What problem does it solve?
A corporate treasury often sees its cash through several bank portals with separate logins, an ERP, a treasury management system and spreadsheets. Atlar's customer story on Bandwidth describes cash positions spread across multiple bank portals, so a full picture meant working through each portal one at a time. The AFP's summary of ServiceNow's submission says its treasury spent 40 to 60 hours each week collecting and reconciling data from disconnected systems, banks and entities as the company expanded across more than 50 countries.
Questions from the rest of the business add load. Microsoft's Worldwide Payment Solutions team received too many emails about financing programs, tools and policies that needed a manual answer. WPP's group treasury set out to give its global team self service access to treasury policy, delegated authority limits and how to guides.
An assistant sits on top of the data and documents treasury already owns, so staff ask a question and get an answer from the live source.
- The AFP's summary of ServiceNow's 2026 Pinnacle Awards submission says its treasury spent 40 to 60 hours each week manually collecting and reconciling data from disconnected systems, banks and entities.Finalists Named for the AFP 2026 Pinnacle Awards: Bandwidth, Google, ServiceNow (2026)
How does it work?
- Connect the sources. Bank balances and transactions come in through a multibank connection or a treasury data layer. Bandwidth uses Atlar's bank connectivity, and ServiceNow first integrated more than 10 source systems into one place. Policy answers need the treasury policy, reporting and committee papers, as in WPP's Treasurer Bot.
- Keep the user's permissions. Atlar's own post on its Claude connector says Claude queries live treasury data the same way a user inside Atlar would, with the same permissions, and that every query is logged in Atlar's audit trail. That describes Atlar's connector in general, not Bandwidth's deployment. Whatever the build, the assistant should inherit each user's access.
- Ask in plain language. A treasurer asks how much sits in euros, what cleared overnight or where the company stands this morning. The assistant turns the question into a query on live data and answers from it, not from a report prepared in advance.
- Answer policy and stakeholder questions. The same kind of assistant answers from policy documents. WPP's team uses it to look up policy and validate proposed actions, and Microsoft's bot, Leonard, answers business questions about financing programs in Microsoft Teams.
- Hand over actions. Most cited assistants answer and explain. Microsoft's Leonard was also trained to cancel financing requests in Microsoft's approval workflow. Moving money, sweeping or hedging should remain a separate, approved step, and read only access is the safer starting point.
- Audience
- Employee facing
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Internal tools, Microsoft Teams
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Employee productivity, Speed and cycle time.
Indicative value
A treasury team of six with relationships at about fifteen banks
USD 9936 to USD 96,600
Treasury time released per year
How this is calculated
Formula: teamSize * hoursPerWeek * weeks * share * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Treasury staff who pull cash and policy information teamSize, people | 6 | 6 | The reference team. |
| Hours per person per week spent gathering balances, stitching exports and finding policy hoursPerWeek, hours per person per week | 4 | 10 | Editorial assumption, replace with your own time study. The AFP's summary of ServiceNow's submission cites 40 to 60 hours a week for the whole team before it centralised its data. Source |
| Working weeks per year weeks, weeks | 46 | 46 | Editorial assumption. |
| Share of that time the assistant removes share, fraction of time | 0.15 | 0.35 | Editorial assumption. Bandwidth's treasurer says the approach saves hours every week without giving a number. |
| Loaded cost of a treasury hour hourlyCost, USD per hour | 60 | 100 | Editorial assumption, replace with your own loaded cost. |
What it leaves out: Leaves out the cost of the data connectivity and the assistant, and the value of faster decisions on funding and idle balances. The time share depends on whether the data is already centralised, which is the larger part of the work in the published cases.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Bandwidth
United States · Telecommunications · 2026
Bandwidth, a Nasdaq listed cloud communications provider, kept cash positions in several bank portals with separate logins, so a full picture meant working through each portal in turn. With Atlar's bank connectivity, its treasury team asks questions in plain language, for example how much sits in euros or what cleared overnight, and gets answers from live data instead of a prepared report. The AFP lists the project as a 2026 Pinnacle Awards finalist and says the team built its treasury platform using Atlar's bank connectivity and Claude. The treasurer says it saves hours every week.
