Budgeting for AI in HR: What to Put in Next Year's Plan

ISG’s HR technology survey, published in November 2025, says AI budgets for HR will average $1.6 million in 2026, a tenfold increase since its 2023 survey. That’s a forecast, averaged across companies from 1,000 to more than 100,000 employees, so it says very little about what a company of a few thousand people should spend.

A second number from the same survey matters more for the plan on your desk. Only 52% of organizations see quantifiable value from HR technology.

Finance will ask what your AI line buys. The line you can’t answer that question for is the one that gets cut.

What to put in next year’s plan, line by line

The table below lists the nine lines most HR AI budgets need. Data repair adds a tenth if your systems need work first. Only the strategy line can carry a quoted fee before the year starts. Every other line gets its number at a known point, and the last column says when.

Line in your plan	What it covers	Who sets the number	When you can write a number
Strategy	Sorting and sequencing your AI ideas into a roadmap, with a named owner for each metric it aims to move	A strategy partner, or your own team. With us: “One fixed fee for the month, quoted after the first call and scaled to the size and complexity of your function”	After a first call
Proof, per use case	A working version on test data and a written go / no-go	Your build partner, or your own team. With us: “Fixed fee per use case, quoted from the roadmap”	Once the roadmap exists
Build, per use case	A production system in your environment	Your build partner, or your own team. With us: “Fixed fee for the agreed scope, quoted after the proof”	After that use case’s proof
Software licences and vendor tools	Any tool the roadmap says to buy	The vendor’s quote	When the roadmap names the tool
Hosting and model usage	Running what’s built	Your own providers, who bill you directly	Estimated from the proof, which reports cost per query on data shaped like yours
Compliance for specific use cases	For example, the annual bias audit and candidate notice NYC Local Law 144 requires for a screening tool	Your auditor and counsel	When the roadmap names a use case that needs it
Your HR team’s time	Owners, data pulls, reviewing outputs	You, in hours	From the start: the strategy month asks for “a handful of working sessions of about two hours, plus one data pull between them”
Enablement for each launch	Onboarding and change work for the managers and employees who’ll use it	You. Each build of ours includes “Onboarding for the people who’ll use it”	With each build
Ongoing support (optional)	Sequencing the next use cases as the program grows	A retainer, month-to-month	Only after the strategy step

The first three lines carry the work, and finance will push on them hardest. With us they run in order, starting with the strategy month, and each is priced at a different moment.

Why the AI line is landing on your desk this budget season

ISG’s same survey found that more than two-thirds of enterprises rank AI adoption among their top three HR priorities. It also found that cost of ownership, which buyers ranked lowest two years ago, has risen to third place amid growing economic pressures. Both pressures land on one line in your HR AI budget: the business wants AI adopted in HR, and finance wants to know what it will cost to own.

Then there’s the AI you’re already paying for.

“We found that 31% of organizations are currently using some kind of AI-enabled technology, ranging from ChatGPT (58%) to AI-supported meeting tools (11%) to HR applications with embedded AI (9%).” That’s Stacey Harris, chief research officer at Sapient Insights Group, reporting their 2025-2026 HR Systems Survey, which drew on 9,886 HR professionals at 4,670 organizations. The base is any AI-enabled technology, including general tools like ChatGPT. General tools lead that list, and HR applications with embedded AI come last.

If your function looks like that list, today’s AI spend may be a few general licences with no HR metric attached. That’s a normal start, and it’s why a strategy line belongs in next year’s plan. Without one, “AI” in the budget means whatever is already on the card statement.

Price the strategy and proof lines the way the work actually happens

The strategy line

With us, strategy takes about a month, on “one fixed fee for the month, quoted after the first call and scaled to the size and complexity of your function.” You can write that number into the plan as soon as the call has happened.

The month produces a sequenced roadmap. Every idea on your list, yours and ours, is scored on three lenses from the 10X HR-AI Framework: Impact, which number it moves; Trust to earn, what a wrong answer costs, who decides and how sensitive the data is; and Readiness, where the data lives and whose name is on the number. Each idea then lands in one of four tiers: Build now, Proof first, Build with guardrails or Not yet. There are two ways to land in Not yet, so the list below has five lines. The framework’s scorer checks the first one before anything else:

• Not yet, by the hard rule. If the idea, as imagined, would make a decision about a person on its own, it stops here. Nothing else gets weighed.
• Build now. It moves a real number, has little or some trust to earn, and the data is ready or nearly ready.
• Build with guardrails. It moves a real number, has real or a lot of trust to earn (sensitive data, or a decision about a person nearby), and the data is at least half ready.
• Proof first. It’s at least useful to your own team, has little or some trust to earn, and the data is at least half ready. A smaller number, or data that’s only half ready, usually lands here.
• Not yet, for another reason. It has the most trust the framework can score, or the data isn’t ready, or nobody can say which number it moves, or the gain is too small to be worth the trust it would have to earn. The scorer names which one.

Here’s the part that changes your budget. Every use case that goes on to production runs a proof first, whichever tier it lands in. Our build step is “quoted after the proof” and “scoped after the proof”, and each step ends in a written go / no-go. A Build now idea doesn’t skip it. The scorer’s own advice for Build now is to “start with the proof step and go to production when the numbers hold.”

