DeepSeek AI Business Insights

This report from ICONIQ focuses on the "“how-to”: what it takes to conceive, deliver, and scale AI-powered offerings end-to-end", mapping out how what startups building AI products are doing - successfully or otherwise - based on a survey of 300 startup executives. They define "AI-native" as where the core product or business model is AI-driven, while "AI-enabled" adds AI capabilities to existing products or new non-core AI products. There is a lot in the deck, worth going through, but a few highlights: 🚀 AI-native companies scale faster and earlier. 47% of AI-native products are already in the scaling phase versus only 13% of AI-enabled products, showing structural advantages in speed and maturity. 💡 High-growth companies build more agentic workflows. 47% of high-growth firms are actively deploying AI agents in production, compared to 32% of other companies. 🧠 External AI is optimized for accuracy, internal AI for cost. Accuracy is the top model selection factor for customer-facing products (74%), while cost leads for internal tools (72%). 🧾 API costs are the biggest budgeting challenge. 70% of respondents rank API usage fees as the hardest infrastructure cost to manage, outpacing inference, training, and storage. 💰 Inference spending explodes post-launch. High-growth companies spend up to $2.3M/month on inference at scale—more than 2x that of their peers. 📊 Coding tools lead in real productivity gains. 65% ranked AI coding assistants as the top driver of productivity, with high-growth companies reporting 33% of code written by AI. 📈 AI engineering headcount is rapidly increasing. High-growth companies expect 37% of engineering roles to focus on AI in 2026, up from 28% in 2025. 🧩 Open-source and inference optimization are key cost controls. 41% are switching to open-source models and 37% are optimizing inference efficiency to combat spiraling costs. 🏷️ AI pricing is still immature and mostly bundled. 73% of AI-enabled companies either include AI in premium tiers or at no extra cost, but 37% plan to revise pricing based on usage or ROI. ⚖️ Explainability is a critical barrier to trust. 42% of companies cite explainability as a top-3 deployment challenge, especially in regulated sectors. 📉 Only half of employees use AI tools regularly. Despite 70% of employees having access to AI tools, only 50% use them consistently—dropping to 44% in $1B+ enterprises. 🧪 Monitoring is common but automation lags. 75% of scaled AI products include advanced monitoring, but few teams have fully automated retraining pipelines. 🛠️ Proprietary models are a high-growth differentiator. 54% of high-growth firms fine-tune foundation models and 32% build proprietary models, compared to 32% and 20% respectively among others. 📦 AI-native firms build more agentic and vertical tools. 79% of AI-native firms focus on agentic workflows, while 56% also build vertical applications tailored to specific industries.

Most companies think they've become AI native the moment their product has an AI feature. Or when the engineering team starts using Copilot to write code faster. That's not AI native. That's AI at the edges. A company truly becomes an AI native when the person running payroll uses it to flag anomalies before they become problems. When the account manager uses it to prep for a client call using data they technically had but never had time to read. When the ops team automates the weekly report that used to eat three hours every Monday. It's not about the product. It's not about the tech team. It's about every person, every function, every decision loop inside the company. Here's what that actually looks like and something that we are trying at Unstop. Still work on progress. - Finance stops doing manual reconciliations and starts asking why did this number moved because AI already flagged it. And gives me a picture on my cash runway automatically. - Sales walks into every conversation knowing which accounts are at risk, what they bought, what they ignored, and what might close next without digging through a CRM. And everything about that company. - HR isn't just posting jobs faster. They're using data to understand what actually predicts retention vs. attrition in their own workforce - Customer success catches problems before the customer writes in because the signals were always there, just unread. - Leadership makes fewer gut calls and more pattern-based ones, not because they got smarter, but because the data they already had finally became usable Most companies are not data poor. They're insight-poor. The data exists. The problem is that no one had the time or the tools to turn it into a decision. AI closes the gap not just for the product team, not just for engineering. For everyone. That's what #AI native actually means.

