer with their own requirements. Module 11 covers them properly.

Four questions for any demo

  1. After I use this, where do candidate records live — your system or mine? Tells you if you are about to end up with two databases.
  2. Show me the data you pulled out of a CV, not the score. Tells you whether they really own layer 2.
  3. What happens after the ranking? Can it message people or book interviews? Separates layer 4 from layers 5 and 6.
  4. What can I export if someone asks me to justify a decision? Tells you whether layer 7 exists at all.

Use this as a shopping list

Before your next sales call, write down which layers you already have and which you are missing. Then ask which layer the tool sits in. Most calls get shorter and more useful after that question.