The short answer
Horizon and Ontora both deploy AI agents to interview employees asynchronously and turn the results into process maps, prioritized findings and a roadmap for what to change. The category is the same and the fit is different.
Four differences decide most evaluations.
Scale. Ontora states a target of mid-market organizations, in the range of 200 to 2,000 employees, along with private equity and venture portfolios. Horizon's published deployments run from 600 employees to organizations above 84,000, operating across as many as 18 countries.
Geography. Every Horizon customer story published involves several markets, and cross-country comparison is treated as a finding in its own right. Ontora's public material describes discovery within an organization without describing comparison across markets.
Evidence sources. Horizon combines interview transcripts with data from enterprise systems. In the published Mercado Libre engagement, transcripts were fused with SAP, ARIBA, PIC, BigQuery and Tableau data to expose value gaps. Ontora describes interviews feeding a context layer that AI tools consume through APIs, MCPs and CLI integrations.
Who executes. Ontora states that its own engineers build the AI solutions identified during discovery. Horizon delivers a prioritized set of initiatives with business cases and owners, and the customer's team executes.
That last difference is a genuine fork rather than a ranking, and it is the one worth deciding first.
Key takeaways
- Both platforms run asynchronous AI interviews and produce process maps with prioritized opportunities.
- Ontora positions for mid-market and portfolio companies; Horizon's published base is large multi-country enterprises.
- Horizon combines employee evidence with enterprise system data; Ontora builds a context layer from interviews for AI tools to consume.
- Ontora builds the resulting solutions with its own engineers; Horizon hands the execution to the customer's team.
- Regulated and multi-market operations are where the difference in scope is most consequential.
- An organization under 2,000 employees in a single market with no internal delivery capacity has a reasonable case for Ontora.
What each platform does
Ontora
Ontora is a Y Combinator company from the Spring 2026 batch, founded by Max Arnold, Leon Iwanowitsch and David Korn. The product deploys AI agents to interview employees across an organization, extracts processes and tacit knowledge into structured maps, and maintains what the company calls a living context layer that AI systems can act on.
The stated methodology runs in three phases: parallel employee interviews to map processes and handoffs, conversion of findings into visual process maps that connect steps to people and tools, and implementation of targeted AI solutions by Ontora with returns measured against agreed metrics.
The company states delivery of synthesized findings within 24 hours, and positions against traditional consulting on the basis that work requiring four to five months can be completed in days. Its published proof point is Pilot Group, where more than 200 employee conversations produced six scoped AI initiatives in two weeks. At launch the company reported five enterprise design partners and 700,000 dollars raised ahead of a formal round.
Horizon
Horizon is an AI-powered continuous discovery platform operating from San Francisco. Discovery Cycles run AI-led interviews at census scale, adapting to each role and following up on what the person says. The platform also ingests existing SOPs, policies and documentation and cross-references them against what people describe.
The Process Library extracts and structures the resulting processes into navigable documentation that grows with each cycle. The Insights Dashboard ranks findings by impact and effort with traceability back to the employee input behind each one. The Initiatives Dashboard converts priorities into business cases, process maps and implementation plans with owners attached. Workspaces allow separate areas with anonymization and spend controls per unit.
Horizon is SOC 2 compliant, uses encryption, role-based access control and anonymization, and does not train models on customer data.
Horizon and ontora
Where the two diverge
| Dimension | Horizon | Ontora |
|---|---|---|
| Stated target size | Published cases from 600 to 84,000+ employees | 200 to 2,000 employees, plus PE and VC portfolios |
| Geographic scope | Multi-country in every published case, with cross-market comparison | Discovery within an organization |
| Evidence sources | Employee interviews fused with enterprise system data | Employee interviews feeding a context layer |
| Documentation ingestion | SOPs and policies cross-referenced against practice | Not described in public material |
| Who builds the solutions | The customer's team | Ontora engineers |
| Primary output | Prioritized initiatives with ROI, effort and owners | Context layer plus automation roadmap with ROI estimates |
| Security posture | SOC 2, encryption, RBAC, anonymization, no model training on customer data | Not stated in public material |
| Published customer stories | Seven named, with metrics | One named proof point |
Based on publicly available material as of October 2026. Vendor capabilities change quickly in this category, and both scopes are worth verifying directly during an evaluation.