Are AI Native companies today showing us the future of the org chart? If you want to know what org charts will start to look like in a few years, consider looking at what the top AI Native companies look like today. —————————— We analyzed functional distributions across 1M+ employees in Pave's real-time dataset, comparing the top AI-native companies (70% of the Forbes AI 50 participate in Pave's data including leaders like OpenAI) against the broader non-AI tech market. 𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲 𝗔𝗜 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 𝗼𝗿𝗴 𝗰𝗵𝗮𝗿𝘁 𝗯𝗹𝘂𝗲𝗽𝗿𝗶𝗻𝘁 𝗹𝗼𝗼𝗸𝘀 𝗹𝗶𝗸𝗲: 1️⃣ 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗱𝗼𝗺𝗶𝗻𝗮𝘁𝗲𝘀. ~50% of headcount at top AI companies sits in engineering vs. ~37% at non-AI tech. 2️⃣ 𝗦𝗮𝗹𝗲𝘀 𝗵𝗼𝗹𝗱𝘀 𝘀𝘁𝗲𝗮𝗱𝘆. ~19% at AI companies vs. ~20% at non-AI tech. Even the most AI-native companies are investing in humans to sell their offerings. The product may be AI, but the GTM motion is still involving humans. 3️⃣ 𝗖𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗦𝘂𝗽𝗽𝗼𝗿𝘁 𝗴𝗲𝘁𝘀 𝗰𝘂𝘁 𝗶𝗻 𝗵𝗮𝗹𝗳. ~3% at AI Native companies vs. ~7% at non-AI tech. 4️⃣ 𝗙𝗶𝗻𝗮𝗻𝗰𝗲, 𝗠𝗮𝗿𝗸𝗲𝘁𝗶𝗻𝗴, 𝗮𝗻𝗱 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗹𝗹 𝗿𝘂𝗻 𝗹𝗲𝗮𝗻𝗲𝗿. Each of these functions runs 1-2 percentage points thinner at AI companies. Not dramatic individually, but collectively it paints a picture: AI-Native companies are concentrating headcount in builders and sellers and generally compressing everywhere else. —————————— 𝗧𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝗿 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻: We recently showed that the entry-level IC workforce is shrinking across all of tech, and that AI-Native companies had ~13% more senior ICs and ~16% fewer junior employees: https://lnkd.in/gYHfxv2Y Today's data adds another dimension. It's not just the level mix that looks different at AI companies. It's the entire functional shape of the org. More engineers. Similar sales investment. But fewer support staff. And a leaner back-office.

Building an AI-native company doesn't mean using AI tools. It means the company wouldn't make sense without them. I'm building something new. Two people. And I'm not going to pretend that's the interesting part. The interesting part is what it looks like under the hood. We have a full CRM dashboard. A client portal. An admin portal. A task engine with natural language input. Pipeline tracking. Revenue analytics with conversion funnels and loss pattern analysis. Automated client research. A consulting process orchestrated by 12 AI skills that chain together. Our company history updates itself every night. Our proposals check what won in the same industry before suggesting a price. Our clients log into a portal that's always current — and nobody on our team manually updates it. I didn't plan this as a series. But every time I describe one of these systems, the response is "wait, how?" — so I'm going to spend the next 4 weeks showing exactly how. Two posts a week. No theory. Just the actual systems running inside a two-person company that has no business having this much infrastructure. If you're building something lean and using AI as more than a chatbot, I want to hear about it. What's the one system you've built that makes people say "wait, there's only two of you?"

While Palantir, OpenAI, and Anthropic generate headlines with their exponential ARR growth, most private companies at the intersection of SaaS and AI struggle to optimize their Go-to-Market strategy. Perfecting GTM is particularly vital for companies beyond the 180 unicorns—those aiming to reach the $100M ARR milestone. Here's an insightful report on the current GTM landscape, especially relevant as vertical SaaS companies increasingly shift toward AI and AI startups pivot from consumer to enterprise markets. Key takeaways from ICONIQ: 🔶 AI-Native vs. Traditional SaaS: Performance Gaps Widening 🔸 AI-Native Outperformance: AI-Native companies significantly outperform peers in conversion rates, especially in the free trial/POC stage. Faster ROI and clearer value help close deals despite market headwinds. 🔸 Team Structure Evolution: AI-Natives allocate more headcount to Post-Sales teams (e.g., forward-deployed engineers supporting customer onboarding/adoption), optimizing for long-term customer value. Non-AI firms are embedding CS functions throughout the GTM org, moving away from standalone CSM teams. 🔶 GTM Motions: Multi-Channel, Hybrid, and Partnership-Driven 🔸 Hybrid Motions Rising: There is a pronounced shift toward blended top-down and bottom-up customer acquisition, reflecting the need to engage multiple stakeholders. 🔸 Partnerships as Key Levers: Investing early in partner ecosystems pays off as companies scale: >80% of $25M+ ARR companies derive at least 10% of revenue from channel sales. 🔶 Internal AI Adoption: Foundation for Lean, High-Performance Go-To-Market 🔸 AI as a Team Multiplier: Founders who invest in embedding AI into GTM operations (especially in Marketing, SDR/BDR, and AE teams) see marked productivity and efficiency gains. 🔸 Core Use Cases: Lead generation (61%), content/campaign creation (58%), and meeting transcription/analysis (71%) are the most common entry points for GTM AI—start there if you haven't already. 🔶 Key recommendations for founders and growth teams: 🔸 Benchmark AI Maturity: Honestly assess where your GTM org stands on AI adoption. Prioritize embedding AI in lead gen, content, and sales workflow automation. 🔸 Invest in Technical Post-Sales: As products become more AI-powered and complex, ensure support and onboarding teams are staffed with technically adept talent who can drive value and adoption. 🔸 Double Down on Partnerships: Build out your channel strategy early, even modest revenue from partners signals scalability and can de-risk revenue concentration. 🔸 Innovate on Pricing: Consider hybrid models if appropriate for your product, especially for AI solutions. 🔸 Track the Right Metrics: Focus not just on lagging indicators like ARR and NRR, but also top/mid-funnel conversion, pipeline coverage, and leading indicators of GTM health (AI adoption, team efficiency, and partnership contribution). #gotomarket #GTM

