Playbook · Market thesis
Vertical AI in 2026: where the dollars actually are
Horizontal AI is a crowded knife fight against $100B incumbents. Vertical AI is a quiet land grab where buyer pain is concrete and incumbents don't show up. Here's the operator's read.
Every AI investor in 2024 was funding "horizontal" products: a better Slack, a better Notion, a better email client, a better CRM. The pitch was that AI changed the unit economics of the same generic productivity software, and a new winner would emerge in each category.
Two years in, that thesis hasn't paid off the way people expected. The horizontal categories that mattered most are still owned by the incumbents who already had distribution, brand, and the data flywheel: Microsoft owns the office, Salesforce owns the sales pipeline, HubSpot owns the SMB CRM, Notion owns docs, Linear owns engineering tickets. A few specialty horizontals broke through (Cursor, Granola, Perplexity), but they did so by being aggressively specific, not aggressively horizontal.
Meanwhile, in the verticals - legal intake, dental office ops, fire-department scheduling, restaurant inventory, immigration paperwork - the incumbents are old, the buyer pain is acute, and almost no one with real AI fluency is shipping there. That's the actual 2026 opportunity.
Why vertical AI works better than horizontal AI right now
The standard objection to vertical software is "TAM is too small." That objection comes from VCs and applies to VC-funded startups. If you're running on $5,000 to $250,000 of self-funded capital with the goal of $1-5M ARR, "small TAM" is a feature, not a bug. Small TAM means low competition, captive buyer attention, and the ability to charge for solving a specific painful problem instead of getting commoditized by a free general-purpose AI.
Look at that last row in particular. Defensibility in horizontal AI is almost impossible because the underlying model is the same and anyone can plug into it. In vertical AI, defensibility comes from knowing the workflow: which forms a paralegal actually fills out, which steps a roofing estimator takes between inspection and quote, which compliance reports a state DEP requires by date. That knowledge is hard to replicate and gets deeper with every customer.
Why 2026 specifically
The vertical AI thesis has been around since 2023. Why is 2026 different?
- Inference costs are 10x lower than 2023. A workflow that cost $0.15 to run on GPT-4-turbo costs about $0.02 on Claude Haiku 4.5 with prompt caching. Margins finally work.
- Vertical buyers got their first taste of AI in 2024-2025 from ChatGPT, found it useful but generic, and now want something that works inside their actual software stack. The discovery happened. The buying readiness is here.
- Incumbent vertical SaaS hasn't shipped meaningful AI yet. Some have bolted on a chatbot. None have re-thought their core workflows. There's a 12-24 month window before the slow incumbents catch up.
- Capital markets are out of horizontal-AI patience. Vertical AI startups can raise on smaller numbers because the metrics are real: paying customers, real ARR, low CAC, high retention.
Verticals worth attacking right now
Not every vertical is open. Some have already been won (vertical sales tools - clay.com, gong, attio). Some are too regulated for an operator-funded play (clinical decision support, FDA-cleared software). Here are the verticals where in 2026 the math actually works:
Legal (specifically: intake, immigration, small-firm operations)
Big-law has Harvey and a few well-funded competitors. The interesting opportunity is the long tail: solo practices, small firms (under 10 lawyers), immigration practices, plaintiffs work, real estate closings. These firms get drowned in intake work, document review, and client communication. Their current software (Clio, MyCase, Smokeball) does not have native AI workflows yet. Buyer pain is daily. Pricing power is real ($300-1500/mo per firm).
One product in our catalog (Counsel AI, now intakecounsel.com) graduated by doing exactly this: after-hours intake for solo and small firms.
Trades and field service (dispatch, estimating, scheduling)
Plumbers, electricians, roofers, HVAC, landscaping. The buyer is a shop owner with 5-30 trucks who's running on ServiceTitan, Jobber, or a spreadsheet. The pain: dispatch is chaotic, estimates take too long, lead routing leaks revenue. AI fits naturally into all three.
Distribution channel is not LinkedIn. It's trade associations, regional trade shows, and ServiceTitan's marketplace. Conferences cost $800-2000 but are how this market actually buys. The buyer is older than the typical SaaS buyer (median age 50+) and prefers a phone call, but they pay $500-3000/mo without flinching once value is shown.
