Choosing a BI Stack for a South African Business: Metabase vs Superset vs Grafana vs Looker Studio (and Power BI)

The question behind the question

Every few months someone asks me which dashboard tool they should use. The honest answer is that the tool is rarely the hard part. The hard part is the data underneath it: whether it’s clean, modelled, and landing somewhere queryable. But the tool still matters, because it decides who can see the data, what it costs you every month, and how much engineering time you burn keeping it alive.

I’ve shipped dashboards on most of this list. I built a real-time Power BI sentiment dashboard for the insurer Hollard (Hollard case study), and the fleet-operations dashboards behind Decarb.Earth’s solar data platform (Decarb.Earth case study). This post is the comparison I wish clients had read before our first call, written for a South African business or technical lead, where the rand-to-dollar exchange rate makes per-seat SaaS pricing sting more than it does in London or San Francisco.

The five contenders, honestly

Metabase is the tool I recommend most often to small and mid-sized teams. It’s open source, self-hostable on a single small server, and non-technical people can actually use it. The question builder lets an operations manager filter and group data without writing SQL, while analysts drop into raw SQL when they need to. It’s opinionated and a little limited at the high end: that’s a feature, not a bug, for most businesses.

Apache Superset is the heavyweight open-source option. More chart types, more control, proper SQL IDE, row-level security, and a permission model granular enough for a real data team. The cost is operational: it’s a multi-service Python deployment (web workers, async workers, a metadata database, a cache), and non-technical users find it noticeably less friendly than Metabase. Superset shines when you have a data team who will own it, and struggles when you don’t.

Grafana is superb at what it was built for: time-series metrics, operations, infrastructure, IoT. If you’re watching sensor readings, server health, or solar inverter telemetry ticking in every few minutes, Grafana with alerting is hard to beat. It is the wrong tool for business BI, and I’ll come back to why.

Looker Studio (Google’s free tool, formerly Data Studio) is the zero-budget option. If your data already lives in Google Sheets, GA4, or BigQuery, you can have a shareable dashboard this afternoon for nothing. It gets slow and awkward as data volumes and modelling complexity grow, and its permission model is basically Google Drive sharing: fine for marketing reports, thin for anything sensitive.

Power BI is Microsoft’s answer, and in South Africa’s corporate landscape it’s often the default. Per-seat licensing, deep Excel and Microsoft 365 integration, and a genuinely powerful modelling layer (DAX) that rewards investment and punishes casual use.

Side-by-side

The ZAR cost reality: servers vs seats

This is where South African context actually changes the answer. I won’t quote prices (they change, and exchange rates change faster) but the structures are stable and they’re what you should reason about.

Per-seat SaaS pricing is set in dollars. Every viewer licence, every month, forever, at whatever the rand is doing that year. Ten people who need to look at a dashboard becomes a recurring USD line item that scales linearly with headcount and non-linearly with your CFO’s blood pressure. Power BI mitigates this inside Microsoft shops because seats often ride along with existing 365 agreements. But if you’re not already in that ecosystem, you’re buying in.

Self-hosting inverts the structure. Metabase runs comfortably on a small VM: the kind of instance that costs about the same whether five people or fifty people log in. Your costs are a fixed, modest hosting bill plus the occasional hour of someone’s time to apply updates and check backups. For a 20-person company where half the staff should see numbers, that maths usually lands decisively on the self-hosted side.

The honest counterweight: self-hosting is only cheap if someone competent owns it. Superset in particular can quietly consume days of engineering time. A “free” tool that eats a week of a senior engineer’s month is not free. Price the person as well as the server.

Embedding dashboards in your product

If you’re a software business and the dashboard is for your customers (inside your SaaS product), the calculus changes entirely.

Metabase’s paid tier offers signed, parameterised embedding that works well for multi-tenant products: each customer sees only their rows, and you control the frame. Superset can do it, but expect to build and maintain more glue yourself. Power BI Embedded exists and is capable, but it’s priced on dedicated capacity, in dollars, and that’s a serious commitment for a small SA product company. Looker Studio embedding is essentially a public-ish iframe: I wouldn’t put customer-scoped data behind it.

