Headcount used to be a reliable proxy for capacity, but AI changed that. Now CFOs are being asked to underwrite a productivity assumption that is difficult to measure. The fundamental question is: Can the organization, as it stands, deliver what has been promised to the board?
This is a question for CHROs and HR leaders, too, because the answer depends on how people and AI share the work.
Answering is difficult because the evidence is spread across multiple systems that have never been joined for a single team member. And now, part of the workforce is no longer human at all.
What makes these questions answerable is an ontology of the business: a shared model in which CRM, ERP, HRIS, and operational tools are resolved to a single record and mapped to every employee or agent within a common data model. It’s the connective tissue that lets Finance and HR view the same data with unified definitions.
HiBob calls this Organizational Intelligence. It’s the direction we’re building toward for the future of work and the future of financial planning.
Key takeaways: The CFO’s challenge in the AI era
- Headcount is no longer a reliable proxy for capacity. AI has changed how much output each person produces. Gartner reports that headcount growth expectations fell from 6 percent to 2 percent while 75 percent of CFOs planned to raise technology budgets in 2026.
- Proving AI’s return is the CFO’s job, and most organizations can’t yet do it. Only 12 percent of CEOs say AI has delivered both cost and revenue benefits (PwC), and only 35 percent of Finance organizations say they’re effective at measuring the ROI of their AI investment (Protiviti).
- Last year’s productivity figure is no longer a safe number to divide by. Planning arithmetic like revenue target divided by revenue per rep worked when output per person barely moved. Productivity now moves at a different rate in every function, so the answer comes out wrong.
- Answering the board’s questions takes an ontology of the business. Work output, employee identity and cost, and financial results sit in separate systems that don’t agree on identity, and AI agents have no record at all. An ontology resolves them to a single record mapped to every employee or agent, so Finance and HR work from the same data.
- AI should interpret and draft, while a deterministic engine does the math and people decide. To win the audit committee’s trust, a platform has to separate AI reasoning from calculation. In Bob, the plan-card protocol keeps people responsible for the final merge.
- Capacity is the metric no one owns yet, which makes it a shared CFO and CHRO question. Whoever defines capacity when part of the workforce is no longer human sets the standard everyone else measures against. A CFO who plans around the capacity of the whole workforce, with the CHRO as a partner, could set the standard for the function.
Your AI budget is reshaping your headcount
According to Gartner, 75 percent of CFOs planned on raising technology budgets in 2026. Nearly half (48 percent) planned increases of 10 percent or more. At the same time, headcount growth expectations fell from 6 percent to 2 percent. That’s not a hiring freeze. It’s a swap.
Work that used to arrive as a person’s salary now arrives as a subscription and a compute bill for an AI platform. The cost of that work is split across three separate places in the P&L that no report puts back together. Every agent has a cost, an owner, and an output, but it sits on a credit card statement outside nearly every roster.
Picture a team of eight running a dozen agents alongside them. People cost is down, software spend is up, and few teams can say whether the work got cheaper or simply changed addresses.
The better question isn’t which line went down. It’s how much more each person, and each agent alongside them, is delivering.
The swap that broke the relationship financial plans rest on
Nearly every financial plan assumes that if the company wants more output, it will hire more people. But today, in the AI era, output per person moves on its own and at different speeds in different teams, so the numbers CFOs present can mean two things at once: a more productive company, or one that has stopped hiring.
Take revenue per employee. It rises when a company gets more productive and when it stops hiring: same number, opposite stories, and nothing in the report says which one it is.
HR leaders see the other side of that number. Is the team getting more productive, or just being stretched thin?
Many leaders believe AI is working. It’s the CFO’s job to prove it.
Only 12 percent of CEOs say AI has delivered both cost and revenue benefits, according to PwC, and 56 percent say they haven’t seen any real financial gain.
On top of that, a Protiviti research survey shows that only 35 percent of Finance organizations say they’re effective at measuring the ROI of their AI investment.
While most functions in an organization can say AI is helping, the CFO has to put a number on it, build the business plan around that number, and defend it when the board asks where it came from.
The two questions the board and CEOs are asking
- Is our AI spend returning more than we put in, and what does that mean for how many people we’ll need next year?
