
Article
No manager ever asked for a dashboard
Total Rewards across the five levels of AI maturity, read by the three people who feel the result: the HR seat, the manager and the employee. Article 1 of the series AI-native HR seen by the people who use it. Available in Portuguese.
Published Aug 17, 2026
What our yardstick missed
Traditional HR was designed to administer the masses, enforce rules and sustain command and control, which is what the technology of the time allowed at scale. Each layer that came after added more policy, more approvals and more control. When software arrived, it optimized that design instead of replacing it, and it was built to make life easier for the people who work in HR (payroll, recruiting, compensation). For the manager, each of these waves arrived as one more screen to log into to find the data for a decision they already knew they would have to make.
In May we published the AI Maturity Map, with the five levels we have used since to diagnose a function. The ladder stands and the yardstick works, and it is what supports this article. What was missing was a second reading: we described the journey by HR subfunction (compensation, recruiting, performance) and measured progress by how much HR’s own work improves. We argued for HR as a product while measuring the experience of the people who operate the product, with little room for those who use it.
- 92% of the HR functions we assessed are at L1 or below, when each level is verified capability by capability. Zivo diagnostic, Aug 2026.
- 8% reached L2, where the AI gain stops being individual and becomes the function’s standard. Zivo diagnostic, Aug 2026.
- 0 functions verified at L3 or above. The levels where managers and employees feel the difference remain empty. Zivo diagnostic, Aug 2026.
This series fixes that. Same ladder, same thesis, with the outcome of each level read from three places at once: the HR team, the manager and the employee. It starts with Total Rewards, because that is where the friction is oldest and where getting it wrong costs the most.
If HR is a product, who is the user?
It is worth unpacking this part of the thesis, because it usually goes by quickly and it holds up the rest of the article.
Treating HR as a product has little to do with adopting product vocabulary. Roadmap, discovery and backlog are consequences. What defines a product is that it exists to solve someone else’s pain, and that is where the analogy starts to cost us.
- The manager End user Decides about people every week, almost always with incomplete information, and is the one who defends the decision in front of the team.
- The employee End user Lives with the outcome: the band they were placed in, the package they received, the explanation they heard or didn’t hear.
- The HR team Who operates it (and signs the contract) A good part of the HR tech market was built for them, because they are the ones who buy. The side effect: the yardstick of success became the comfort of the operator, and the gain at the edge became a second-order concern.
Neither of the first two works in HR, and neither asked to be put into yet another system.
That is why this article reads the five levels three times. If a maturity gain changes nothing for the manager or the employee, it improved the function’s operation, and that counts for less than we usually say. When the gain shows up in all three readings at once, the investment is easy to defend internally: the function gains capability, the manager decides better and the employee feels the difference in their own package.
Why Total Rewards first
Total Rewards was designed around averages, annual cycles and standardized programs because that was the only model that could run at scale, not because it was the best way to reward people. Market data came as a snapshot, equity was audited when someone asked, and the merit budget came out once a year (twice, at the more organized companies). Benefits were designed per persona because building and administering a proposal per person was unworkable operationally and, in some cases, legally.
AI removed that constraint, and without it you can decide compensation case by case, at the moment the case comes up. Written like that, it is HR jargon.
The translation that matters
For the manager, it means no longer finding out they lost someone over money only after the resignation.
For the employee, it means no longer depending on an annual window to get their own package reviewed.
Total Rewards, level by level
On to the five levels. The ladder is the same as the AI Maturity Map, and at each one the question repeats in three forms: what changes for the HR team, what changes for the manager and what changes for the employee.
L1 · Individual productivity
The gain that does not cross the border
At L1, someone on the compensation team has figured out how to use AI. They put together the positioning analysis in minutes, write the leveling justification without opening the spreadsheet and close the total package cost calculation, with payroll taxes and benefits, in an afternoon.
For the manager, none of this exists. They ask for the information by email or chat, wait, and get a spreadsheet or a PDF, with luck a day sooner. For the employee, even less: they don’t know there was anything to change, and when they find out they were out of band, they almost always find out by accident.
Email · manager’s outbox · Mon, 2:32 pm
To: remuneracao@empresa.com | Can you send me Maria’s position in the band and her total cost? I need it for Thursday’s conversation.
awaiting reply · 2 business days
planilha_maria_v3.xlsx · received Thursday, 4:51 pm (the conversation was at 2 pm)
The AI gain exists inside the function. From the outside, nothing changed.
Level 1 illustration. Fictional data.
This is the level where most of the HR functions that went through our maturity diagnostic sit, and it is also where the illusion is greatest, because the improvement the function feels is real and measurable, while the path between the manager and the decision stays the same.
Where HR functions stand, when each level is verified capability by capability (Zivo diagnostic · Aug 2026)
- L1 or below (Individual productivity): 92%
- L2 (Team productivity): 8%
- L3 (Contextualized operating system): 0%
- L4 (Decision intelligence): 0%
- L5 (Adaptive intelligence): 0%
No function we verified reached L3. The levels where managers and employees start to feel the difference remain, in practice, unexplored.