There’s also a fifth option worth naming: for a product where the dashboard is the product, skip embedded BI entirely and build the views in your frontend against your warehouse. That’s what I did for Decarb.Earth’s public impact map and internal fleet dashboard. When you need full control over UX and performance, a BI tool in an iframe is a ceiling, not a floor.

When Power BI is the right answer

I’m not reflexively anti-Microsoft: the Hollard dashboard I built ran on Power BI and it was the right call. Power BI makes sense when:

• Your organisation already lives in Microsoft 365, so licensing is incremental and users sign in with their existing Microsoft account.
• Your power users are Excel people. The mental model transfers, and adoption is real rather than aspirational.
• You need enterprise governance (Entra ID groups, sensitivity labels, audited sharing) and someone to blame with an SLA.
• Reports need to circulate through Teams and SharePoint, because that’s where your organisation actually reads things.

Where it doesn’t make sense: a startup on Google Workspace with a Postgres database, no Microsoft footprint, and rand-denominated revenue. There, per-seat USD licensing plus the DAX learning curve buys you very little over a Metabase container.

When Grafana is the wrong tool

Grafana keeps getting pulled into business BI because someone on the team already loves it, and it’s genuinely free and pleasant to run. Resist this.

Grafana’s model is metrics-first: time on the x-axis, values streaming in, alert when a threshold breaks. Business BI is different work: joining a sales table to a customer table, grouping revenue by region and month, letting a non-technical manager change the question. Grafana’s table and transformation features can be bent into that shape, but every step fights you: no friendly question builder, weak ad-hoc exploration for non-engineers, and a permissions model organised around dashboards and folders rather than data.

Use Grafana for what it’s for. Watching machine telemetry, uptime, queue depths, energy production ticking in real time: excellent. Answering “which product line grew fastest in the Western Cape last quarter?”: wrong tool, and your ops engineer becomes a bottleneck for every business question.

Governance, permissions, and POPIA

For South African businesses, governance isn’t optional paperwork: POPIA makes you accountable for who can access personal information and why. Your BI tool is often the widest door into that data, because it’s the one tool the whole company logs into.

Concretely, that means three questions for any tool you’re evaluating:

1. Can you restrict at the row level? A branch manager should see their branch. Superset and Power BI do this well; Metabase does it via sandboxing in paid tiers; Looker Studio effectively can’t.
2. Where does the data live? Self-hosted tools let you keep data in-country on local cloud regions. With SaaS tools, check the hosting region and your processing agreements.
3. Can you audit access? If the Information Regulator ever asks who viewed a dataset of customer identity numbers, “we shared a link” is not an answer you want to give.

A plain decision framework

Work down this list and stop at the first line that describes you:

1. Data in Google Sheets/GA4, no budget, marketing-style reporting only → Looker Studio. Ship today, revisit in a year.
2. Microsoft 365 shop, Excel-heavy analysts, enterprise governance needs → Power BI. You’re already paying for most of it.
3. Monitoring machines, sensors, or infrastructure, not business questions → Grafana. And keep it out of the boardroom.
4. Small-to-mid team, mixed technical ability, cost-sensitive, own database → Metabase, self-hosted on a small VM. This is the default answer for most SA SMEs I work with.
5. Dedicated data team, complex permissions, heavy analytical workloads → Superset, but only if an engineer genuinely owns it.
6. Dashboards are a feature of your product → Metabase embedding for speed, or custom frontend views on your warehouse for control.

Whatever you pick, spend the saved money on the layer underneath: a modelled, tested warehouse. A mediocre tool on good data beats a great tool on a swamp, every time.

If you’re weighing this decision for your own business, I can help you make it concrete: run your requirements through the scoping tool to get an indicative shape and cost for the work, or just get in touch and we’ll talk through your stack.