From dashboards to real-time, personalised answers
Unifying Hunter.io, Apollo, HubSpot and your own databases — for people and AI agents — with AgentData and MCP.
BiSkilled has spent the better part of two decades building data warehouses and data lakes for banks, telecoms and government — and shipped a lot of dashboards along the way. Here's the uncomfortable truth that experience leads to: the dashboard era is ending.
Not because dashboards are useless — but because they answer questions you decided to ask months ago, in a layout someone designed for an average user who doesn't exist. The real questions show up in the moment, they're specific to one person's job, and they cut across systems that were never meant to talk to each other. A static dashboard can't keep up. This is the use case AgentData was built for: ad-hoc, personalised, real-time answers across all of your data — relational databases, warehouses and lakes alike.
Why the dashboard model breaks
Every data team knows this cycle. Someone requests a dashboard. It takes weeks. By the time it ships, the question has changed. So they request another one. You end up with 200 dashboards, nobody trusts half of them, and the answer to "which enterprise accounts are at risk this week?" still isn't on any of them — because that question crosses your CRM, your billing database and your support tool at once.
The three structural problems:
- Pre-baked, not ad-hoc. A dashboard answers a fixed question. Real decisions need a follow-up, a filter, a "now show me just the ones in the UK" — instantly.
- One-size-fits-none. An account manager, a founder and a finance lead need different cuts of the same data. A single dashboard serves none of them well.
- Siloed. Your most valuable answers live between systems — enrichment tools, your CRM, your product database — and no dashboard spans them safely.
The shift: a semantic layer your people and your AI can both query
The fix isn't another BI tool. It's a semantic layer — one shared model of your business entities (customers, accounts, orders, leads) that sits over all your sources, so anyone (or any AI agent) can ask a question in plain language and get a governed, correct answer in real time.
That's what AgentData is. It connects to your sources read-only, auto-discovers the business entities inside them, and serves both structured and natural-language queries over a REST API and an MCP (Model Context Protocol) endpoint. Crucially, only the model and the question ever reach the LLM — never your row data. The SQL is generated from the model and runs locally against your sources; only the result comes back.
The unlock most people miss: because the same governed query path is exposed over MCP, your AI assistants become first-class data consumers. Claude, ChatGPT or your own agent can answer "which of these leads already exist in HubSpot?" without anyone writing SQL and without handing raw tables to a model. Here's what that looks like in practice.
Scenario 1 — Go-to-market: one question across Hunter.io, Apollo, HubSpot and your product DB
A salesperson is looking at a target company. Today that means four tabs and a spreadsheet. With AgentData connected to Hunter.io, Apollo, HubSpot and the product database, they ask one question — in chat, or through an AI advisor — and the agent works across all of it:
- Find the right contacts at the domain (Hunter.io).
- Enrich them with role, seniority and company firmographics (Apollo).
- Check who already exists in the pipeline, the last touch, and open deals (HubSpot).
- Cross-reference whether anyone at that company is already a product user or on a trial (your database).
- Advise — the LLM advisor reads all of that through MCP and suggests the next best action: "two contacts are net-new, one is an existing trial that went cold 3 weeks ago — reconnect with the champion first."
No exports, no copy-paste, no stale CSV. The data never leaves your environment — the agent only ever sees the semantic model and the answers it's allowed to get. That's the difference between "an AI that guesses" and "an AI that knows."
Scenario 2 — The personalised report that replaces a dashboard
A customer-success lead doesn't want "the churn dashboard." She wants the answer to her question, right now: "Show me enterprise accounts renewing in the next 60 days that have an open invoice and a support ticket raised this month." That spans billing, CRM and support. With AgentData she just asks it — in plain language, in any language — and gets a live, personalised result:
- It's generated on demand, not pre-built — and she can immediately follow up: "just the ones above £50k," "now group by owner."
- It's governed — the same validated model every team uses, so the numbers reconcile.
- It's hers — and the next person gets their own version of "right now," without a ticket to the data team.
How MCP ties it all together
MCP is the connective tissue. AgentData exposes a small set of tools over a secure MCP endpoint — list the available entities, describe one, run a metric, ask a natural-language question — gated by a per-user API key. Any MCP-capable client (Claude, ChatGPT, an IDE, or a custom agent) connects with one line and can then reason across every connected source through a single, validated path:
- One contract, many sources. Hunter.io, Apollo, HubSpot and your warehouses look like one queryable model to the agent.
- Safe by construction. Read-only adapters, encrypted credentials, and only metadata plus the question reaching the LLM — so you can do this even with regulated data, on-prem or air-gapped.
- Composable. The agent chains steps (find → enrich → check → advise) because each source answers through the same interface.
Why this is the future, not a gimmick
Static dashboards assume you can predict the questions. AI assistants in every workflow mean you can't — and you no longer need to. When your data is exposed as one governed semantic layer over MCP, the interface to your business becomes a conversation: any person or agent, any source, any language, in real time, with the governance intact. Dashboards become a fallback for the handful of metrics that genuinely never change.
This is where the work gets interesting, and it's why BiSkilled built AgentData on top of two decades of warehouse and lake experience: the hard part was never the charts — it was making every source, from relational databases to lakes, trustworthy and queryable at once. Get that right, and personalised real-time answers stop being a project and start being the default.
Want to see AgentData on your own stack?
Book a demo and we'll connect a sample of your sources — or explore the live app and MCP endpoint.