- Can we execute this strategy with the organization we have? If not, what has to change?
Now, answering either question means being able to say, “This person (or agent), in this role, at this cost, produced this much.” But for most of corporate history, we could answer questions like these with basic arithmetic. Take the revenue target, divide by revenue per rep, and the result is how many reps we have to hire. Check it against the budget, open the requisitions, and we’re done.
It worked because output per person barely moved from one year to the next, so last year’s figure was a safe number to divide by.
The revenue-per-rep divisor is no longer stable. Divide by last year’s figure now, and the answer is wrong. Today’s business reality doesn’t tell the CFO anything reliable about how productive people are, because productivity is moving at a different rate across functions.
The point isn’t to monitor individuals to get the answers. It’s to understand your capacity well enough to plan and to put the right tools and support behind the people doing the work.
Where the data lives today
| System | What it holds |
| Operational tools (CRM, code repos, service desk) | Work output |
| HR | Employee identity and cost |
| General ledger | Financial results |
Because this data is siloed across disconnected systems, no single platform can tell the complete productivity story.
Access isn’t the problem. Almost any CFO can get these exports today. The roadblock is that the systems don’t agree on identity: CRM users don’t match HR records, email addresses change, and contractors convert. HR exports only show today’s snapshot, erasing historical roles and costs.
Worst of all, AI agents have no record at all. Without a common key, Finance is left with disconnected files and no way to trace output to its true cost and owner.
Connecting these systems is now possible. Making that connection useful to the people who rely on it, from Finance to HR to business leaders, is the harder part.
What Finance has tried, and what’s still missing
AI adoption for financial forecasting rose from 58 percent to 76 percent in a single year, according to Protiviti. But faster forecasts haven’t answered the board’s and CFOs’ questions: What AI is really returning, and whether the organization can deliver what it has promised.
Many teams still don’t own a planning tool at all. For them, the company model lives in a workbook only a few people understand, so the real competition is Excel, increasingly with an AI assistant in it.
Microsoft’s own guidance emphasizes keeping humans in control of AI-assisted work. For its Copilot function in Excel, it recommends using ordinary formulas for any task that requires accuracy or reproducibility, and it advises against using AI output for financial reporting.
Its Copilot in Excel FAQ also advises against using Copilot for decisions in sensitive areas, and it specifically names finance.
For numbers that have to reproduce next quarter and survive an auditor, Microsoft’s guidance makes the case for something more deterministic.
This is why the conversation around AI in finance is shifting from whether to use it to whether it can be trusted. To win the trust of the audit committee, a platform must separate the reasoning of the AI from the calculation of the math.
What is required are agents that compute, not chat. In a model Finance professionals can trust, the AI never generates a number. It interprets human intent, translates that intent into structured queries, and passes those queries to a deterministic, version-controlled calculation engine. The engine does the math. The AI drafts the plan and explains the result. People decide what to do with it.
The missing layer: An ontology of the business
To answer the board, a CFO needs more than a data warehouse or another planning tool. They need an ontology of the business: a connected model of who does what, at what cost, and with what result.
Bob Finance combines the deterministic calculation engine of Mosaic, a cloud-based Financial Planning and Analysis (FP&A) platform, with HiBob’s employee record. This architecture strictly separates AI reasoning from mathematical calculation.
Large language models (LLMs) interpret intent, draft models, and explain variances, while the underlying metric engine computes all financial figures to support audit-level reproducibility.
An effective ontology connects the dots that siloed systems can’t. It binds a person’s (or agent’s) role, team, manager, cost, and tenure directly to the work they produced and the financial results that followed. Because the employee record is the spine of the organization, every piece of work can be traced back to a person or agent.
AI is only as good as the context it has. This infrastructure delivers context that turns raw data into insight leaders can act on.
This enables “Numbers that Confess.” With this ontology, every figure on a statement or dashboard becomes a living value. Clicking any cell shows its full lineage: the connected ERP transaction data, the active driver assumptions, the scenario version, the people who approved the inputs, and the agents that helped draft them. This shifts the finance function from manual “data work” to strategic steering.
This ontology isn’t just a data structure. It’s the semantic foundation that allows AI agents to “understand” the business. Without it, agents process context-free data.
