Semilattice
Get started
NEW Announcing Custom

User in the loop

Semilattice predicts what your users want in minutes, so you can ship fast without shipping blind. Fix problems while they’re cheap, not after launch.

Get started See how it works
Usage alerts PRD v1

Usage alerts PRD v1

Maya Simons · Updated 12m ago

Premium subscribers regularly exceed usage limits without warning, driving support tickets and surprise bills. This PRD proposes usage alerts to give subscribers visibility and control over their spend.

Requirements

  1. 1. Real-time alerts on usage thresholds.
  2. 2. Configurable per workspace; weekly digest default.
  3. 3. Alert history export in every plan.
  4. 4. Alerts locked to the Enterprise tier.
  5. 5. Usage data retained for 12 months.

Success metrics

  1. 1. Upgrade rate to Custom (primary).
  2. 2. Alert opt-in within the first week.

Open questions

  1. 1. Weekly digest or real-time default?
Premium Subscribers
Simulacra
752
Alignment
0.775
Engine
anacreon
Simulate
83% would upgrade 1 in 3 would churn
MONEYME logo
“Using Semilattice has been game changing when developing product features. We’ve been able to source early, frequent, and instant customer feedback that’s almost identical to real life customer survey responses. It’s been instrumental in helping us make the right product decisions at every stage of ideation and discovery.”
Jem Corden · Head of Product, MONEYME
Read the case study
Use cases

Ask users at every stage

At every layer of detail and every stage of the process, Semilattice lets you check with users. Explore assumptions, validate concepts, prioritise features, and optimise flows.

01 · Scoping

Explore assumptions and concepts

Drop in a PRD and Semilattice tests every requirement against simulated users. You see which work well, which confuse, and what’s missing, all before a single ticket is written.

Guest checkout · PRD
Guest checkout
Draft 2 Maya Okafor·Product
Requirements

1. Offer guest checkout: no account required before payment.

2.Collect a single-screen delivery address, autofilled from postcode.

3. Returns policy stays as a link in the footer.

4.Saved cards are optional, opt-in after the first purchase.

Success metrics
Checkout completion rate (primary) · guest vs. account split.
72% of UK adults would use it 4 in 5 UK adults miss this
02 · Development

Validate decisions and UX details

Put the design in front of simulated users and run them through the prototype. See what they understand, what they miss, and where they go, all while changes still cost minutes, not sprints.

Checkout flow Share
Layers
Flow
Cart
Details
Pay
Components
1 · Cart
2 · Details
3 · Pay
22% of users drop off here “Pay now” lifts completion
03 · Testing

Test how users will interact

Drop a staging URL and watch simulated users explore the real flow. You get predicted completion, drop-off by step, and the fixes to make before launch.

staging.folk.shop/checkout
folk.
Checkout
alex@example.com
Delivery address
Address line 1
Address line 2
City
Postcode
UK shoppers stall at the address
Free 30-day returns UK adults find this reassuring
Card detailsVISA
How it works

Describe the change. See the consequences.

Tell Semilattice what you’re planning and it predicts the outcome while you’re still working. User research mode shows what your users think, user journey mode what they’ll do.

01

Describe the change

A decision, a PRD, a design, a staging URL. Drop it in, or just say what you’re thinking of doing. The planner clarifies and plans simulations to run.

Projects / Checkout redesign ＋ New project
checkout-prd.md
Guest checkout
Planner
M
⎘ checkout-prd.md
The guest path is the risky change.
Read it. Are we validating the whole flow, or just the guest path?
M
Just the guest path.
Planning three simulations against the guest flow.
Planning 3 simulations
checkout-prd.md
02

Semilattice runs simulations

Review and edit the plan, then hit run. User research and user journey simulations run in parallel, against models of your users.

Projects / Checkout redesign ＋ New project
Simulations · checkout-prd.md
Running
Guest-checkout appetite
UK Adults 2026 · 12 questions
Discovery User research
Checkout walkthrough
staging.folk.shop/checkout · cart to confirmation
Conversion User journey
Single-screen guest concept
UK Adults 25–34 · 9 questions
Concept validation User research
Planner
M
⎘ checkout-prd.md
The guest path is the risky change.
Read it. Are we validating the whole flow, or just the guest path?
M
Just the guest path.
Planning three simulations against the guest flow.
⎘ Checkout PRD
3 simulations · running
03

Get validated recommendations

Results are synthesised into specific changes to make. The data behind them, charts and user flows, is one click away.

