A law firm we spoke with last quarter had every partner paying for ChatGPT Plus. They were still hiring a paralegal to clean up the citations.
That gap — between what general AI does and what your work actually requires — is the whole story of horizontal versus vertical AI. The question isn't which one is better. It's which one your business needs, and when.
What horizontal AI is good at
Horizontal AI means the general-purpose tools: ChatGPT, Claude, Microsoft Copilot, Gemini. One model, trained on the internet, applied to anything you ask it.
These tools are excellent at a specific slice of work. Drafting first versions of emails. Summarizing long documents. Restructuring a messy spreadsheet. Brainstorming. Translating. Cleaning up rough notes into something presentable.
For knowledge workers, they save real time. They're the cheapest, fastest AI investment you can make — usually $20 to $30 per seat, per month.
Horizontal AI gets you to roughly 60% of the value AI can deliver in a typical small or mid-sized business. The first 60% is enormous. It's also where most companies stop, which is why most AI projects look like a productivity rounding error instead of a strategic shift.
Where horizontal AI falls short
The remaining 40% is everything that requires your industry's language, regulations, and workflows. That's where general models start producing confident answers that are subtly wrong.
A few patterns we've seen this year:
- A marine services client asked ChatGPT to interpret a survey clause. It gave a clean answer that confused a P&I claim with a hull claim. Different policies, different consequences.
- An accounting firm tried to use Copilot to categorize transactions against their chart of accounts. The model invented account codes that didn't exist.
- A construction GC ran site reports through Claude to extract delay risks. It missed the variation orders entirely because nobody had told the model what a variation order was.
None of these are bugs. The model is doing what it was trained to do: producing plausible language. Plausible isn't the bar in regulated, technical, or money-handling work. Right is the bar.
What vertical AI is
Vertical AI means tools built for one industry, trained on that industry's data, wired into that industry's workflows. They cost more. They do less. And inside their lane, they outperform horizontal AI by a large margin.
A few examples that read as familiar:
- Legal: Harvey and Thomson Reuters CoCounsel. Trained on case law, contract patterns, and legal reasoning. They cite real cases. They flag the clauses that matter.
- Healthcare: Abridge and Nuance DAX. Listen to a clinical conversation and produce a structured note in the doctor's voice, with the medical vocabulary and billing codes attached. Abridge is now mainstream inside major US health systems.
- Property and housing: EliseAI. An AI leasing agent that handles tenant communication, applications, and maintenance — and crossed $100M ARR in a few years, backed by a16z's Big Ideas 2026 view that vertical AI is hitting its inflection point.
- Accounting: Black Ore and Fieldguide for audit and assurance work. They understand the audit workflow as a series of permissions, sign-offs, and evidence trails — not as a generic document task.
The pattern is consistent. Vertical AI wins where the work has its own vocabulary, its own rules, and its own audit trail. Anywhere a wrong answer carries real cost.
A decision framework
Skip the abstract debate. Ask these questions about the specific work you're trying to improve:
1. Does the work have a private vocabulary? If your team uses words that only mean something inside your industry — variation order, P&I, SOAP note, GAAP adjustment — horizontal AI will guess and get it wrong some of the time. Vertical AI will know.
2. Does a wrong answer cost real money or risk? If the output goes to a regulator, a client invoice, a court filing, or a patient record, vertical AI's higher cost is a rounding error against the cost of being wrong. If the output is an internal email or a brainstorm, horizontal AI is fine.
3. Is the workflow already well-defined? Vertical AI tools assume your workflow looks like the industry-standard workflow. If you're a marine surveyor, that's probably true. If you run a uniquely structured operation, a vertical tool may force you into someone else's process.
4. How sensitive is the data? General models route prompts through shared infrastructure. Vertical tools in regulated industries are usually built with the compliance posture the industry expects — HIPAA, SOC 2, jurisdictional data residency. Worth checking, not assuming.
5. Is there a vertical AI for your industry that's actually mature? Some verticals — legal, healthcare, property, accounting — have serious vertical AI today. Others are still early. If the vertical option is one funded-but-unproven startup, it may be safer to combine horizontal AI with custom-built workflows on top.
If you answered yes to questions 1 through 4, you need vertical AI for that workflow. If you answered no across the board, horizontal AI is fine.
When you combine them
The realistic answer for most small and mid-sized businesses is both. Horizontal AI for the long tail of general knowledge work — every team member, every day, $20 a seat. Vertical AI for the two or three workflows where industry depth matters and a wrong answer is expensive.
That combination is also where Kursol does most of our work. The horizontal layer comes off the shelf. The vertical layer — for industries that don't have a mature AI product yet — gets built on top of general models with your data, your workflows, and a human in the loop. That's the augment-not-automate position in practice: keep the general-purpose tools, build the industry depth around your team rather than replacing it.
The position
Horizontal AI is a productivity floor. Vertical AI is where the strategic gains live. Buy the floor cheap. Invest seriously in the one or two verticals where your business actually competes. If a vendor is selling you horizontal AI dressed up as a vertical solution — a thin wrapper on ChatGPT with industry-themed marketing — you'll find out within the first month. Ask to see how they handle your specific edge cases before you sign.
FAQ
Some vertical AI products are thin wrappers — a generic model with a custom system prompt and industry branding. Real vertical AI is trained or fine-tuned on industry data, integrates with the systems of record in that industry, and has been validated against the work that industry actually does. The way to tell the difference is to ask the vendor about their training data, their integrations, and their accuracy benchmarks on real workflows.
Start with horizontal AI across the team. It's cheap, it lifts everyone's productivity, and it shows you which workflows generate the highest payback. After three to six months, you'll have a clear picture of where a vertical tool — or a custom-built vertical layer — is worth the investment.
Most industries don't have a mature vertical AI product. That's normal. The build path is to combine horizontal AI with your own data, workflows, and review steps — keeping a human in the loop on anything that leaves the building. That's what most of our client work at Kursol looks like.
Let's build your AI advantage
30-minute call. No sales pitch
Just an honest look at what autopilot could mean for your operations.