Skip to content
Tallyfy

Run analytics with Claude

Your first analytics question, start to finish

You can get real analytics out of Tallyfy just by asking Claude. Connect the Tallyfy MCP server to Claude, ask a question in plain English, and Claude reads your live workflow data and answers. Often with a chart you can drop straight into a document.

This is the fast, conversational way to explore your data. No SQL, no dashboard to build. For the bigger question of when to ask Claude versus when to build a standing dashboard, see one-time questions vs recurring dashboards.

Step 1: ask a hello-world question

Once Claude is connected to your Tallyfy account, start small. Something you can eyeball to confirm it’s reading the right data:

Connect to my Tallyfy account. How many templates do we have, what's
the average number of steps per template, and can you show me a bar
chart of templates by step count?

Claude calls the Tallyfy MCP server, pulls your templates, counts them, works out the average, and draws the chart. You might see something like:

You have 14 templates, averaging 9 steps each. Here’s the spread. Your heaviest is “Field Service” at 29 steps. Three templates have 3 steps or fewer, which sometimes means they’re really checklists rather than full processes.

That’s it. If the count matches what you expect, the connection is reading your data correctly and you can trust the harder questions.

Step 2: ask the questions that actually matter

Now ask the things you’d normally wait on a report for. Each of these is a one-time question. Ask, read, act, move on.

  • Bottlenecks. “Across our active processes, which step sits unfinished the longest on average?”
  • Throughput. “How many processes did we complete last month compared with the month before?”
  • Cycle time. “What’s the average time from launch to completion for our Quotation process, and has it changed since we simplified it?”
  • Adoption. “Which templates were launched in the last 30 days, and which haven’t been used at all?”

Claude works through these in steps. It reads your data, counts, and usually double-checks itself before answering. When a question is vague (whose tasks? which date range?), it asks you to confirm before it touches anything.

Step 3: reach for full-dataset SQL when you need it

The Tallyfy MCP server answers through Tallyfy’s API, which is plenty for most questions. When you need to query your entire history with custom SQL (joins across every task, form field, and run over several years), that’s the job for Tallyfy Analytics. It copies your workflow data into a private Amazon Athena environment.

You can point Claude at that data too. If you have the Analytics add-on, you already have the Athena credentials. Community MCP servers can run read-only SQL against Athena, so Claude writes the query, runs it, and explains the result in plain words. Search the MCP directories for “Athena”, and see AWS Labs’ official collection of AWS MCP servers1 for where this is heading.

Keep any database connection read-only, and handle the credentials with the same care as any other system login.

Tips for answers you can trust

  • Ask Claude to show its work. “List the templates you counted” turns a number into something you can verify.
  • Save prompts you’ll reuse. If you ask the same question every month, word it the same way each time so the answer stays comparable.
  • For numbers people depend on, lean on the governed dataset. Exploring is Claude’s strength. For figures that drive money or targets, Tallyfy Analytics gives you one definition computed the same way every time.

Footnotes

  1. A growing set of official AWS connectors for AI assistants, including read-only query access to data services. [1] ↩

Get in touch
About Tallyfy