For whoever leads People, the problem with this level is that the gain stays locked in the person who found it and leaves when they do.
L2 · Team productivity
The same answer, no matter who you ask
At L2, the rule leaves the head of the person who knows it and becomes a shared skill: merit distribution doesn’t close if it blows the function’s budget, total cost always comes out with payroll taxes, provisions, benefits and long-term incentive built in, and nobody moves forward without a justification per person. One HR person’s learning becomes the whole function’s learning, and the criterion becomes an institutional, auditable asset that survives turnover and stops depending on who is on call.
Internal chat · any BP on the team · single rule
How much does it cost, in total, to promote Diego to senior?
running skill · total-package-cost
Total cost: +R$ 38.2k/year (salary, payroll taxes, provisions, benefits and long-term incentive). Fits within the function’s budget. Written justification still needed to proceed.
Any BP, same yardstick. The criterion stopped varying.
Level 2 illustration. Fictional data.
For the manager this shows up as consistency, and it is worth more than it seems. Before, the answer depended on who they talked to: one BP calculated cost one way, another a different way, one asked for a justification, another didn’t. Over time the manager learned whom it was worth talking to, and that is how HR taught the business to work around it. Now the yardstick is the same. They may not like the answer, but they don’t have to decode who it came from.
For the employee, the effect is quiet and important: whoever delivers better earns more than whoever just negotiates better. The criterion is still invisible to them, with the difference that it now applies equally. And the manager still asks and waits in the same channel at the same hour.
L3 · Contextualized operating system
Asking stops costing something
From L3 on, there is a layer connected to payroll, the band table, performance history, the budget and the market, and it answers the manager directly.
People layer · chat · 5 connected sources
Where is Maria in her band, how much of her equity has vested, and how does her total package compare with her internal peers?
Maria is at 82% of the midpoint of band 7. Equity: 50% vested, next cliff in November. Total package 6% below the median of the 8 comparable internal peers.
Sources: payroll · band table · equity plan · performance · market benchmark
Sunday, 11:07 pm · answered in seconds
The request died, and with it the cost of asking.
Level 3 illustration. Fictional data.
The answer comes with the source of every number, without anyone exporting anything. The manager asks before the career conversation, not two weeks after promising to check. If the answer doesn’t add up, they rephrase and ask again, without feeling like they are spending someone’s goodwill.
What dies at this level is the request, and with it the cost of asking, which nobody measures. A good share of bad management decisions comes from data being expensive to get. If getting the information means sending an email, waiting and asking for a favor, the manager decides without it.
For the employee, the gain is hearing an answer with substance, not getting one more screen. Transparency is not publishing the band and letting everyone compare themselves alone. It is understanding why you are where you are and what would change your position. That is sustained in a conversation with whoever leads, and what L3 changes is that this conversation stops running into “I’ll check.”
For whoever leads the function, the gain is of a different order. Analyses that wouldn’t have fit in the month start to exist, and you can answer the CEO in the meeting itself, with the source of every number, instead of promising a deck for the following week.
One caveat: at this level the person asking the question is still the manager, and what the layer solved was access to data. It answers what they ask and doesn’t point out what they failed to ask.
L4 · Decision intelligence
The decision arrives before the request
But here, at L4, the layer stops waiting. The manager gets a message on Teams or Slack:
Message · Tuesday, 9:12 am · arrived on its own
Maria is at 82% of the midpoint, below comparable internal peers, with strong performance in the last two cycles, and her role has moved in the market. A 9% adjustment would take her to 89%, within the approved budget, company policies and equity limits.
Do you want to approve, adjust or discard?
Approve · Adjust · Discard
Nobody asked for the analysis. It arrived ready, with its source, in the channel where the manager already was.
Level 4 illustration. Fictional data.
The proposal is not always a raise. Sometimes it is an equity refresh, a one-off bonus or a specific training, when that is what solves the case. Three things change places at this level, and the analysis itself is not one of them.
- The cycle loses its calendar owner For decades we organized compensation around the annual merit cycle because it was the only viable way to process thousands of decisions: collect market data, freeze the budget, build the spreadsheet, ask for recommendations, calibrate and push to payroll. With that cost lower, any relevant event (a promotion, a big change in scope, a premium for a scarce skill, growing inequity, flight risk) can trigger a recommendation. The cycle still exists, but it is no longer the only moment when the company can make a smart decision.
- Who starts the conversation changes Before, the decision only existed if the manager remembered to raise it. Now it arrives, and that changes what they are held to, because “I didn’t know they were out of band” stops working as an explanation.
- HR’s work flips toward judgment The function doesn’t get smaller. Instead of spending the month building the analysis, which now arrives built, it spends the month on what requires judgment: compensation philosophy, business strategy, talent priorities and appetite for differentiation. It is the senior part of the work, and precisely the part the cycle calendar always squeezed.
