How to free finance from ad hoc asks with an AI agent

The questions themselves are rarely hard. "What did we spend on contractors last quarter?" "Why is my line over budget?" "What changed since the last forecast?" Each one takes a few minutes. The cost is that they arrive all day, from every department, and they land on the same few people who are supposed to be doing analysis. The FP&A Trends Survey 2025 found that 46% of FP&A time still goes to data collection and validation rather than analysis. Ad hoc asks are a big part of that.

Bottom line: An AI agent can take most "what is the number" and "why did it move" questions off finance's plate, but only if you sort the asks first, fix the definitions the agent answers from, and put it in the channel where people already ask. Judgment calls stay with finance.

Which ad hoc asks can an AI agent take?

Most of them, if you sort them first. Ad hoc asks fall into four types, and an agent handles the first two well.

This table is the whole playbook in one view: automate the lookups and variance questions, share the scenarios, and keep judgment with people.

The variance row is the one that matters most. "What changed?" is the question finance teams hear more than any other, and it's the one a good agent answers best, because the answer is a structured drill-down rather than an opinion. Our guide to AI variance detection in FP&A software covers how that analysis works.

Step 1: Run a two-week ask audit

Before you automate anything, find out what you're actually being asked. For two weeks, log every request that reaches finance:

• Who asked and which team they're on
• Where it arrived: Slack DM, channel, email, meeting
• The ask type from the table above
• How long it took to answer, including the time to find the right file

Two weeks is enough to see the pattern. Most teams find that a handful of questions account for most of the volume, that the same people ask the same things every month, and that a surprising share are variance questions after close. That list becomes the agent's first job description, and the time column gives you a baseline to measure against later.

Step 2: Fix definitions before you switch anything on

An agent answers from your data exactly as it's defined. If "headcount" means one thing in the HRIS and another in the budget, the agent will give a confident answer that finance then has to correct, which is worse than no agent at all.

The AFP's best practices for managing ad hoc requests make the same point: agree on shared definitions before debating performance. Before launch, lock down:

1. The metrics people ask about most. OpEx, headcount, budget remaining, ARR. Write down one definition for each, and make it the one the agent uses.
2. One source of truth per number. Actuals from the ERP, plan from the planning model, headcount from the HRIS. The agent should answer from connected live data, not an exported copy.
3. Mappings. Vendors to departments, accounts to budget lines. Most wrong answers trace back to a bad mapping, not a bad model.
4. Who can see what. Budget owners should see their own lines, not their peers' salaries. We cover the permission model in detail in permission-aware AI agents for budget owners and role-based access controls for FP&A.

Step 3: Put the agent where the asks already arrive

People already send their questions to finance in Slack or Teams. Put the agent there too. A separate portal is one more place to remember, and people will go back to messaging a human.

A launch that works usually looks like this:

• One channel to start. A single #ask-finance channel, or the budget owners' channel you already have. Don't turn it on everywhere on day one.
• A "try asking" list. Pin five real questions from your audit, word for word. People copy examples faster than they read instructions.
• A clear handoff. When the agent can't answer, or the question is a judgment call, it should say so and route the thread to a named person with the context attached.
• Traceable answers. Every number should link back to its source, so a budget owner can check it and finance can defend it.

With Aleph Agent, a budget owner tags the agent in Slack or Teams and gets an answer drawn from the same connected data the finance team models on, with cited sources and a follow-up that keeps the context.

Step 4: Answer "what changed?" before anyone asks

The best ad hoc ask is the one that never gets sent. Once the agent handles questions reliably, push the most common answers out on a schedule:

• Monthly variance notes after close. Send each budget owner a short note on their lines: what moved, by how much, and the vendor or transaction behind it. AI flux analysis can draft that note in seconds, and finance reviews it before it goes out. See automating month-end close commentary with AI.
• Threshold alerts. Flag a department when spend passes a set level, or when billing revenue and ERP revenue stop matching. Aleph's Checks and Slack alerts do this without code.
• Budget-remaining snapshots. A weekly post in each department's channel answers the most common lookup before anyone types it.

This is where the time comes back. Answering faster helps. Getting the answer out before anyone has to ask helps more.

Step 5: Measure how many asks the agent handles

If you can't show the deflection, you can't defend the rollout. Track four numbers against the baseline from your audit:

1. Deflection rate: the share of asks the agent answers without anyone in finance touching them.
2. Time to answer: minutes for the agent, compared with hours or days before.
3. Escalation rate: how often the agent hands off to a person. It should fall as definitions improve.
4. Wrong-number incidents: any time a stakeholder acted on an answer that turned out to be wrong. Investigate every one, because most point to a definition or mapping to fix.

Review these monthly for the first quarter. Expect deflection to start modest and rise as you add questions to the "try asking" list and fix the mappings the escalations expose.

Which tools can answer finance questions in Slack or Teams?

Several FP&A and finance platforms now ship an agent or assistant, but they differ on where people can ask and what data they answer from. As of September 2026:

• Aleph: best for finance teams that want budget owners to ask in Slack, Teams, dashboards or the web app and get answers from live data connected across the ERP, HRIS, CRM and billing. It also connects to Claude and ChatGPT through MCP. Aleph Agent.
• Cube: best for spreadsheet-first teams. Its AI agents answer questions in Slack, Teams, Excel, Google Sheets and the web.
• Datarails: best for teams that want a human in the loop. Datarails Desk collects requests from email, Slack and Teams into one queue and drafts answers for finance to review before sending.
• Abacum: best for teams that want stakeholders to self-serve inside Abacum's web app. Abacum Intelligence answers within the permissions finance sets, and connects to AI assistants through MCP.
• Anaplan: best for enterprises already on Anaplan. CoPlanner answers plain-language questions on live Anaplan models, inside Anaplan.
• Microsoft Finance Agent: best for finance staff working in Excel and Outlook, for tasks like reconciliation and variance work. It's built for the finance team rather than for business users asking questions (Microsoft Learn).
• Slack enterprise search and general AI assistants: good at finding documents and messages, but they don't sit on a governed finance model, so treat their numbers as unverified.

For the wider category, see AI agents for finance and FP&A.

What should a one-person finance team do first?

Start with the variance question. If you're the only finance person, you can't run a two-week audit and a launch plan in parallel with close. So do the one thing that removes the most interruptions:

1. Connect the ERP and the budget to one place, so actuals and plan sit side by side.
2. After each close, send a short "what changed" note per department, drafted by AI and checked by you.
3. Open one Slack channel where people ask, and let the agent take the lookups.

That covers most of what reaches a solo finance lead. The rest of the playbook can wait until the second quarter. Our guide to automating FP&A workflows lists the other automations teams usually add next.

Deloitte's Q4 2025 CFO Signals survey found that 54% of CFOs say integrating AI agents into finance will be a transformation priority. The teams that get value from it will be the ones that treat the agent as a process change, not a feature to switch on.

Hand the routine questions to Aleph Agent

Aleph Agent answers budget owners' questions in Slack, Teams and dashboards from your live, connected finance data, with cited sources and your existing permissions. Finance keeps the judgment calls and gets the rest of the week back.