Real estate (specifically: tenant communication, listing copy, agent backoffice)
Residential real estate is owned by Zillow and Realtor.com. Commercial is owned by CoStar. The opportunity is in the operator middle: rental property managers (50-500 units), commercial brokers at boutique firms, agent teams that need backoffice automation. Buyer pain is communication overhead and listing creation. AI fits cleanly into both.
Education operations (analytics, not chatbots)
The "AI tutor" space is crowded and hard to monetize (parents won't pay, schools can't, ed-tech is a 5-year sales cycle). The interesting vertical play is in school operations: principal-level analytics, learning data interpretation, curriculum auditing, IEP report generation. The buyer is a district administrator or independent school operations lead. The pain is real (reporting requirements are crushing). The check size is $5-50K annually per district.
Compliance and audit (specifically: small-firm SOX, regional banks, healthcare audit prep)
Compliance software is dominated by legacy vendors (AuditBoard, MetricStream, Resolver). The middle of the market (small audit firms, regional banks, mid-size healthcare orgs) is paying $50-200K/year for software that mostly stores documents and reminds people. AI-native workflows for evidence collection, control testing, and report drafting compress that work by 3-5x. The pain is annual and acute. Pricing power is strong.
Hospitality operations (multi-location restaurants, hotel groups, salons)
Toast and Lightspeed own POS. The opportunity is at the operator-of-multiple-locations level: scheduling, inventory predictions, customer follow-up, review responses, cross-location reporting. Buyer is a regional ops manager. They pay $200-800/mo per location.
The honest case against vertical AI
Not everyone should attack a vertical. Skip this approach if:
- You don't have or can't acquire vertical expertise. If you've never worked in or with the industry, you'll spend 6 months learning vocabulary before you can sell a single seat. Either pick a vertical you naturally belong to, or hire/partner with someone who does.
- You need $50M+ ARR to satisfy investors. Most vertical AI plays cap out at $5-20M ARR without geographic expansion or product expansion. That's a great cash business, a poor venture business.
- You need fast feedback loops to learn what to build. Vertical buyers move slowly. Procurement cycles are 30-90 days. Pivoting based on real signal takes months, not weeks.
- The vertical is regulated in ways you don't understand. HIPAA, FINRA, FERPA, GDPR, and state-specific rules can sink a product. If the dossier doesn't address regulatory risk explicitly, treat the vertical as higher-risk than it looks.
And the boring observation: vertical AI is harder to write headlines about. "We're building AI for fire-department scheduling" doesn't trend on Twitter. If your motivation is being part of a hot scene, attack horizontal. If your motivation is paying buyers and durable margins, attack vertical.
What good looks like in vertical AI 2026
- Specific named ICP. Not "small businesses." A specific role at a specific size of organization with a specific pain. "Operations manager at a 12-truck plumbing shop in a Tier 2 city" beats "SMB owner."
- One workflow, deeply done. Don't ship 10 features. Ship 1 workflow that compresses a 90-minute task to 5 minutes, and own that workflow for that ICP.
- Distribution that doesn't involve paid social. Trade associations, industry publications, conference circuits, embedded partnerships with existing vertical SaaS. Direct outbound at 100 named accounts.
- Pricing anchored to value, not seats. If your product saves a paralegal 10 hours/week at $40/hour internal rate, that's $1,600/month of saved labor. Charge $400-800/mo and you're a no-brainer.
- Native integration with the dominant incumbent tool. Pull data from the vertical's current system of record (Clio, ServiceTitan, AppFolio, etc). Don't ask the buyer to switch core software.
- Compliance-aware from day one. Whatever regulatory or audit requirements exist in the vertical, your product should be in compliance or have a clear path. This is where many AI products will lose deals.
How to use this thesis
If you're a builder, this is the time to pick a vertical you understand, find the specific workflow that's most painful in that vertical, and build the AI-native replacement for it. The catalog has a curated list of 10 vertical-specific ideas with named ICPs, suggested integrations, and pricing models.
If you're not sure which vertical to attack, the answer is "the one where you have an unfair advantage." The dossier on each catalog product names the specific buyer and the workflow. Read 3-5 of them in verticals you have some exposure to. The one where you immediately think "oh, I know how to find that buyer" is your answer.
If you'd rather hand the dossier and have us run the launch, the operator partnership tier is designed for exactly this. Many of the most successful vertical AI launches in 2026 will have a domain expert and an operator running them, not a solo founder.