With a unified ontology, agents inherit the meaning of the organization, including the relationships between roles, cost centers, and financial drivers. That gives them the context a senior human operator would bring, with people still making the call.
HiBob’s HCM module holds workforce data from the employee record, including roles, teams, cost centers, and tenure, performance, and compensation. It’s the backbone the rest of the ontology is built on. For HR, this means people data is what the plan runs on.
Governance: How the plan-card protocol keeps people in control
To keep this operating model secure, HiBob enforces a plan-card protocol (a multi-step workflow that ensures human oversight and approval before an AI agent executes a complex task). This means embedded AI agents in Bob are structurally blocked from altering the plan of record (i.e., the company’s approved financial and headcount plan).
Instead, agents work in sandboxed drafts, package their work into “plan cards,” and submit them to human operators. A human must review the row-level differences, assess the
FP&A workflows AI can help with
As AI capabilities mature, they’ll help improve FP&A processes and workflows. At HiBob, we know what the ideal integrated human-AI workforce looks like in practice, and we’re building toward it in Bob:
- Integrated headcount planning: Would use live roster data to forecast person-level costs as the roster changes.
- Automated financial reporting: Would generate statements automatically from connected actuals. The agent would help with account mapping and data onboarding.
- Agile scenario modeling: Would compose complex “what-if” scenarios without building new models. The agent would even “speak” new models into existence from approved intent.
- Variance analysis: Would decompose the “why” behind variances quickly. The agent would draft the commentary, automatically identifying and attributing drivers down to the transaction level.
- Secure budgeting: Would give budget owners a self-serve way to participate within a secure, permissioned framework. The agent would help draft and recalibrate budget proposals.
How Organizational Intelligence can change FP&A in the AI era
Picture the Finance team moving beyond spreadsheets and onto a governed model where AI does the assembly work and people make the decisions. Planning, reporting, scenarios, variance analysis, and the budget cycle all run on it.
What changes for the CFO is a plan built on capacity and context. For HR, it’s a workforce plan built on the same numbers as Finance’s.
On top of the business ontology, a deterministic metric engine would give the same figure every time. The AI would never generate a number. And because every metric would resolve against the employee record, dimensions the source systems never had could come built in: win rate by manager, gross margin by site, cost per ticket by tenure band.
The ontology we’re building toward would give AI agents the context a senior human operator would bring. For example:
- A mapper agent would understand your specific ERP schema and automatically resolve data joins that are easy to miss in spreadsheets
- A builder agent would understand the dependencies between your headcount plan and revenue model and draft models that are structurally sound
- An analyst agent would understand the context behind a variance and attribute financial drivers to specific human or agent actions
- A planner agent would understand budget constraints and draft proposals that respect the organization’s permissions and policy guardrails
By bridging the gap between raw data and business context, an ontology like this would empower agents to execute strategy, not just generate text.
AI can help identify risk before it reaches the P&L
Imagine alerts running in the background, continuously analyzing the interplay between HR reality (workforce data) and financial targets. Instead of lagging metrics, like a red dot on a dashboard that appears after the quarter has already drifted, alerts would act as proactive, early-warning sensors for the business. They’d bridge the critical gap between HR and Finance.
For example, if hiring in a revenue-generating team stalls, alerts wouldn’t just report “vacant roles.” They’d calculate the downstream financial impact. An alert might read, “Hiring lag in the SDR team now puts $1.5M in ARR at risk.”
Similarly, if attrition spikes in a critical department, the system would flag not just a turnover percentage, but a projected variance in the headcount budget, the resulting pressure on the delivery roadmap, and the people carrying it.
This approach would fundamentally change the CFO’s posture. Rather than reacting to variances after the close, the leadership team could identify risks while they are still forming, allowing for targeted interventions like reallocating budget, accelerating recruiting, or re-phasing hiring plans before any of it impacts a single line of the P&L.
Why measuring capacity matters now
Eighteen months ago, CFOs were handed this question and expected to deliver answers: How do you measure productivity in the AI era?
Most are still working on it. Many Finance teams still spend a disproportionate part of their week buried in reconciliation and manual work rather than strategic steering.
The business is changing faster than the close calendar, and many traditional planning tools, which see the workforce as a flat file and data as a series of disconnected tables, weren’t built for this.
