Two-name trial checklist for a new alternative data source

Neon Blue builds simulated audiences of a company's customers, AI models that answer the way a defined group of people would, grounded in Neon Blue's own paid studies and AI-moderated interviews with real people. For funds testing a new survey-based alternative data source, this checklist is a practical two-name trial: pick names you already know well, ask for a fixed deliverable, and decide whether the feed belongs in your process.

Why two names you already know?

One name can look good by luck. Two names show whether the vendor can repeat the same method on different category structures. Pick:

1. A name where your customer thesis recently moved, such as a price increase, a launch, churn or trade-down.
2. A quieter coverage name where you still have a clean view of recent shopper behavior from calls, card data or your own work.

You are checking whether the vendor's process produces decision-useful tables on both, not whether it "called" the last print.

What should you ask for in the trial?

Work this list with the vendor before any data is delivered.

1. Written audience definition. Who is in the sample (recent buyers, paid users, category buyers), geography, lookback window and exclusion rules.
2. Exact question list. Including any price, offer or product text shown to respondents, and which steps are test levels you chose.
3. Segment cuts you use in the model. Heavy vs light, tenure, income band, plan tier or channel. Agree these up front.
4. Output layout. A segment table plus coded reasons with sample verbatims, delivered in a form you can paste into a model note.
5. Dating. Field window, delivery date and whether the read is a one-off or the start of a repeat series.
6. Method note. How simulated or surveyed respondents are built, how live interviews are used if at all, and what the vendor says the read is not (no stock calls, no price targets).
7. Your own prior. For each name, write down one assumption you already hold, so the readout can show where the table agrees or differs and why.

Neon Blue's product shape for this work is a simulated audience study with optional live interview follow-up. How simulated audiences work is the method page, and what is synthetic data for investment research covers the category.

What does a two-name trial look like in practice?

Hypothetical: you cover a casual-dining burger chain where a value platform drove traffic and is about to lap, and a frozen-food maker that just raised prices. You hold one prior on each.

Name A, the restaurant. Prior: "Most value-platform visits would happen anyway." The trial question is how many visits depend on the offer.

• Audience. US adults who ate at a casual-dining burger restaurant in the past 3 months, split by offer buyers, other guests and non-guests, cut by lunch vs dinner.
• Questions. Was the offer the main reason for your last visit? If it were gone for a month, how would your visits change? Which part of the offer matters most? What would bring you in without it, in your own words?
• What comes back. Visit dependency by daypart and frequency, ranked value drivers, coded reasons. A full version of this study, written from a real earnings call, is on our burger chain value-platform study page.

Name B, the staples maker. Prior: "Frozen volume falls about as much as the company planned." The trial question is units bought at each price step.

• Audience. US adults who bought frozen single-serve meals in the past month, split by value-brand, premium-brand and store-brand buyers, cut by income.
• Questions. At each test price step, how many would you buy on a typical trip? At the higher price, what would you buy instead? What would make the price feel worth it, in your own words?
• What comes back. Units-at-price curves by brand tier and income, the share moving to store brand, coded reasons. Compare the shape with our frozen-meal price-ladder study.

Both names use the same skeleton: intent against a recent behavior, a substitution question, an open-ended why, and the same income and frequency cuts. If the vendor can't run that skeleton on both, the trial has already told you something.

How do you run the trial step by step?

For between-print framing, keep consumer channel checks open. For how a table becomes a model line, see can a simulated consumer survey feed an investment model.

Example trial scorecard (illustrative, not a vendor rating)

The layout a fund can use to score any vendor's two-name trial. Cells are blank on purpose. Nothing here rates Neon Blue or any other vendor.

How to read it: if both names clear the process bar, keep the vendor on a repeat cadence for those inputs. If only one does, fix the template before you add coverage. Map the restaurant read to foot traffic and the staples read to trade-down behavior so the trial lands on lines your team already models. For a second stickiness read, the is trade-down sticky checklist uses the same skeleton.

What should a trial not be used for?

• Calling the next print or the stock. A consumer intent table is an input, not a recommendation.
• A one-name beauty contest. One flattering table is not a process.
• Unbounded question sprawl. If the vendor can't reuse the same template on name two, the trial won't scale across coverage.
• Declaring a vendor the winner on stock outcomes. Keep the decision on method clarity and model fit.

Questions about alt-data vendor trials

How long should a two-name trial take?

Usually one agreed field window plus a few days for delivery and readout. A trial that drags usually means the question list was never locked.

Do both names need to be in the same sector?

No. A restaurant and a staples name work well together if the skeleton (intent, substitution, an open-ended why, the same income and frequency cuts) stays the same.

What if the vendor's table disagrees with my prior?

That can still be useful. Ask whether the audience definition or the question text drove the difference before you discard either view.

What does Neon Blue deliver in a two-name trial?

Neon Blue delivers one segment table per name in the cuts you set, coded reasons with sample verbatims, and a method note on the audience and questions. Live AI interviews can follow up on any segment.

Does Neon Blue make stock calls in a trial?

No. Neon Blue returns consumer answers by segment as labeled inputs. It does not make stock calls or give price targets.

How do I start a study?

Bring two names, one prior per name, and the segment cuts you always want. Start a study.

More from consumer research for funds

• Consumer research for funds
• How funds evaluate a new alternative data source
• Consumer channel checks
• How simulated audiences work