AI Category Creation Strategy: When to Create a New Category vs Compete in an Existing One
TL;DR
Category creation is the glamorous strategy — but it's expensive. Drift, Gong, and Datadog each burned $50M+ before category-defining payback. Entering an existing category (PLG into a known buyer pain) is faster and capital-efficient but caps your terminal multiple. The decision should be data-driven, not founder ego: it depends on buyer awareness, incumbent strength, your AI advantage, and how much capital you actually have. Cursor entered an existing category (code editors) and won. Decagon arguably created one (AI agents for support). Both can be right — choose deliberately.
What Category Creation Actually Costs
The Play Bigger / category-king narrative — popularized by Drift, Gong, Snowflake, Datadog — undersold a critical detail: category creation is expensive in capital, time, and team. Before you decide to create a new category for your AI product, know what the path actually costs.
Drift (Conversational Marketing)
Investment: Raised $107M before profitability. Spent an estimated $40M+ on category-defining marketing — books ("Conversational Marketing"), conferences (HYPERGROWTH), branded podcast, owned media. 4+ years to category leadership.
Outcome: Acquired by Vista Equity in 2021 for ~$1.2B. Category exists, but several competitors (Intercom, Zendesk) absorbed the playbook.
Gong (Revenue Intelligence)
Investment: Raised $584M total. Created "Revenue Intelligence" from a more crowded sales-tools space. Heavy investment in analyst relations (Forrester Wave creation), branded research, conferences.
Outcome: Last private valuation ~$7.25B (2021). Revenue Intelligence is now a Gartner-recognized category. Total time to category-creator status: ~5 years.
Datadog (Observability)
Investment: Reframed "monitoring" into "observability" with active analyst push and a re-engineered platform pitch. Public company since 2019. Marketing investment is hidden in opex but is clearly nine figures cumulatively.
Outcome: Public market cap fluctuates around $30–50B. Best-in-class category creation outcome — but on the back of >$1B in cumulative S&M.
The pattern: category creation is a 4–6 year, $50M+ marketing investment on top of product investment, and it only pays off when the category becomes a Gartner Magic Quadrant or Forrester Wave with you at the top. If you don't have the capital or the patience, don't start.
The 3-Question Test: Create or Enter?
Before you decide, run your product through these three questions. Two yeses or fewer = enter an existing category. Three yeses = category creation is on the table (but still optional).
Question 1: Is the buyer pain so new that no existing category covers it?
Decagon's pitch — "AI agents that fully resolve support tickets" — doesn't fit cleanly inside "help desk software." The pain (deflection of human-handled tickets via autonomous AI) didn't exist in 2022. Versus Cursor: developers wanted a code editor with better AI. "Code editor" is a real category with millions of buyers.
Question 2: Will the incumbent category leaders structurally fail to copy your AI advantage?
If Zendesk could ship the same AI agent that Decagon ships, Decagon doesn't need a new category — it just needs to win the existing one. The case for a new category is strongest when incumbents have channel conflict, architecture lock-in, or business-model conflict that prevents fast-following.
Question 3: Do you have the capital and patience for a 4-6 year category build?
Category creation requires sustained marketing investment in analyst relations, branded research, conferences, and content. Without $30M+ runway and a board that won't panic in year 2 when growth looks "niche," you will pivot mid-creation and end up with the worst of both worlds.
The data-driven version of this test: count how often your ICP types your category name into a search box. If the answer is "zero," you're creating. If it's "dozens of comparable terms," you're entering. Most founders convince themselves they're category creators when the search data says they're entering. Be honest. Our AI product positioning guide goes deep on how to test this without burning a year.
How to Name and Stick a Category
If you decide to create, the category name is product-defining. The best AI category names share three traits: they describe what the buyer is trying to do (not what the tech is), they're 2–3 words max, and they're defensibly new (not a near-synonym for an existing category).
Buyer-job framing
"Revenue Intelligence" (Gong) is what the buyer is doing — winning more deals. "AI for Sales" describes the tech. Buyer-job framing scales; tech framing doesn't.
2-3 words
"Conversational Marketing" (Drift). "Customer 360" (Salesforce). "Observability" (Datadog). Longer than 3 words and the category name dies in analyst reports.
Avoid 'AI-powered X'
"AI-powered CRM" isn't a category — it's a feature claim attached to an existing category. Categories that anchor on "AI-powered" date themselves to 2023–2024 and signal lack of conviction.
Own the URL and the analyst entry
Buy the .com. Lobby Gartner and Forrester to recognize the term. Sponsor research that uses your category name. The category doesn't exist until an analyst writes a report titled with it.
Repeat it everywhere, ruthlessly
Every email, every deck, every blog post, every podcast. If you, your sales team, and your customers can't say the category name in their sleep, it isn't a category yet.
Defensible novelty
"AI Workforce" vs "Digital Workforce" — close enough to old categories to invite copying. "Agentforce" (Salesforce) — distinct enough to own. Run the trademark search before you commit.