The measure of capacity is currently unowned, and it may be the most important metric of the next decade. Whoever defines what capacity means when part of the workforce is no longer human gets to be the one everyone else measures themselves against.
CHROs are asking the same question from the people side: How do we divide work between people and AI? How can we build the roles and skills that make this division work?
This is why reinventing how we measure capacity is so urgent now. It’s hard to manage a workforce that’s part people and part agents with a planning tool that treats them as two different files. We need an operating model that integrates the employee record, the financial plan, and the permission layer into a single system. And we need the technology to support it.
Recommended For Further Reading
Why HiBob?
If you’re a CFO reading this, you may be wondering why you’d even consider FP&A software from a platform like HiBob. It’s a fair question.
Our core belief is that, in the AI era, financial planning works best when it’s combined with the context of employee data, productivity data from systems of work, and financial data.
This is what it takes to plan for a workforce of people and agents.
With the acquisition of Mosaic, we brought in pioneers of strategic finance to revamp FP&A around the employee record.
This approach gives AI more to work with. Traditional FP&A tools and disconnected systems tend to see surface-level metrics like attrition rates or financial totals. The qualitative layers, such as performance ratings, feedback, and whether turnover was voluntary or involuntary, usually live somewhere else. AI requires this complete operational and employee context to deliver truly effective insights.
Bringing the employee record, the financial plan, and the permission layer into one system does more than improve planning. It also keeps AI inside the same security model. Because the system inherits the platform’s native role-based permissions, a CFO would take AI out of the sandbox and into the audit committee with confidence. Agents would work with sensitive people and financial data under the same role-based rules as the people they assist.
Many organizations are looking to bring people and financial data together. The ontology we’re building at HiBob, as part of our direction toward Organizational Intelligence, is what that makes it possible.
Consider what happens when attrition on the Sales team doubles. Disconnected tools don’t integrate HR, CRM, and performance data, so it’s hard to see that the departing reps were top performers who left because of an unrealistic plan (such as over-relying on a partner channel that never launched), rather than individual performance.
Ultimately, bringing these dimensions together helps Finance teams move from reporting on the past to helping shape business strategy for today and tomorrow.
The finance function is at a crossroads. We can continue to plan against a disconnected headcount, or we can build a plan based on the actual capacity of our entire workforce. The CFO who makes that shift first, with the CHRO as a partner, won’t just improve their forecasting. They could set the standard for the future of the function.
So the question may be less “Why HiBob?” and more “Why plan without the context of your people?”
FAQs
You need to trace work output to a specific person or agent, along with their role and cost. Today that data is split across systems: operational tools hold work output, HR holds employee identity and cost, and the general ledger holds financial results. The goal isn’t to monitor individuals. It’s to understand capacity well enough to plan and to put the right tools and support behind the people doing the work.
They’re asking the same question from two sides: how much work a workforce of people and agents can deliver. CFOs have to put a number on AI’s return and build the plan around it, while CHROs are asking how to divide work between people and AI and how to build the roles and skills that make the division work. Both depend on one view of capacity, which is hard to get when a planning tool treats people and agents as two different files.
It’s a connected model of who does what, at what cost, and with what result. It resolves CRM, ERP, HRIS, and operational tools to a single record and maps that record to every employee or agent in a common data model. That gives Finance and HR the same data with unified definitions.
AI can help when it’s kept separate from the math. In the model described here, the AI never generates a number. It interprets human intent, translates it into structured queries, and passes them to a deterministic, version-controlled calculation engine. Microsoft’s own guidance recommends ordinary formulas for any task that requires accuracy or reproducibility, and people decide what to do with the result.
Through the Plan-Card Protocol. Embedded AI agents in Bob are structurally blocked from altering the plan of record. They work in sandboxed drafts, package their work into plan cards, and submit them to human operators, who review the row-level differences, assess the computed impact, and execute the final merge.
Financial planning works best when it’s combined with employee data, productivity data from systems of work, and financial data. HiBob acquired Mosaic to revamp FP&A around the employee record. Disconnected tools tend to see only surface-level metrics like attrition rates or financial totals, while qualitative layers such as performance ratings, feedback, and whether turnover was voluntary or involuntary usually live somewhere else.