The manager remains the owner of what the layer can’t know: that the person said last week they want to change areas, that there is an agreement made in a conversation nobody recorded, or that this adjustment, on that team, in that month, breaks a balance that isn’t written anywhere.
For the employee, this is the level where they are seen without having to raise their hand. The inequity that took a year to be noticed is flagged while it is still small, and recognition arrives without depending on them holding a competing offer.
L5 · Adaptive intelligence → frontier
The end of the average
At L5, personalization leaves the pitch deck. For years Total Rewards has talked about it while running standardized programs, and there was a good reason for that: the cost of personalizing at scale. That cost has dropped.
Two people in the same scope may value very different combinations of cash, equity, flexibility, benefits, development and long-term incentive. A layer with judgment understands that preference and recommends different combinations within the same philosophy, the same total cost and the same governance rules.
Package design · same band, same total cost · 2 proposals
Person A
values liquidity and routine
Cash · 70%
Equity · 15%
Flex benefit · 15%
Person B
values upside and career
Cash · 45%
Equity · 40%
Development · 15%
Standardization moves out of the package and into the limits: philosophy, cost and governance stay the same.
Level 5 illustration. Fictional data.
The layer also learns from what happened after each decision it proposed: who stayed, who left, the effect of merit distribution on retention, which benefits cost more and get used less. If the recommendations in a category produced higher-than-expected attrition, it adjusts its own calculation. This learning is what speeds up fine-tuning, because each round of decisions calibrates the next proposal better than the last.
For the manager, the effect is to stop administering exceptions, and conversations about why the package can’t be different shrink, because the package is born different. For the employee, the destination is to feel correctly incentivized, recognized and paid, which is what the function set out to deliver from the start. And for whoever leads the function, policy corrects itself from the results of its own decisions, which even changes what gets measured: retention and equity as outcomes, instead of how many cycles were closed.
Frontier territory: no HR operation we know of works like this today, and what blocks it is measurement. Confidently attributing a business outcome to a specific decision remains unsolved at scale.
The three readings, side by side
| Level | For the HR team | For the manager | For the employee |
|---|---|---|---|
| L1.Individual productivity | A real gain that is also invisible: it stays locked in whoever found it and leaves with them. | Nothing changes. They keep asking and waiting, with luck a day sooner. | Nothing changes. When they find out they were out of band, they find out by accident. |
| L2.Team productivity | One person’s learning becomes the whole function’s learning, and the criterion survives turnover. | The answer stops depending on who they talked to. | Whoever delivers better earns more than whoever just negotiates better. |
| L3.Contextualized operating system | Analyses that wouldn’t have fit in the month start to exist, and the decision is grounded on the spot. | They ask instead of filing a request, and asking stops costing something. | Transparency becomes possible, and policy decides whether it arrives. |
| L4.Decision intelligence | The team judges recommendations instead of producing analysis, and returns to the senior part of the work. | The decision arrives before they realize they needed it. | They are seen without having to raise their hand or hold a competing offer. |
| L5.Adaptive intelligence→ frontier | Policy corrects itself from the results of its own decisions. | They stop administering exceptions. | They feel correctly incentivized, recognized and paid. |
Total Rewards across the five levels of the AI Maturity Map, read by the three people who feel the outcome.
The three columns advance together, and that is what makes the decision to invest easier to defend internally. A gain that shows up only in the first column improves the function’s operation and stops there.
What this asks of the HR team
If the unit of measure changes, where it is worth investing changes too.
- Measures the function’s activity “How many skills did the compensation team create this quarter?” An L2 metric disguised as strategy: it counts effort and stops there.
- Measures what the business solves on its own “How many managers asked a compensation question this month, and how many got the answer without opening a request?” Measures the gain that crossed the function’s border.
The manager also doesn’t arrive with a compensation problem on the table. They arrive with a person, who is at once a salary band case, a performance case, a flight risk case and a succession case. When HR hands that back in four separate conversations, with different owners and systems, it falls to the manager to piece the context together by hand. In practice, they are the middleware of our architecture.
In practice, the manager is our middleware, and nobody asked for their opinion.
It is also worth pointing out what our own yardstick didn’t measure: when we first published the five-level framework, in May, we organized the ladder by HR subfunction and confirmed progress by how much HR’s work improved. That ladder is still right (and what the diagnostic evidence shows is that almost everyone is at the start of it), but we failed to also confirm the level by what improved in the lives of people who no longer needed to talk to HR about certain topics.
The goal was never to automate the compensation decision. It was to automate everything that has to happen so that whoever decides can decide better, and whoever decides, in most cases, is not HR.
References
- Zivo, AI maturity diagnostic: base of HR functions assessed with the diagnostic tool, snapshot of 08/10/2026, measured capability by capability per level.
- Zivo, What it means to be AI-native: 5 levels of organizational maturity in AI adoption, white paper, May 2026. Available at zivo.ai/ai-maturity-map.