Product2025-07-23T16:35:12+0000

The Semilattice API: user insights as infrastructure

Predict user behaviour like you make database queries.

Joseph WrightCo-Founder, Product & Engineering

UI mock showing a B2B Agent chat interface, with the agent running a simulation with Semilattice plus the code showing how that simulation would be generated

There’s a particular moment that happens in every product meeting that probably costs businesses billions annually. It goes like this:

Product Manager: “Should we prioritise this feature or that feature?”
Executive: “What does the data say?”
Data Scientist: “We don’t have any reliable proxy metrics for these features.”
Executive: “What does our customer research say?”
Product Manager: “We don’t have research on these features.”
Executive: “Well, let’s commission a study.”
Research Team: “That’ll be $35,000 and take 6 weeks.”
Executive: “Never mind, ship this feature.”

This is not hyperbole. This is Tuesday. And this is the exact kind of problem Semilattice solves.

The artisanal nature of traditional research

If you work in product, marketing, or data, you’ve probably internalised a particular cost equation without realising it. Market and user research is expensive, slow, and finite. You can have answers to three important questions per quarter, maybe five if you really push the budget. Unless you have the right data (which you don’t), everything else gets decided by intuition, internal politics, or whoever speaks loudest in the meeting.

The maths on traditional research is unforgiving. A basic survey with 1,000 respondents costs $5,000-$15,000. Add demographic targeting, and you’re at $25,000. Want to understand a niche population? $40,000. Need results in less than a month? Add 50%. Want to test multiple concepts? Multiply everything.

The result is that most product decisions—the ones that collectively determine whether your company succeeds or fails—are made without any customer input whatsoever. This is systematically expensive. A single wrong assumption about user preferences can tank a product launch. Misunderstood customer segments lead to features nobody uses. Poor messaging choices kill conversion rates for months. The aggregate cost of decisions made without customer input is invisible but massive.

User insights as infrastructure

Semilattice changes this arithmetic entirely. Through months of research, we discovered a method of prompting an LLM which reliably outputs accurate predictions about specific populations. Our approach takes ground truth data about an audience and algorithmically reassembles it into thousands of prompts which together mine accurate answers to new questions about that audience. Instead of paying $5,000 to survey 1,000 people over 4 weeks, you pay $1 to query an AI model of your target population and get results in 45 seconds.

When user behaviour data has the cost structure of a database query instead of a consulting engagement, it prompts a rethink. A big rethink.

We believe customer insights should be infrastructure, not services. When insights are expensive and slow, they're reserved for major decisions—feature launches, campaign strategies, product pivots. But when they're cheap and instant, they become something else entirely: an always-on feedback system that can inform every decision, from high-level strategy down to individual user interactions.

This transformation enables two fundamentally new integration patterns, each with its own multiplier effects:

Process integration replaces expensive quarterly research with real-time validation in human workflows. Data scientists query user preferences before making targeting recommendations. Product managers validate feature concepts during roadmap planning. Marketing teams test messaging variations before campaign launches. The insights feed into human decision-making, turning every product conversation from speculation into data-driven analysis.

Programmatic integration embeds behavioural predictions directly into automated systems. Recommendation engines query user preferences before serving suggestions. AI writing tools test generated copy against target audiences. Onboarding systems choose optimal flows based on user characteristics. The insights feed directly into code, eliminating hardcoded assumptions about user behaviour and enabling systems that adapt to real human preferences.

These integration patterns compound in different ways. Process integrations save money and improve decision quality—measurable wins that justify themselves quickly—while programmatic integrations create entirely new product capabilities that were previously impossible at scale. The former means making better decisions about users, while the latter means building products that understand users. But they’re not mutually exclusive. Both patterns point toward the same future: software that doesn't just collect user data, but actively predicts and responds to user behaviour.

Getting started with the Semilattice API

The Semilattice API starts with a simple foundation: two primitives which can be combined to support a range of use cases. As we learn more about how teams integrate behavioural predictions into their workflows, we'll expand the API with more sophisticated tools.

Population Models represent specific user profiles—from general consumers to software developers to niche segments you define with your own data. Think of them as behavioral fingerprints for different types of users.

Answers simulate how those populations would respond to questions. You can apply this concept across any use case where you need to predict user behaviour: copy testing, feature prioritisation, pricing sensitivity, onboarding optimisation, or content personalisation.

1from semilattice import Semilattice
2
3semilattice = Semilattice(api_key=API_KEY)
4
5answer = semilattice.answers.simulate(
6    population_id=POPULATION_MODEL_ID,
7    answers={
8        "question": "Which is most important to you?",
9        "answer_options": ["Integrations", "Latency", "Cost"],
10        "question_options": {"question_type": "single-choice"},
11    }
12)

1from semilattice import Semilattice
2
3semilattice = Semilattice(api_key=API_KEY)
4
5answer = semilattice.answers.simulate(
6    population_id=POPULATION_MODEL_ID,
7    answers={
8        "question": "Which is most important to you?",
9        "answer_options": ["Integrations", "Latency", "Cost"],
10        "question_options": {"question_type": "single-choice"},
11    }
12)

These primitives adapt to remarkably diverse scenarios. For example, marketing teams can use Population Models for different customer segments and Answers to test messaging variations ("How would enterprise buyers respond to this value proposition?") Or product teams can model user personas and simulate feature preferences ("Would power users prefer advanced customisation or simplified workflows?")

We provide a REST API as well as Python and TypeScript SDKs. Sign up for an API key at semilattice.ai and visit the docs to get started.

Get in touch

If you're working on problems where user behaviour matters—whether that's optimising conversion funnels, personalising experiences, or building AI agents that understand humans—we want to learn from you. Please send us a note or schedule a call to talk to us.

Product

Company