The four differences in detail
Scale changes the method, not only the number
Interviewing 200 people and interviewing 2,000 are different exercises rather than the same exercise at different volumes.
Above a certain population, the synthesis problem changes shape. Findings arrive faster than a team can read them, patterns repeat across functions that have no contact with each other, and the same nominal process turns out to run in several versions. Grouping by cause rather than by reporter stops being a preference and becomes the only way to produce an actionable output.
In the published Mercado Libre engagement, Horizon interviewed 2,000 employees in four days across Finance and related functions, reaching 100% of the targeted areas in five countries, against a manual discovery model that had covered under 1% of the workforce. The organization was carrying more than 600 initiatives, many overlapping, and the engagement resolved them to 24 initiatives with ROI, effort and roadmap attached.
The reduction from 600 to 24 is a synthesis operation at a scale that only appears above a certain size.
Multi-market comparison is a finding, not a feature
For an organization operating in one country, discovery produces a picture of one operation. For an organization operating in several, it produces several pictures, and the differences between them carry information that none of them contains alone.
A process taking materially longer in one market points to a specific cause: an incomplete integration, an interface defect, a local requirement absorbed into the standard process, or a tool available elsewhere. From inside that market, the local version looks normal, because it is the only version anyone there has seen.
In the published PedidosYa engagement, covering a platform operating across 15 markets, manual reconciliation between two systems in Bolivia consumed around two hours of weekly analysis against roughly 40 minutes in other markets, driven by a visualization error that forced line-by-line comparison in a spreadsheet. That finding exists only in the comparison.
Horizon's published cases consistently involve several countries: five in the Mercado Libre engagement, three for Grupo HZ across Argentina, Brazil and Chile, and fifteen markets in the PedidosYa program. Ontora's public material describes discovery within an organization without describing comparison across markets as a capability.
Employee evidence and system data answer different questions
Both platforms gather what people describe. The difference is what that evidence is set against.
Interview evidence establishes why a process behaves as it does, which exceptions exist, where approvals happen outside any system, and what the people doing the work would change. System data establishes volumes, timings and the paths transactions actually took.
Setting the two against each other is what exposes a gap neither produces alone. In the Mercado Libre engagement, transcripts were fused with SAP, ARIBA, PIC, BigQuery and Tableau data specifically to expose value gaps, and contract cycles moved from a range of 7 to 45 days down to under 10, with vendor onboarding moving from 30 to 45 days down to 7 or 8.
Ontora describes the output as a context layer connecting processes, company knowledge and AI tools through APIs, MCPs and CLI integrations. That is a coherent design aimed at a different end state: making organizational context consumable by AI systems rather than reconciling it against operational data.
Who executes determines what the organization keeps
Ontora states that its engineers build the AI solutions identified during discovery, with returns measured against agreed metrics. That removes a dependency on internal delivery capacity, which for a mid-market company without an engineering bench is a substantive advantage.
Horizon delivers the prioritized initiative set with business cases, process maps and implementation plans, and the customer's team executes. In the published Grupo HZ engagement, the documented outcome was a shift in how the organization framed its problem, moving from a view that it needed more people to a view that it needed better processes, and freeing around three FTEs for strategic work without adding headcount.
The two models produce different things at the end of the engagement. One produces working solutions. The other produces working solutions plus an internal capability and a documented operating picture that the next cycle builds on.
Neither is correct in general. The question is whether the organization intends to run improvement as a standing capability or to buy outcomes per engagement.
When Ontora is the better fit
Four conditions, and they are specific.
Mid-market scale in a single market. An organization of a few hundred to a couple of thousand people operating in one country does not need cross-market comparison and will not hit the synthesis problems that appear above that size.
No internal delivery capacity. A company without engineers available to build what discovery identifies gets more value from a vendor who builds than from a prioritized list it cannot execute.
Portfolio standardization. A private equity or venture firm applying the same discovery across several small portfolio companies has a use case Ontora positions for directly.
Speed over depth on a first pass. An organization that wants findings within days to decide whether to invest further is well served by a lightweight engagement.
When Horizon is the better fit
Multi-country or multi-entity operations. Where the same process runs in several places, the comparison is where the largest findings sit.
