Anthropic released a new artefact referred to as Claude Dashboards, alongside a short introductory video that showcases how the feature works and how teams can collaborate using it.
It is a great new feature that has unfortunately been plagued by a lot of LinkedIn hot takes. So, I thought it would be useful to give a slightly more balanced take on what the Claude Dashboards capability is and what it is not.
Here, I will not be comparing individual features of Claude Dashboards with existing Business Intelligence (BI) tools. Rather, I will discuss the feature in the context of the BI ecosystem as a whole.
What is a Claude Dashboard?
The new Dashboard artefact is a Claude-generated artefact that is able to connect to various platforms, pull live data and metrics, and generate a dashboard. All it takes is a few clicks and a prompt.
On top of the general dashboard functionality, it also has some useful features, such as:
- Comments and collaborative work within a Project;
- Exporting the dashboard to a connected tool: for example, we can export a Claude Dashboard into Omni, and whilst it is not like-for-like, Claude and Omni’s MCP will take care of the conversion;
- Investigating where numbers have come from, including the SQL used to generate them.


Common Claude Dashboards misconceptions
By itself, Claude Dashboards is a really useful product. But before going into the use cases, it is important to address some common misconceptions around it.
This is the death of BI
The common misconception that I have seen across LinkedIn and X recently is that Claude Dashboards will somehow result in the “death of BI".
The notion that the release of a BI tool will cause the death of BI is paradoxical but let’s nevertheless entertain this one.
BI is generally defined as the process of helping an organisation make better business decisions through the use of data and technology. Unfortunately over a number of years, it seems to have morphed into just being about dashboard builds.
Of course, dashboards can be incredibly useful for displaying information at a glance and providing the ability to drill into certain areas. However, in reality, not everything needs to be a dashboard.
Here are just a few examples where there is a much better-suited solution than a dashboard:
- To be able to course-correct when things seem to be going wrong, set up alerting and anomaly detection instead.
- If you are looking for a quick answer on your data whilst in the middle of analysis, use conversational analytics.
- If you are giving regular updates every Monday morning to set the tone for the week, schedule a newsletter with all of the numbers and data embedded into it.
- When you are building bespoke financial calculations that do not lend easily to dashboards, set up a spreadsheet with a connection to the base data.
The ability to choose the right tools for the job is part of what an effective BI strategy entails.
At Infinite Lambda, we often advise clients in this space to move away from the tunnel vision that every question needs a dashboard. Instead, we help them focus on what the use case actually is and how it embeds into existing workflows.
A final point here is that almost all major BI platforms already offer functionalities to address the use cases above. You can, of course, vibe code your own versions of them, but that might be a can of worms you do not really want to open.
Claude Dashboards will surely improve efficiency
The next misconception, and one that applies to any other AI use, is around speed for speed’s sake.
BI requests take too long, so throwing AI at it must definitely improve efficiency, no?
Many people claim that Claude Dashboards will lead to a significant reduction in bottlenecks around the speed of delivery.
Indeed, we often hear from clients that the process of raising a request, that request landing with the BI team, and ultimately coming back in the form of a delivered data product can stretch to several weeks.
However, when we investigate what causes the delay, we tend to find it has nothing to do with the dashboard build itself.
Often, it is blocked by the numerous activities that surround it, such as:
- Stakeholder interviews
- Requirements gathering
- Defining metric logic
- Wireframing & iterations
- Working with data engineers if the data is missing in the first place
- Data model or semantic layer updates
- Testing and validation
The list could go on. Whether using Claude Dashboards or a BI platform, the activities will not go away.
If we take an example four-week timeline, the dashboard build would likely only take 1–2 days, the rest of the time being spent on the list above. So, the dashboard build is not the time-consuming part in the overall process.
Finally, almost all modern BI tools make it relatively simple to create dashboards with point-and-click functionalities. AI can definitely help speed up the bulk of the build, but I believe it is still quicker to change a bar chart to a line chart using your mouse than it is to re-prompt and wait for the model to produce an output.
More importantly, the majority of BI solutions already offer AI-based approaches to dashboard building.
Omni, for example, has two distinct approaches:
- Omni’s AI, Blobby, which can help you build dashboards quickly
- The newly released Omni Apps for an AI-based HTML or JavaScript approach
So, you should not expect any miracles from Claude Dashboard in terms of speeding up the delivery.
You do not need a semantic layer
The concept of a semantic layer has been discussed fairly extensively this year, but in case you have not had a chance to catch up, here is an Omni blog post on what makes a semantic layer good.
Why does this matter when it comes to Claude Dashboards, though?
Under the hood, Claude connects to your dataset and processes your prompt to figure out what it should build. If you do not have a well-defined semantic layer as part of that connection, Claude will essentially just take its best guess at what it thinks you need.
Here, scale is key. For relatively minor use cases, you could get away with it by re-prompting it until it provides the output you expect. But try to scale this approach to an entire organisation and the risk increases significantly because dashboards will start to disagree with each other.
The rate at which AI labs are implementing new features is frighteningly high and makes it hard to forecast what the future of the BI market is going to look like. But right now it is very clear that the BI tools that are going to succeed are the ones with a robust semantic layer that coherently integrates into the rest of the tech stack.
When to use Claude Dashboards
Have I been trying to say Claude Dashboards are pointless? Absolutely not.
In fact, I have already used them for a couple of use cases.
Non-semantic-layer use cases
A handy use case I have found whilst testing Claude Dashboards is the ability to quickly spin up a dashboard to inform me about how my data quality is looking at different stages of the medallion architecture.
There are two main prerequisites:
- The dashboard does not really need to touch any of the semantic information since it is primarily looking for nulls / referential integrity / custom tests I am defining.
- I do not necessarily need historical data because I am just doing some basic investigative work.
Here is what my dashboard looks like:

This was incredibly easy to set up, and during my first iteration, I even let Claude figure out the checks by itself for testing purposes.
Of course, it took a few iterations as the intentional vagueness resulted in needing to modify some of the more nuanced checks, but it allowed me to be up and running within about 15 minutes.
Prototyping in Claude
This is another use case that I find particularly useful. A lot of my workflows now live within Claude, and being able to work there until the last second has a lot of advantages when it comes to consistency.
Most of my meeting notes are processed in Claude to generate a to-do list and a requirements list.
Within the same session, I can quickly transition to building out a dashboard based on the above information, especially if there is no new data modelling work needed. Finally, once I am happy with the prototype, I can export the same dashboard to Omni for productionising.
Final deliverables still end up in Omni because, amongst other things, the production-grade infrastructure is much better suited for the hosting and sharing of dashboards across an enterprise.
However, the journey there can vary depending on the individual’s workflows and needs. In addition, the outputs are all being driven by a central semantic layer (Omni’s Shared Model here), so the numbers between both should be in sync.
The verdict on Claude Dashboards
In this article, I have been looking to share a balanced view on the new Claude Dashboard functionality.
Whilst there is no doubt that Claude Dashboards will help in certain areas, they do not address the underlying challenges in BI, especially the ones around process, semantic layer, and scalability.
If you need a hand adopting advanced analytics in your organisation, explore Infinite Lambda’s data analytics and BI modernisation solutions.
And if you would like to learn from peers' experience with AI-powered analytics, join us at the upcoming Omni AI Analytics Meetup in London.