Pick the Right Category Strategy Before You Burn Capital
The AI PM Masterclass covers category strategy, positioning, and GTM sequencing in depth — taught live by a Salesforce Sr. Director PM who has shipped in both modes.
Mistakes That Kill Category Creators
Most attempted category creations don't fail because the category was wrong. They fail because of execution mistakes that compound. The four that kill the most teams:
Vague positioning that tries to cover too much
"The AI platform for everything" is not a category. Successful category creators pick a narrow ICP and a narrow first job-to-be-done. Drift was "chatbots for B2B marketing funnels," not "conversational AI for business." Narrow wins, broad dies.
Slow ICP commitment — chasing every revenue dollar
Category creators who say yes to every deal in year 1–2 dilute the category narrative. If Decagon had sold to mid-market SMB instead of enterprise from the start, "enterprise AI support agents" wouldn't be a recognized category today. Pick one ICP and refuse the others.
Premium pricing before the category exists
Category-king pricing ($100K+ ACVs) only works once analysts and buyers recognize the category. Pricing premium too early reduces deal velocity and slows the case-study flywheel that makes the category real. Price market-clearing in year 1; raise as recognition grows.
Letting incumbents define your category for you
If Salesforce decides "Agentforce" means "agent features inside CRM," your standalone agent startup is now positioning against a feature of a $300B incumbent. Get to analyst recognition before incumbents reframe your category as their feature. Speed matters.
The structural risk in 2026 specifically: foundation model providers (OpenAI, Anthropic, Google) and platform incumbents (Salesforce, Microsoft, ServiceNow) are aggressively reframing emerging AI categories as features of their existing platforms. A category creator that takes more than ~24 months to lock in analyst recognition risks being absorbed.
Case: How Cursor Entered (Didn't Create) and Won
Cursor is the canonical 2025 counter-example to category-creation doctrine. They entered the world's most mature software category — code editors, where VS Code already had >70% market share — and won faster than most category creators do. Their strategy is worth studying as the alternative path.
Step 1 — Forked, didn't reinvent
What happened: Cursor is a fork of VS Code. Day-one users got every keybinding, every extension, every muscle memory from their existing editor.
PM Implication: Zero switching cost = zero adoption friction. Category-entry strategy is to neutralize all reasons not to switch, then offer one compelling reason to switch.
Step 2 — One concentrated wedge: AI-native UX
What happened: Cursor didn't pitch "a new category of AI development environments." They pitched "the same editor you use, with better AI built in." The wedge was Tab autocomplete, then Composer.
PM Implication: Don't try to redefine the buyer's mental model. Slot into it and over-deliver on one dimension.
Step 3 — Bottom-up PLG distribution
What happened: Free tier, viral developer word-of-mouth, every engineering team self-adopting. No analyst push, no category-creating book, no conference brand.
PM Implication: When you enter an existing category, distribution is product-led. Save the analyst spend for when you're selling to a CIO.
Step 4 — Result: $500M+ ARR in 2025
What happened: Cursor reportedly crossed $500M ARR by late 2025 — faster than any category creator at comparable revenue scale. Did it without inventing a category name or buying a Gartner report.
PM Implication: Category entry can win bigger and faster than category creation when the AI capability gap vs incumbents is large enough that buyers self-select.
The lesson: in 2026, when AI capability gaps between products are huge, category entry can be more capital-efficient than category creation. The right move depends on whether your wedge is "a new way to do a known job" (enter) or "a job that wasn't possible before" (create). Pressure-test against the broader AI PMF signals framework and the AI GTM playbook.
The Decision Matrix
A practical scorecard you can run in an afternoon to pick between create and enter. Score each factor 1–5; sum the column.
Score CREATE high if…
Buyer can't name an existing category for the pain · Incumbents have structural reasons not to ship your solution · You have >$30M runway · Your team has analyst relations experience · The problem is enterprise (CIO-bought, analyst-influenced) · Your wedge requires a new buyer category (e.g., Head of AI).
Score ENTER high if…
Buyer already searches for a related category by name · Incumbents are slow but technically able to copy you · You're capital-constrained · Your team is product-led / developer-marketing strong · The buyer is bottom-up (developer, IC, end-user) · You have a clear 10x capability gap vs incumbents on a known job.
Hybrid: Enter, then Create
Snowflake entered "cloud data warehouse" (existing) and only later helped define "data cloud" (created). Many AI products should follow this pattern: win in an existing category first, then expand the category narrative as you scale. Lower risk, higher option value.
Red flag scenarios
Trying to create a category while pricing low and selling SMB · Trying to enter a category with a product that doesn't fit the buyer's mental model · Spending category-creation money without category-creation runway · Choosing based on what sounds impressive on Twitter rather than data.
The honest answer most AI startups should reach in 2026: enter first, create later. The capital efficiency of riding existing buyer demand beats the slow burn of category invention — unless your wedge truly requires a new buyer to exist.