Projects / Checkout redesign ＋ New project
Synthesis Updated just now
Export ▾ Share
Strong appetite for guest checkout across both markets, with the account wall the biggest source of drop-off. Ship it, with three fixes.
Recommendations 3
1Surface guest mode: the account requirement buries it.
2Autofill the address from postcode to cut drop-off.
3Move the returns reassurance above the fold.
Key predictions
Would you use guest checkout?
User research
UK Adults 2026 US Adults 2026
Definitely
Probably
Unsure
No
Planner
M
⎘ checkout-prd.md
The guest path is the risky change.
Read it. Are we validating the whole flow, or just the guest path?
M
Just the guest path.
Planning three simulations against the guest flow.
⎘ Checkout PRD
3 simulations · done
MCP

Add users to your agentic loops

Connect the Semilattice MCP to Claude, Claude Code, Cursor, or any agent you work with. Your agent plans and runs the simulations, interprets the results, and can apply the improvements to your work.

Claude Opus 4.8 ▾
We're moving to usage-based pricing. How will free, pro, and enterprise customers react?
Run simulation·Free tier
Run simulation·Pro tier
Run simulation·Enterprise
Reactions split sharply by tier, so a single rollout will not land.
  • •Free users barely notice; lead with the lower entry price.
  • •Pro users fear runaway bills; ship a visible spend cap.
  • •Enterprise wants predictability; offer an annual commit.
Reply to Claude…
Claude
Connect →
✳ Claude Code - ~/app
> simulate the new onboarding
⏺ Run simulation
  ⎿ 200 simulated users
⏺ Fetch results
  ✓ completion 64% → 78%
The new flow lifts completion, but the phone step is where it leaks.
• Move phone verification to the end; most drop-off is there.
• Keep the single-field email start; it tested well.
• Add a skip on the workspace step for solo users.
> ▍
Claude Code
Connect →
Cursor
Plan · annual billing toggle
Thought 4sRead pricing.tsx List simulations Fetch results
Checked what we already know: two past runs cover this.
  • •71% prefer the monthly default; keep annual opt-in.
  • •Show the saving inline; it doubled toggle intent.
  • •No new run needed; wiring the toggle now.
Add follow-up…
Plan Composer 2.5
pricing.tsx
1export function Pricing() {
2 const [annual, setAnnual] = useState(false)
3 const price = annual
4 ? plan.yearly
5 : plan.monthly
6 return (
7 <BillingToggle
8 annual={annual}
9 onChange={setAnnual}
10 saving={plan.saving}
11 />
12 <PriceLabel value={price} />
13 {annual && <SaveBadge amount={plan.saving} />}
14 <PlanFeatures tier={plan.tier}
15 compact
16 />
17 )
18}
Cursor
Connect →
User models

Validated against real people, question by question

Our user models are built from real human data and validated on held-out questions. We benchmark against published academic baselines for digital twins, and show every test result in the product.

answers-1 Free

Built on real human survey data and validated question by question. Every result is an audience-level distribution: what your market thinks, in proportions you can act on.

UK Adults 2026 US Adults 2026 Global Developers UK Gen Z B2B Software Buyers +44 more
49
public models
88%*
average accuracy
Free
get started now
*Average Mean Absolute Error (MAE) across held-back test predictions.
anacreon Custom Technical report (PDF)

Trained on unstructured data: hundreds of individual simulacra, each one a model of a single person. You get individual-level answers, so you can segment and re-cut the results any way you need.

Ahead of the best published academic baselines.*
Top-1 accuracy
Held-out individual-level questions
anacreon
67.0%
Park et al.
65.7%
Ordinal alignment (1-MAD/range)
Held-out individual-level questions
anacreon
0.77
Mega-Study
0.75
Socrates
0.74
Twin-2K-500
0.71
*Benchmarked on held-out questions against published academic baselines: Park et al. (2024), arXiv:2411.10109 · Toubia et al. (2025), arXiv:2505.17479 · Peng et al. (2025), arXiv:2509.19088 · Kolluri et al. (2025), EMNLP 2025
Get started

Free

Get started free

Create an account and run your first simulation now.

Pricing
Free
49 free public answers-1 models
User research and user journey modes
The app, the MCP, and the API
Unlimited simulations
Create account

Custom

Get a custom model of your users

Semilattice Custom is a partnership, not a licence. We build a model of your users and meet regularly to help you simulate your decisions.

Pricing from
$1,799 / month
Everything in Free, plus:
A custom anacreon model of your users
Organisation features with unlimited seats
Regular working sessions and support

Book a call
As seen in
Audio