Evidence before opinion: how a data product earns trust

A commercial intelligence product can turn daily changes across public regulatory datasets into ranked opportunity briefs. People may use those briefs to decide who to call and when. If a brief is wrong, someone can waste a week.

So the product has to earn trust on every page. We set one rule early, and it shaped the whole design.

The rule

Observed facts stay separate from interpretation.

A permit date, an ownership change or a removal record is an observed fact, and each one should link back to its source record. A commercial score, a “why now” statement or a decision to engage is an inference, and the product should label it as inference.

The reader can always see which is which.

What the rule changed

Every factor is listed. A campaign score uses remaining scope, permit activity, deadlines, lease status, production history and agreement between datasets. The brief shows each input. There is no black box.

Confidence is explicit. Every ranked campaign carries a confidence level. When datasets disagree, the confidence drops, and the brief says so.

Decisions are plain. Each campaign ends in one of three words: watch, investigate or engage now. A team can act on that without reading a model description.

Limits are stated. An intelligence tool is not the authoritative regulatory record. A published cost estimate is not a contract value. The product says this in the brief, not in a footnote.

Provenance should be part of the interface

Data provenance is often treated as backend documentation. In a decision product, it belongs in the user experience. A reader should be able to move from a conclusion to the factor behind it, then from that factor to the original record. The path should take seconds, not a support request.

That does not mean filling the screen with citations. Use progressive disclosure. Show the conclusion first, list the important factors beside it, and make the underlying records available when the reader wants to check them. A brief can stay concise while remaining defensible.

Provenance also needs time. Display when a source was last checked, when a record changed and whether the product has processed the latest release. “Current” is not a useful claim unless the reader knows what current means.

Why it matters commercially

A brief gets forwarded. The person who receives it did not buy the product and has no reason to trust it. When the evidence is one click away, the brief can defend itself inside the customer’s team. That is how a $19 brief turns into a subscription.

The same principle applies beyond commercial intelligence. A risk score may reach a manager who never opened the application. A policy summary may appear in a board pack. A recommendation may be copied into a ticket. Once the conclusion leaves the original interface, branding and onboarding cannot carry its credibility. The evidence has to travel with it.

Design exports accordingly. Include the observation date, the source names, the factors that affected the result and a permanent link back to the live record. A screenshot without context is fragile. A self-explaining brief can survive forwarding.

Handle disagreement instead of hiding it

Public and internal datasets often disagree. One table may show an asset as active while another records a removal. The tempting response is to choose the source that looks newest and suppress the conflict. That creates false certainty.

A better product keeps the disagreement visible. Define a source hierarchy, lower the confidence level and tell the reader what needs checking. In some cases, the conflict itself is the useful signal. It may reveal a delayed filing, a changing plan or a record that deserves attention.

Confidence should never be a decorative percentage. Explain what raises it and what lowers it. Readers do not need the full statistical model, but they do need enough information to judge whether the recommendation fits the decision they are making.

What we learned

Narrow scope lets you go deep. One basin and one data source made it possible to link wells, structures, pipelines, leases and operators into real campaigns, not just a list of changes.

Trust is a design feature. Source links, confidence levels and stated limits take screen space. They are also the reason a buyer forwards the brief.

Let the buyer test on one target. A free search and a one-time brief let a team prove value on one real pursuit before paying monthly.

Make corrections visible. Data products will be wrong sometimes. Keep a correction history, update derived conclusions and make it easy for a knowledgeable reader to challenge a record. A transparent correction can strengthen trust; a silent change weakens it.

Review the decision language. “Watch”, “investigate” and “engage” are clearer than an unexplained score because they connect the model to an action. The words should match the user’s authority and the cost of being wrong.

If you are building a product that scores, ranks or recommends, decide early how the user will check your work. The products that get adopted are the ones that make checking easy.