The question I ask every morning isn't "what can #AI do today?" It's "what should I still own?" Running an AI-native company looks nothing like using AI tools. The difference is architectural. Tools augment tasks. An AI-native operating system — what we're building at Dorje AI — coordinates agents across the entire workflow. One handles data analysis. Another drafts and debugs code. Another processes documents before they reach a human. Another clears the queue so I can focus on decisions that require judgment. When you orchestrate five specialised agents simultaneously, the founder's job transforms. You stop being a doer and start being a director of judgment. That shift is harder than it sounds. Every morning, I run a mental triage: → What gets fully automated? → What gets delegated to an agent with human review? → What requires my judgment — and why? → What should never touch an agent at all? The third and fourth categories are where most founders underinvest. Not everything is meant to be optimised. Client relationships. Strategic ambiguity. The call that has to land right. The decision where being wrong carries consequences no SLA can absorb. Some days this feels magical — five parallel streams compressing two weeks into two hours. Some days it's chaos — catching the edge case an agent confidently got wrong, rebuilding trust in a pipeline that surprised you. Most days, it's both. That's what building on the frontier actually feels like. Across the enterprises we work with across Asia-Pacific, the pattern is consistent: the firms winning with AI aren't the ones with the most tools. They're the ones who made an architectural decision — humans own judgment, agents own execution. The next generation of companies won't say "we adopted AI." They'll say "we redesigned the organisation around it." The question isn't whether your company will run on an AI-native operating system. It's whether you'll design that system — or let it design itself by default. AI + Human. Not a platitude. An engineering problem.

AI agents are taking digital transformation to the next level, redefining enterprise automation and unleashing innovation with new possibilities. While robotic process automation (RPA) excels at automating repetitive, rule-based tasks in structured environments, agentic process automation (APA) addresses the key limitations of RPA. AI agents bring in context aware reasoning, adaptability, and decision-making capabilities that support dynamic optimization and autonomous workflows.     In a new Deloitte report, my colleagues Prakul Sharma, AJ M., Patricia Henderson, and Camille Chicklis examine the innovative strides AI agents are making in automation and consider what the future holds.     Today’s AI agents are capable of understanding environments, intentions, and adapting to user preferences. In the future, we’ll see multi-agent collaboration, self-healing correction of inefficiencies, and generalist GenAI systems working as strategic advisors across multiple domains. AI agents will expand automation into other business areas, complementing rather than replacing existing technologies. Future AI agents will manage strategic negotiations, supply chains, investment management, and customer experiences, shifting the focus of human work to higher value activities like oversight, ethical decision-making, and continuous innovation.     It’s incredibly rewarding and awe-inspiring to collaborate with organizations as they embrace these innovations and unlock new efficiencies and resilience.

