2025 - the Moment Everything Changed
An HR director opens the Lattice dashboard. They notice the AI-powered functionality and think, "Okay, that's a nice feature."
Then they see the invoice. Then they put their face in their hands.
Six months later, that same HR director looks at a competitor whose AI features are included in the base plan by default. No extra fees. No separate "AI Package." It's just there.
That's not a coincidence. It's a signal that a paradigm shift has taken place.
What Happened: AI Stopped Being Optional
A few years ago, AI in HR software was positioned as a premium feature.
A luxury add-on. Something like leather seats in a car—nice to have, but not essential.
Companies paid $50–100 per employee per year for core functionality (assessments, PDPs, organizational structures) and viewed AI as an exotic extra.
Lattice was one of the first major players to recognize where things were headed.
In 2024–2025, Lattice made a move that only looks rational if you understand what comes next: They embedded AI directly into their entry-level plan.
Not as a separate "AI Premium" package. Not as an add-on. Into the plan that everyone pays for. At first glance, it looked irrational. Why reduce average revenue per user when you could sell AI as an additional module?
Because the base plan stops being attractive without AI.
Why It Happened So Quickly: The Vocational School Effect
By 2026, every employee uses between three and eight AI tools as part of their daily work.
This isn't a hypothesis. It's an observable reality: ChatGPT. Claude. Perplexity. GitHub Copilot. Specialized internal bots. People have become accustomed to AI being part of the background fabric of work itself.
Now imagine an employee opening a talent management system that feels like a web application from 2015. No AI. No assistant. No personalized recommendations.
When someone has Claude in one tab and your system offers checkboxes in a form, that's not a contrast. It's incompatibility.
Lattice understood this. If they didn't integrate AI into the core product, customers would eventually move to someone who did.
This wasn't a marketing decision. It was a survival decision for the category.
Pricing Strategy: How This Changes ROI
Let's talk about the economics.
The Old Model (2020–2023)
Base Tier: $50/user/year
└─ Оценки, PDP, org structure
Premium Tier: $100/user/year
└─ + Аналитика, интеграции
AI Add-on: +$30/user/year (опция)
└─ Рекомендации по развитию, автоматизация ревью
The problem: Only 10–15% of customers paid for the AI add-on.
Most customers chose the base plan and stayed there.
The New Model (2025–2026)
Base Tier: $65/user/year
└─ Assessments, PDPs, organizational structure,
AI assistant included by default
Premium Tier: $130/user/year
└─ + Advanced analytics,
Custom LLM,
Enterprise integrations
At first glance, ARPU appears to decline in the base tier. It looks like a strategic mistake.
In reality, it was cross-subsidization.
Here's what actually happened:
1. Reduced Churn Customers who previously viewed AI as too expensive no longer needed to make a separate purchasing decision. Retention improved by 40–50%.
2. A Clear Upgrade Path Customers experienced AI in the base plan and started asking: "What could we do with even better recommendations?". Conversion from Base to Premium increased by 2–3x.
3. The Data Network Effect When everyone uses AI, more data is generated. More data improves machine learning models. Better models produce better recommendations. Better recommendations improve the product. It's a self-reinforcing cycle.
4. A Market Signal The message is straightforward: "Without AI, you're no longer competitive." That puts pressure on Workday, Personio, and everyone else in the category.
Real ROI for Customers
Previously:
- The system helped users fill out assessment forms.
- PDPs were created manually by managers (4–6 hours per employee annually).
- Leveling and grading required HR meetings (40–80 hours annually per team).
Now, with AI included in the base plan:
- AI generates assessment drafts (30 minutes instead of 2 hours).
- AI recommends PDP items based on competencies (saving roughly 3 hours per employee).
- AI suggests grade levels with supporting rationale (reducing alignment time by about 50%).
For a team of 100 employees, that's 250–300 hours saved per year. At an HR labor cost of $50–100 per hour, that's $12.5K–30K annually. Moving from $50 to $65 per user costs approximately $1.5K per year.
The payback period is measured in weeks.
How Competitors Are Responding: Patterns in 2026
No one can afford to ignore this trend. Here's what's happening across the market.
Lattice
- AI is integrated everywhere: assessments, development recommendations, attrition prediction, talent discovery.
- Pricing remained stable while premium tiers expanded with cloud functionality and integrations.
- _Strategy: _dominate the mid-market segment (500–2,000 employees) through consistently high baseline quality.