The tier sets the order and the depth of the work. Build with guardrails ideas usually come later, because “the guardrails, the notice and who decides are designed before anything is connected.” Not yet ideas get no proof line until the one thing to fix is fixed and they score again.

If you don’t have a shortlist yet, the map of 122 HR agents is an inventory to score. We’ve also scored three real use cases end to end.

The proof line

The proof is “fixed fee per use case, quoted from the roadmap.” It takes a few weeks per use case, on test data in a sandbox, and ends with a written go / no-go plus accuracy, failure cases and cost per query on data shaped like yours.

There’s no per-use-case number until the strategy step has produced the list. That makes the defensible figure for the proof line a range, bounded by how many Build now, Proof first and Build with guardrails ideas the roadmap sends to a proof. A round number picked before the list exists has nothing under it.

Put a name on each line, too

Each line also needs an owner and the number it’s meant to move. Mercer’s Global Talent Trends 2026 shows why. The C-suite’s top people priority for return on investment is redesigning work to incorporate AI and automation, at 63%. HR’s top priority is enhancing the employee experience to attract and retain talent. That’s two answers to “what is this money for?”, and in a budget review the line with no owner and no target gets questioned first.

“Whose name goes on each number” is one of the five questions the strategy month answers, and you name the owner on your side. Before you assume your current team can absorb the work, check who on it has HR tech or people analytics skills. AIHR reports that only 10.5% of the HR payroll budget goes to role families with the fastest-growing capabilities in the tightest talent markets, such as HR tech, people analytics and total rewards.

The build line stays a placeholder until the proof reports back

Build is a “fixed fee for the agreed scope, quoted after the proof.” It typically takes a few weeks to a couple of months per use case, inside your environment and under your security team’s controls. A support period after handover is included, scoped with the build.

Hosting and model usage follow it, billed to you by your own providers, and the proof’s cost-per-query figure sizes that line.

A precise build number written into next year’s plan before any proof has run is a guess. If finance wants one total for the AI budget for HR, give them a bounded range for the proof step and mark the build line “priced after proof.” Don’t invent a figure to fill the cell.

Some finance teams won’t accept that. An annual plan with a placeholder in it doesn’t pass their process, and they’re entitled to that rule. Then you’re choosing between a padded build estimate now and a request for more money halfway through the year. Neither one is comfortable, and we haven’t found a version of this that makes it go away.

Attach guardrail costs to the use cases that need them

Compliance sat on its own line in the first table, tied to specific use cases. Some use cases need almost nothing extra. Others need an audit before day one. Three examples from the map:

Use case	What a wrong answer touches	Extra guardrail cost	Where it goes in your budget
HR Policy Question Answering Agent	An answer to a policy question, from your approved documents. Grievance, health, harassment and pay disputes go to a person unanswered	HR samples a share of the logged answers	Inside the proof and build lines
Application Screening Support Agent	Who advances toward an offer. In NYC it’s an automated employment decision tool; in the EU it’s high-risk under the AI Act	Annual bias audit and candidate notice in NYC; works-council consultation in the EU	Its own line, attached to this use case
Pay Transparency Compliance Agent	What gets disclosed about pay. Total rewards and legal approve every response before it goes out	Legal review of every disclosure, and an owner who keeps the jurisdiction rules current	Its own line, attached to this use case

A governance percentage spread evenly would give the policy agent more than it needs and the screening agent less. Pricing each one where it sits fixes both.

If your data needs work first, add a line for it: cleanup, join work, and an owner for each system of record. Those are the repair costs that belong in next year’s plan as their own line, and they come before a use case can be funded. The security review, the production data pipeline and the adoption work belong inside the build and enablement lines. Left out, they’re the costs that show up after a pilot stalls.

Budgets are rising. Confidence is falling.

ISG expects AI budgets for HR to average $1.6 million in 2026, a tenfold increase since its 2023 survey, across companies from 1,000 to more than 100,000 employees. Mercer found the share of C-suite leaders who say their organization is well prepared to succeed in the human-machine era fell from 65% in 2024 to 51% in 2026. And ISG found only 52% of organizations see quantifiable value from HR technology at all.

Each source reports its own number on its own base, and none claims the others are connected. Our reading of the three together: the money is rising faster than the confidence behind it.

Mercer’s President and CEO, Pat Tomlinson, put the condition plainly: “realizing a true return on AI investments depends on intentionally redesigning work and building AI-enabled operating models that amplify the workforce’s capabilities and experience.” Redesigning work doesn’t fit inside one planning cycle.

What sits inside your control this budget season is narrower: whether each line in next year’s plan can be defended on its own, with an owner, a number it’s meant to move, and a plain note on when its price becomes real. The industry’s average is outside your control.

---

The strategy line is the one Culturro line in next year’s plan with a number before the year starts. In the strategy month we score your shortlist against the 10X HR-AI Framework and sequence it. You leave with a written roadmap, a business case for each use case in the numbers your CFO tracks, and an owner on your side for every number. Each proof after it is a fixed fee quoted from the roadmap, and each build a fixed fee quoted after its proof.

The business case for your AI budget for HR starts with that roadmap. Book a 30-minute call and bring the numbers you already report on.

Frequently asked questions