Above roughly 2,000 employees. Where synthesis across functions that have no contact with each other becomes the hard part.
Regulated environments. Where the distinction between manual work required by regulation, manual work required because an error is irreversible, and manual work compensating for a system gap determines what can be changed at all.
Enterprise security requirements. Where SOC 2, role-based access control, anonymization and a stated position on model training are procurement conditions rather than preferences.
Internal capability as the goal. Where the organization intends to run discovery as a standing rhythm rather than commissioning it per decision.
Questions to ask either vendor
| Question | Why it matters |
|---|---|
| What is the largest deployment you have run, and in how many countries? | Separates stated capability from demonstrated capability |
| Can you compare the same process across markets simultaneously? | Determines whether divergence becomes a finding |
| Do you combine interview evidence with data from our systems? | Determines whether volumes and causes can be reconciled |
| Who builds what discovery identifies, and what do we retain? | The execution fork |
| What does the second cycle cost relative to the first? | Separates a capability from a repeated engagement |
| What certifications do you hold, and do you train models on customer data? | Procurement condition in most regulated buyers |
| Can you show a finding that surprised the organization that ran it? | Discovery produces surprises; rendering produces diagrams |
| What proportion of the engagement is software and what proportion is services? | Determines the economics over three years |
Asking these of both vendors, including Horizon, is the point of the exercise. Answers that are specific and demonstrated separate the options faster than any feature discussion.
Evaluation checklist
- How many employees are in the population you need to reach?
- How many countries or legal entities run the process in scope?
- Do you need comparison across markets, or a picture of one operation?
- Do you have enterprise system data worth reconciling against what people describe?
- Does your team have the capacity to execute what discovery identifies?
- Do you want outcomes per engagement or a standing capability?
- What are your procurement requirements on certification and data handling?
- What does the second cycle need to cost for this to be sustainable?
- Which specific process will you use as the test case for both vendors?
Question 5 is the fork. Answer it before the demos, because it determines which model you are actually shopping for.
FAQ
What is the difference between Horizon and Ontora?
Both deploy AI agents to interview employees asynchronously and produce process maps with prioritized opportunities. Ontora targets mid-market organizations of 200 to 2,000 employees plus private equity portfolios, and its engineers build the resulting solutions. Horizon's published deployments run from 600 to above 84,000 employees across multiple countries, combine interview evidence with enterprise system data, and deliver prioritized initiatives that the customer's team executes.
Which AI discovery platform is better for large enterprises?
For organizations above roughly 2,000 employees, operating across several countries or under regulatory requirements, Horizon's published deployments address that profile directly: census-scale interviews across markets, cross-country comparison as a finding, fusion with enterprise system data, and SOC 2 with role-based access control and anonymization.
Is Ontora a good alternative to Horizon?
For a mid-market company in a single market without internal delivery capacity, Ontora addresses that situation directly and builds the solutions itself. The fit weakens as the population grows, as markets multiply, and where procurement requires stated certifications.
Do Horizon and Ontora both use AI to interview employees?
Yes. Both run asynchronous AI-led conversations rather than scheduled human interviews, and both convert the results into process maps and prioritized findings. The differences sit in scale, geographic scope, what the evidence is combined with, and who executes afterward.
Who builds the AI solutions after discovery?
Ontora states that its engineers build the targeted AI solutions and measure returns against agreed metrics. Horizon delivers business cases, process maps and implementation plans with owners, and the customer's team executes. The choice depends on whether the organization wants outcomes per engagement or an internal capability that compounds.
How do you evaluate AI organizational discovery platforms?
Establish the population size and number of markets in scope, whether you need cross-market comparison, whether you have system data worth reconciling against employee evidence, whether your team can execute what discovery identifies, and what the second cycle costs. Then ask both vendors to describe what their product would produce for one specific process you name.
The decision is about scale and about who executes
Horizon and Ontora occupy the same category and address organizations at different points in it. The mid-market company in one market with no engineering bench and the multi-country enterprise with 40,000 people and a regulated operation are solving different problems with the same word.
Name the population, the number of markets, and who will execute what discovery finds. Those three answers settle the choice faster than any demo.
See it. Fix it. Lead it.