15Five / Leapsome
- Started following the same approach by including AI in their base plans.
- Reduced prices by 15–20% as a defensive move.
- Challenge: they lack the data volume available to Lattice for model training.
- Positioning: "similar to Lattice, but cheaper"—effective in price-sensitive segments.
Workday / SAP SuccessFactors
- Attempted to add AI, but their underlying architecture works against them.
- Their core systems are built around jobs and organizational hierarchies rather than competencies.
- Adding AI on top feels like putting a bandage on a tank.
- Market perception: increasingly viewed as late-moving incumbents.
Eightfold / Gloat
- Positioned themselves as AI-first platforms from the beginning.
- But their AI offerings primarily target Fortune 500 organizations.
- Smaller companies compare them to platforms like Lattice that work effectively for organizations with as few as 50 employees.
- Challenge: expensive and difficult to scale downmarket.
Personio, Zelt, and Other European Vendors
- Adoption has moved more slowly due to the regulatory environment, particularly the EU AI Act.
- But they recognize the same reality: failing to integrate AI means falling behind.
- Approach: localize AI capabilities, ensure compliance, then scale.
Why This Trend Is Irreversible
There are three reasons why embedding AI into entry-level plans isn't a temporary strategy. It's the new baseline.
1. Customer Expectations Have Permanently Shifted
Every day, employees:
- Write emails with AI assistance.
- Debug code with AI assistance.
- Generate images with AI assistance.
Then they open a performance management platform, fill out forms manually, and are asked to pay extra for AI-assisted development planning.
That feels absurd. The expectation shift is permanent.
No company can retrain users to stop expecting AI. Trying to move AI back into an optional add-on would damage credibility.
2. Data Has Become a Competitive Asset
Previously: More users → more server load → higher costs.
Today: More users → more training data → better models → better recommendations.
By embedding AI into its base plan, Lattice effectively made a long-term bet on data. ARPU may have changed, but training data grew exponentially.
Organizations that fail to recognize this lose speed-to-insight.
3. Compute Costs Have Fallen
In 2020, processing a single assessment through an LLM could cost $0.10–0.50. Today, it's closer to $0.01–0.05.
This is similar to what happened with cloud infrastructure in the late 2000s. Once cloud became inexpensive enough, maintaining on-premise servers stopped making economic sense.
The same thing is happening with AI.
What This Means for Companies Evaluating HR Platforms in 2026
If you're evaluating HR software today, pay attention to the following signals.
Red Flags
❌ AI is sold as a separate add-on.
❌ The "AI assistant" is merely a chatbot that waits for user prompts.
❌ AI is advertised without specific use cases.
❌ The company still markets AI as a luxury feature.
Green Flags
✅ AI is embedded directly into core workflows (assessments, PDPs, recommendations).
✅ No additional payment is required for basic AI usage.
✅ Time savings are clearly quantified.
✅ The vendor openly discusses limitations and keeps humans responsible for final decisions.
Where HR Pulsar Fits Into This Landscape
We recognized this shift before the market officially acknowledged it.
Our Position
We're building a platform designed around competencies rather than job titles. That's an architectural choice. And it makes AI integration feel natural rather than bolted on.
Why does that matter?
Lattice integrated AI into an architecture built around jobs and organizational structures.
It works. But it's a bit like adding a passenger seat to a cargo truck—functional, but not elegant.
We designed a competency graph from the start. Within that model, AI helps:
✅ Match competencies (pgvector embeddings against 4,800+ role models)
✅ Recommend development paths (which competencies should be developed for a target role)
✅ Suggest career moves (internal talent marketplace: employee + AI toolkit → project)
✅ Manage hybrid teams (people and AI agents visible within a single system)
And this is not a premium feature. _ It's foundational._
We also made a decision that would be difficult for Lattice to make without rebuilding its core architecture: We embedded an enterprise AI tools registry. The Workforce Map doesn't just show people. It shows which AI tools are approved, who owns them, and where risk exists.
This is more than workforce management. It's visibility into the reality of hybrid teams, something no other platform currently provides.
The Brutal Math: What Happens Next
Let's extrapolate.
2026–2027 (Current Phase)
- Every top-10 HR platform will include baseline AI functionality.
- ARPU will decline by 20–30%.
- Retention will increase by 35–50%.
- Startups that fail to integrate AI will lose rankings and relevance.
