Ask better questions

When founders first use thola, the most common feeling after a week is "the answers are OK but not great." Almost always, the answers are great — the questions are vague. This recipe is the small set of patterns that turn vague into sharp.

The scene

You typed "how are we doing?" and got a four-line summary. It's accurate but doesn't tell you what to do next. You want the answer to feel like talking to a smart ops consultant, not reading a status report.

Three patterns that work

Pattern 1 — Ask broad, then narrow

The fastest way to get to a useful answer is to start wide, then narrow based on what comes back.

The Planner remembers the context across turns — you don't have to keep naming the deal.

Pattern 2 — State the outcome, not the steps

Tell the agent what you want done, not how to do it.

✅ "Send a payment reminder to all customers more than 30 days overdue." ❌ "Open Finance, filter invoices by status overdue, sort by age, then send each one an email…"

The agents know how to use their tools. You don't have to walk them through click-paths.

Pattern 3 — Use names, not pronouns when you change topic

Pronouns work inside a single thread of thought. They drift when you change topic.

✅ "What's the status of the Globex deal?" → after looking → "What about the GlobalTex one?" ❌ "What's the status of that one?" (which one?)

Names always work; pronouns sometimes do. When in doubt, name the thing.

Two patterns that don't work

Anti-pattern 1 — Asking for opinion, not analysis

✅ "What's our biggest risk this month, with evidence?" ❌ "Should we hire?"

"Should we hire" is a values question, not a data question. The agent will hedge, ask clarifying questions, eventually punt back to you. Save your tokens and your time. Reframe: "Given our current runway and pipeline, would a ₹15L/year hire bring runway below 6 months?" — that the agent can answer.

Anti-pattern 2 — Open-ended creative requests

✅ "Draft a re-engagement email for the Acme deal, in the customer's tone, mentioning the May timeline we discussed." ❌ "Write me a marketing strategy."

"Write me a marketing strategy" is a 50-page document the agent has no business writing. Reframe into a specific, scoped task. The agent shines at one task at a time, not strategy documents.

Useful question templates

Status checks

"How are we doing in [module]?" "What's the [metric] this [period] vs last [period]?" "Show me [entities] with [filter]."

Diagnoses

"Why is [metric] down?" "What's contributing to the drop in [metric]?" "Show me the [top N] [things] by [criterion]."

Forecasts

"Forecast [metric] for [period]." "What's the probability of hitting [goal] by [date]?" "If [scenario], what happens to [metric]?"

Actions

"Draft a [message type] for [recipient] about [topic]." "Run the [Playbook name] Playbook." "Update [entity] [field] to [value]." "Reassign [entities] from [owner] to [new owner]."

The "show me your work" follow-up

When the agent gives you a number you don't immediately trust, ask:

"Where did this number come from?"

The agent will show its query — which data, which time window, which calculation. If something looks off, you can fix it (often it's a date-format issue or a category mis-tag).

The "make it shorter" / "expand on that" follow-ups

The agent's first reply is usually mid-length. You can shape it:

• "Shorter." — gets you the headline.
• "Expand on point 2." — dig deeper on one specific thing.
• "Show me as a table." — for data that wants tabular form.
• "Just give me the three things I should do this week." — when you want action, not analysis.

These follow-ups are cheap (a fraction of a token). Use them generously.

What to do when the agent is wrong

It happens. When it does:

1. Say it's wrong, and why. "That's wrong — INV-104 was paid on 12 May, see the bank statement." The agent will recheck, often finding the issue (a mis-categorised bank row, a duplicate invoice).
2. If the data is wrong, fix the data. The agent doesn't make up numbers. It uses what's in your workspace. If a number is wrong, the source data is wrong; trace it back.
3. If the agent's interpretation is wrong, say so explicitly. "That's not 'stuck' — we're waiting on a client signature, that's expected." The agent will remember and not flag it next time.

A note on tokens

Every reply costs a fraction of a token. You can see your usage in Profile → Usage. Most founders run on 50–500 tokens/month, well within the Starter plan.

Be a little generous with follow-up questions, frugal with very long analyses. "Draft a comprehensive marketing plan" costs ~10 tokens; "Show me my five biggest expense categories" costs 0.1.

→ See: Token economy

What's next

• Upload data the right way — better data leads to better answers
• Pick your first Playbook — when ad-hoc chat becomes a repeatable workflow
• Reference → Memory & history — what the agent remembers and how to teach it