Data
November 17, 2026

What to look for in a data team

What to look for in a data team

If you've already spent hours searching for data today, it's okay to admit it. You aren't alone: employees spend 10% of their work week searching for information they need to do their job.

So what should you actually be looking for in a data team, whether that's a hire, an agency, or the talent you already have in-house? It is not a list of tools, but rather something closer to a way of thinking. These should involve how they get curious about a problem before reaching for a method, whether they can explain a finding in plain language, and whether they tell you what they're sure about versus what's still a guess.

At Portage Labs, we are no stranger to the world of data. In this guide we'll walk you through the benefits of experimentation culture, show why data maturity is a big boon for business economics, and share how we approach (and how you can too) data problems with AI.

Why dashboards don't explain business trends

This is the complaint we hear most often, and it's worth addressing directly: the dashboard is live, the numbers are right there, so why can't anyone tell you why they're moving?

The honest answer is that a dashboard is built to display your data, not to decipher it. It can show a dip in a metric. But it can't tell you whether that dip is seasonal, a data quality issue, or an early signal of something real; that itself takes asking the right follow-up question. A data team worth its salt treats the dashboard as the start of that conversation, not the end of it, and builds systems that make dashboards functional for real-life problems (and business needs!).

1. Can they tell if two records are actually the same person?

Imagine this scenario. Your customer churns and rejoins under a slightly different email or using a different phone number. On your system that's noted as two clients. In reality, it's one, and every metric downstream (lifetime value, churn, retention) is quietly distorted by this duplication.

A data team that knows what they're doing will reach for entity resolution, using packages like Splink, weigh the evidence across fields rather than looking for an exact match, and produce a single, reliable view of each customer instead of fragmented ones.

If you want to find out the where, why, and how to reach the right customers with your data, this is usually where it starts. Keep in mind that it is not necessarily with a bigger dataset, but a cleaner one.

2. Do they assume one answer fits every market, or do they check?

A centralised strategy, run from one team, with one control group, is easy to manage and often wrong. For example, should a business treat European customers the same way as customers in a much smaller, lower-volume market? Without in-depth testing, it often can feel like a free-for-all, guessing at a regional strategy with no real evidence either way.

A serious data team reaches for a Bayesian hierarchical model here, borrowing signals from high-volume markets to make sense of low-volume ones, so a market with less data isn't left guessing and a market with a clear signal isn't drowned out by the global average. In practice, that can mean something as concrete as applying different data standards to French-speaking and English-speaking markets rather than forcing both through the same template.

Anyone in the market for data support would do well to look to teams who ask this question before recommending a rollout, not after.

3. Do they check their own experiment before believing it?

Here's a mistake that's more common than it should be: filtering an A/B test to only the users who actually engaged with the treatment, then comparing that group to the control. It might look like a clean result. But it's actually post-treatment bias, and it can make an ineffective program look like it's not.

The fix is a framework called intent-to-treat: look at everyone assigned to the treatment, whether they are engaged or not. It's a less flattering number most of the time, but it's the true one.

Equally impressive are the guardrails a serious data team builds around every test, not just the model. The discipline to check the test itself before trusting what it says is so important.

4. Can they turn a gut feeling into something provable?

Your sales team doesn't call every lead without gauging which prospect is a priority. A data team approaches the same task with a propensity score model, with historical patterns fitted to a statistical model that assigns every lead an explicit conversion likelihood.

In a solid data team, they are light on guesswork, and heavy on a ranked list.

5. Can you actually ask a dashboard a question and get a straight answer?

Your dashboard's insights should be everything a good data system is: usable, insightful, and shot through with the kind of detail that actually explains why a number moved.

For a few years, static dashboards have been giving way to something more conversational, like clients asking direct questions in plain language and getting an answer back. AI data insights have helped to broaden this feature (see our FAQ below).

Why experimentation culture matters more than most businesses assume

In the current data experimentation climate, it really matters. Take Kameleoon's 2025 Experimentation-led Growth Report, which found that 65% of businesses running experiments haven't fully integrated their testing technologies across teams and channels: disconnected tools and processes that keep experimentation limited to simple, page-level tests rather than something embedded across the business.

Experimentation occupies a strange middle ground for a lot of businesses: everyone agrees data matters, but planning around how to actually test and validate decisions is still slow and painfully drawn out. Having a data culture that not only experiments, but iterates on new techniques is key for businesses.

How Portage Labs approaches data problems

Portage Labs' standards mean clients get a more joined-up treatment of their data than a patchwork of one-off dashboards can offer. We start from the business question, so we are always clear on your priorities. We're upfront about any blockers. And we build on solid data foundations first, because unreliable data won't get you to your end goal.

That last point matters especially in Mauritius, where we're increasingly working with companies making the jump from basic reporting to something genuinely actionable. If your business hasn't laid that groundwork yet, that is the natural place to start, since none of the five capabilities above work well on top of data nobody trusts yet.

Whether you're evaluating your own data team against this list, or you'd rather bring in one that already works this way, that's the difference worth looking for.

Frequently asked questions


My dashboard is live and the numbers are right there, so why can't anyone tell me why they're moving?

Because a dashboard is built to display data, not interpret it. We treat this as an exploratory data analysis problem first, uncovering the patterns and anomalies underneath a number before anyone touches the dashboard again, so the explanation is grounded in what's actually happening in the data.


Can you actually ask a dashboard a question and get the correct answer, with AI?

Increasingly, yes. AI summary and query features are a native part of tools like Power BI and Tableau now. Portage Labs builds custom dashboards and automated reporting on top of a properly consolidated data source.


What is entity resolution, and why does it matter?

Entity resolution is the process of confirming whether two records (a customer, a lead, a transaction) actually represent the same real-world person or thing, even when the details don't match exactly. Without it, duplicate records distort nearly every downstream metric.


How do you avoid bias in A/B testing?

The most common mistake is filtering results to only engaged users after the fact, which introduces post-treatment bias. The fix is an intent-to-treat framework, measuring everyone assigned to the treatment group regardless of whether they are engaged, so the result reflects the real effect of a decision rather than a flattering version of it.


Do I need a data scientist on staff to do any of this?

No. Portage Labs works as your trusted external team, handling everything from consolidating scattered data to building the data lakes and configuring automated reporting. You can request help for any custom data projects.


What's the difference between a dashboard and real data insight?

A dashboard shows you what happened, whereas data insight explains why it happened and what to do next.