2027–2028 (The Next Wave)
- AI systems will require specialized operators and configuration experts.
- New categories will emerge around AI accountability, governance, and auditability.
- The EU AI Act and NIST AI RMF will become operational requirements rather than optional frameworks.
- Companies will discover they're using dozens of AI tools that nobody tracks.
2028 and Beyond
- AI will become as invisible and as essential to HR software as SQL is to backend systems.
- The question will no longer be whether AI exists.
- The question will be how effectively it operates within your architecture.
- Companies that layered AI onto legacy systems will struggle.
- Competency-centric platforms with built-in hybrid workforce management will gain structural advantages.
For Builders: Why This Matters
If you're developing HR software today, here's what you need to understand.
1. Architecture Is Destiny
If your platform is fundamentally built around job titles, integrating AI will be painful. Start with a competency graph.
2. LLM Errors Are a Feature, Not a Bug
Treat AI output as a recommendation, not a decision. Humans must retain final authority. This reduces legal exposure and increases trust.
3. Data Flows Matter
Where does the AI get its data? From a single organization? Or from an anonymized population? The answer determines which features are possible. Self-hosted systems can never provide industry-wide benchmarking. That's not a design limitation. It's mathematics.
4. Open Source Signals Trust
Technical buyers evaluate open-source AI systems differently. Not as black boxes.
As tools they can inspect, understand, and extend.
# Пример: как мы думаем о встраивании AI в HRPulsar
class CompetencyMatcher:
"""
AI Fluency — это не магия. Это pgvector embedding.
Сотрудник имеет компетенции. Роль требует компетенции.
Косинусное расстояние между векторами = match score.
"""
def recommend_development(self, employee_id, target_role_id):
# LLM помогает сформулировать смысл рекомендации
# Но мэтчинг — это чистая математика
employee_embedding = self.get_embeddings(employee_id)
role_embedding = self.get_embeddings(target_role_id)
gaps = role_embedding - employee_embedding
# ^ Это — список компетенций для развития
recommendation = self.llm.generate(
prompt=f"Сотрудник {name} хочет перейти на роль {role}. "
f"Ему нужно развить: {gaps}. "
f"Напиши конкретный PDP."
)
# ВАЖНО: это — предложение, не приказ
return {
"recommendation": recommendation,
"confidence": self.calculate_confidence(gaps),
"final_decision_owner": "manager" # Не LLM!
}
An Honest Assessment: What Could Go Wrong
Embedding AI into the base plan isn't risk-free.
**Risk #1: **Margin Erosion Without Scale
If you've integrated AI but only have 50 customers, revenue decreases while LLM infrastructure costs increase.
You're underwater.
Mitigation: Integrate AI once you reach meaningful scale (roughly 200–500 customers with 50+ employees each). Until then, keep it optional or in alpha.
Risk #2: Poor AI Can Destroy Trust
If your LLM generates invalid leveling recommendations and a manager blindly follows them, you've created a potential legal problem.
Mitigation: Be transparent about limitations.
"AI recommendation" does not mean "final decision." Educate customers. Log everything.
Risk #3: Regulation Moves Faster Than You Do
The EU AI Act requires documentation for high-risk AI systems. The NIST AI Risk Management Framework is increasingly becoming the de facto standard. Ignoring compliance while integrating AI creates future pain.
Mitigation: Start compliance work now, in parallel with AI development.
Closing: This Isn't the End of the Story
AI in the base tier of HR software isn't the peak of the wave. It's preparation for the next phase.
Because within the next 18 months, another shift will occur: M*anaging hybrid teams (people + AI agents) will become a core capability rather than a peripheral feature.*
Companies that currently use AI merely as an assistant—"here's a development recommendation"—are only preparing for the transition.
When organizations begin working alongside real AI agents that actively perform work, workforce management systems will need to see and manage those agents the same way they manage people.
That's when the real competition begins.
Until then, if you're building or selecting an HR platform, remember: AI in the base plan is no longer a differentiator. It's the cost of entry.
Everything else comes down to execution quality and architectural decisions that either enable—or prevent—the next stage of evolution.
Interested in how we're building this inside HR Pulsar?
Open source. 228 endpoints. Fully typed OpenAPI.
Spin up Docker, import your employees, launch an assessment, and see for yourself.
If you don't like something, open an issue.
Use it completely free https://app.hrpulsar.com/
Top comments (0)