You’re not behind on AI. You’re just not racing a Formula 1 car on a go kart track yet. That’s fine. Most companies aren’t either. I spoke with Chadi Nicola recently and it got me thinking about what it actually takes to bring AI into a business and spoiler alert: it’s not just throwing tools at the problem. It’s about maturity. Not the emotional kind. The Gartner AI Maturity Model kind. Here’s how the journey plays out: Level 1: Awareness AI sounds impressive. People talk about it. But no one’s doing anything yet. • No budget • No tools • Still asking “Can we trust this?” Level 2: Active The pilots begin. One team’s trying Copilot. Another’s playing with prompts in a spreadsheet. • Siloed experiments • Trial and error • Governance? Not a chance Level 3: Operational AI is doing actual work now. It’s not just a pet project. • Multiple teams involved • ROI starts showing up • Internal champions emerge Level 4: Systemic AI is no longer bolted on. It’s built in. • Governance is real (ethics, compliance, risk) • Data pipelines are stable • AI shapes both operations and product Level 5: Transformational This is where things get interesting. AI isn’t helping the business . It is the business. • New revenue streams • Fast innovation cycles • Culture and talent centred around AI Most companies? Somewhere between Level 2 and Level 3. And that’s perfectly normal. It's up to your leadership team to take the right steps to take them higher. Because the hardest part isn’t buying the tech. It’s building the capability. Shifting how people think, work, and make decisions. So, that said, where would you place your organisation today? 👇 Let’s talk.

What is an AI-First company? One where AI isn’t a side project, a bolt-on feature, or a cost-cutting tool; it’s the primary engine for how the company creates value, makes decisions, and operates. Instead of asking “Where can we apply AI?”, AI-First companies ask “How would we design this product, process, or business model if AI were native from day one?” Here’s what defines them: 𝟭. 𝗔𝗜 𝗶𝘀 𝗰𝗼𝗿𝗲 𝘁𝗼 𝘁𝗵𝗲 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗺𝗼𝗱𝗲𝗹 Products and services are built on AI capabilities, not just enhanced by them. Revenue and competitive edge depend on AI performance. 𝟮. 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻-𝗺𝗮𝗸𝗶𝗻𝗴 𝗶𝘀 𝗔𝗜-𝗮𝘂𝗴𝗺𝗲𝗻𝘁𝗲𝗱 Strategic, operational, and customer-facing decisions leverage AI insights. Leaders treat AI outputs as first-class inputs in planning and execution. 𝟯. 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 𝗮𝗿𝗲 𝗿𝗲𝗱𝗲𝘀𝗶𝗴𝗻𝗲𝗱 𝗮𝗿𝗼𝘂𝗻𝗱 𝗔𝗜 Processes are re-engineered so AI can operate at the center; removing human bottlenecks, not just automating existing steps. 𝟰. 𝗗𝗮𝘁𝗮 𝗶𝘀 𝗮 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝗮𝘀𝘀𝗲𝘁 The company actively designs systems to collect, structure, and own proprietary, high-quality data to feed models. 𝟱. 𝗖𝘂𝗹𝘁𝘂𝗿𝗲 𝗶𝘀 “𝗔𝗜-𝗹𝗶𝘁𝗲𝗿𝗮𝘁𝗲” Everyone, from executives to front-line staff, understands AI’s capabilities, limits, and ethical implications. What critical parts of your business today cannot run without AI?

18 months as VP of customer success at an AI-native company has fundamentally changed how I think about this role. Here are 5 key distinctions from traditional SaaS that every CS leader should know if they step into an AI-native company: ✨Model-driven evolution. You're not just deploying code updates—you're working with evolving AI models that fundamentally shape customer experience. Every model improvement (or regression) directly impacts how your product performs in production. ✨AI-fluent talent is non-negotiable. You need team members who combine core CS skills with genuine curiosity about AI and technology. The space evolves too rapidly for passive learners. Success requires comfort with experimentation, ambiguity, and continuous learning. ✨ Embrace imperfect outputs. AI won't behave perfectly 100% of the time. You're working with probabilistic systems, not deterministic code. Your team needs to help customers set appropriate expectations and build trust through transparency, not promises of perfection. ✨ Strategic AI literacy matters at every level. When I joined Siena AI, I initially dismissed the importance of tracking LLM releases like Claude updates—that seemed like an engineering concern. My CEO pushed back, and he was right. Model releases can dramatically shift what's possible for our customers overnight. As a VP, you need this literacy. We use AI in everything we do: from renewal proposals to churn reviews. Every day we come up with unique ways to improve our effectiveness with AI. ✨ The mission shifts from adoption to transformation. It's not enough to drive product adoption. You're helping customers fundamentally transform how they operate—reimagining processes, redefining roles, and building AI-native organizations. That requires a different level of strategic partnership. The future of CS in AI native organizations isn't just about helping customers use software. It's about guiding them through one of the most significant technological shifts in decades. ****** Chad is the author of "The Strategic Customer Success Manager". Follow me here or at Customer Success & Failures: https://lnkd.in/ew6Jhx-w