Back to PromptMates
Free Resource

Comp Intelligence
for Recruiters & HR

Where the offer process breaks down, what data you're leaving on the table, and where AI is genuinely useful versus overhyped.

Giac Soliman

Giac Soliman

12+ years at Snap, TikTok & Monzo

6

Key Insights

12+

Years Experience

6

Action Items

What You'll Learn

About the Expert

Giac spent over 12 years leading Compensation & Rewards at some of the most talent-competitive companies in the world. At Monzo, he built comp infrastructure for a rapidly scaling fintech. At TikTok, he operated in one of the most complex and fast-moving talent markets globally. At Snap, he worked at the intersection of high-growth hiring and aggressive comp strategy.

These are the insights that come from doing that job at that level, for that long.

Follow Giac on LinkedIn

“Most recruiters only speak to comp at offer stage. By then, it's firefighting.”

This session breaks down exactly where the process fails and what to do about it.

Insight 01

The offer approval process is a compliance check. It shouldn't be.

The fix isn't faster approvals. It's moving the conversation earlier.

The reason offer approvals feel like a blocker is because comp is being asked to review decisions that are already made. By the time a recruiter submits for approval, the candidate has been through every stage, the hiring manager has a favourite, and the number is already in someone's head. Comp's job at that point is just to say yes or no, and that's not how good comp decisions get made.

When comp strategy (the band, the philosophy, the exception criteria) is documented and shared at role kickoff, the offer stage becomes a confirmation, not a negotiation.

Try This

At your next role kickoff, ask comp three questions: What's the band? What's the philosophy behind it: are we leading, matching, or lagging market? And what would an exception need to look like to get approved? Write it down. You'll save two weeks at offer stage.


Insight 02

You're sitting on better market data than you think. You're just not capturing it.

Triangulating all three sources gets you materially closer to real market rate.

There are three data sources that matter for comp benchmarking. First: your internal employee salary distribution, meaning what you're already paying people in comparable roles. Second: your own offer and candidate data, including what candidates have told you they're earning, what offers they've received elsewhere, and what they've accepted or declined. Third: what other companies are actively offering right now to candidates who haven't applied to you yet.

Most teams rely only on annual pay surveys, which are already 12 months old by the time they're published. The second and third sources are live, specific, and sitting inside your ATS and your recruiters' heads, largely uncaptured.

Try This

Start logging candidate comp expectations and competing offers in a shared doc or ATS field. Even six months of consistent data gives your comp team something no survey vendor can provide: live intelligence from your actual market.


Insight 03

LLMs will give you a salary number. Here's how much to trust it.

Specific beats synthetic every time.

LLMs are probabilistic. They're predicting a likely answer based on patterns in training data, not calculating from a real-time database. For comp benchmarking, that means they're useful for orientation and rough sense-checking, not precise decisions. The bigger problem is that LLMs can't show their working in a way you can verify.

Tools like Compa's Partner AI go further by embedding verified comp data directly into the recruiting workflow, so the guidance you're getting is grounded in real benchmarks rather than probabilistic output. But even without a dedicated tool, the strongest move is to find two or three real, current job ads from direct competitors hiring the same role and use those as your anchors.

Try This

Next time you need a rough benchmark with no survey data, search for the exact role at three to five companies competing for the same talent. Screenshot the salary bands from their live job ads. Then use an LLM to help you interpret and contextualise, not to generate the number in the first place.


Insight 04

Hiring for AI skills is quietly pulling your pay bands out of shape.

"AI skills" is too vague to build a comp architecture around.

Most companies right now are hiring AI engineers or AI-adjacent roles at a significant premium, and attaching them to existing job families that were never designed for that market rate. The result is pay compression, equity complaints, and parity issues that take years to unwind.

Until you define the role in terms of business impact, complexity, and knowledge requirements (not just job title), you can't enforce parity, and you can't build a band that holds. The companies getting this right are doing systematic job sizing before they open the requisition, not after they've already made three exceptions.

Try This

Before opening any AI-adjacent role, ask comp to size it using your existing job evaluation framework, even roughly. If you don't have one, use three questions: What's the business impact if this role is done well versus badly? What level of technical and strategic complexity does it involve? Does it require managing people, decisions, or both?


Insight 05

The offer approval loop has a data problem. You can fix your half of it.

The conversation shifts from "we need another exception" to "the band is wrong."

One of the least-solved problems in comp is closing the loop between offer decisions and comp strategy. Knowing that you deviated above band on several offers last quarter is useful, but knowing why, and whether the pattern points to a broken band rather than individual exceptions, is where the insight actually changes behaviour.

When you can see that deviations are consistently 15% above midpoint for a specific job family, the conversation with comp shifts from requesting another exception to challenging whether the band itself is wrong. That's a much more productive place to be, and it's a conversation recruiters are much better positioned to have when they have the data.

Try This

Start tracking your own offer exceptions in a shared log: role, level, band midpoint, offer made, percentage deviation, and outcome (accepted/declined). Three months of this gives you the evidence base to challenge a band rather than just work around it.


Insight 06

In merit cycles, the AI use case isn't the spreadsheet. It's the manager conversation.

A well-prompted LLM with the right context can draft a personalised brief in minutes.

The formulaic part of a merit cycle (applying percentage increases based on performance rating and position in band) is already mostly automated in any modern HRIS. That's not where AI adds value. The two places it genuinely does are risk flagging and manager preparation.

When a manager has to deliver a 3% increase to someone who expected 8%, the conversation goes badly if they don't understand why. Not just what the policy says, but what actually happened in the approval chain. An AI system with full context on the merit cycle can generate a personalised Q&A brief for each manager before those conversations happen.

Try this (for HR Business Partners):

Before merit outcomes are communicated, ask yourself: does each manager know why their team members got what they got, not just what the number is? If not, that's the brief to write. A well-prompted LLM with the right context (performance rating, original recommendation, what changed in approval) can draft it in minutes.