No outcome disclosed.
ServiceNow
United States · Technology and software · 2026
As ServiceNow grew across more than 50 countries, its treasury team spent 40 to 60 hours a week collecting and reconciling data from disconnected systems, banks and entities. It first built a central treasury data architecture over more than 10 source systems, then added real time dashboards, a treasury chatbot, machine learning cash forecasting, automated FX hedging across more than 40 currency pairs and AI powered counterparty exposure monitoring. The AFP names the project as a finalist for its 2026 Pinnacle Awards. No result figures are published.
No outcome disclosed.
WPP
United Kingdom · Professional services · 2025
WPP's group treasury built a Treasurer Bot on the company's own AI environment. Its knowledge comes from WPP's treasury practices and policies, core reporting and past treasury committee papers, and the global treasury team uses it to look up policy, validate proposed actions and onboard new team members. WPP plans a lighter version for finance teams that manage FX risk in the agencies. It won the Judges' Choice award at the Adam Smith Awards 2025. No result figures are published.
No outcome disclosed.
Microsoft
United States · Technology and software · 2018
Microsoft's Worldwide Payment Solutions team received many emails from the business that needed a manual answer about financing programs, internal tools and policies. A finance analyst turned an Excel FAQ library into a Q&A chatbot on Azure Bot Service, deployed in Microsoft Teams within weeks. The bot, named Leonard, uses natural language processing with active learning, and was later connected to the financing approval workflow so users can check or cancel requests. Microsoft reported time savings without figures; the write up won the Adam Smith Awards 2018 Judges' Choice.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Bank balances and transactions in one queryable store, with account and entity metadata
- Treasury policy, delegated authority limits and how to guides in current, versioned documents
- A list of the questions stakeholders ask most often, taken from email and chat history
Systems to integrate
- Multibank connectivity or a treasury management system
- ERP for payables, receivables and forecasts
- Identity provider, so the assistant inherits each user's access
- Microsoft Teams or the treasury portal as the place staff ask
Complexity: Medium
The build needs bank connectivity or a treasury data layer with permissions that carry through to the assistant, a governed query path over cash data, and a cleaned set of treasury policy documents. ServiceNow built its central data architecture before it added AI. A policy only assistant, such as Microsoft's FAQ bot, needs far less.
- 1
List the questions
Collect two months of questions the team and its stakeholders ask, split into data questions (balances, flows) and policy questions (limits, procedures). Microsoft's bot started from an Excel FAQ library.
- 2
Centralise the data first
Put balances and transactions from all banks and entities in one governed store before adding the assistant. ServiceNow did this before it added a chatbot.
- 3
Start read only
Release an assistant that answers and shows the query or document behind the answer. Do not connect it to payment or trade initiation.
- 4
Test on known questions
Build a test set from the question list with answers checked by treasury, and run it after every change to data, prompts or model.
- 5
Extend with approvals
Add any action, such as preparing a payment batch or a sweep proposal, only as a draft for a named approver.
Guardrails
- The assistant sees only the accounts and documents the asking user may see
- Every answer from data shows the query or source behind it, and every query is logged
- No payment, sweep or trade is initiated by the assistant
- Policy answers cite the current policy version and say when no document covers the question
KPIs to instrument
- Hours per week the team spends gathering data and answering stakeholder questions, before and after
- Share of questions answered correctly on the test set, and the number escalated to a person
- Active users per month and questions per user
- Time from question to answer for the ten most common cash questions
Human in the loop
Treasury staff use the answers as input and remain accountable for decisions. Questions the assistant cannot answer, or where it finds no policy, go to a named treasury owner. The policy owner reviews the answers the bot gives on policy at a set interval.
Common failure modes
- Fragmented data behind a fluent answer
- Accounts at some banks are missing, and the answer reads as complete. Show coverage, list the accounts included and warn when material accounts are absent.
- Outdated policy
- The assistant answers from an old limit or procedure. Version the documents, retire old ones and show the version date in the answer.
- Leaking access
- The assistant retrieves data the asking user may not see. Pass the user's identity through and test with accounts of different permission levels.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A treasury assistant interacts directly with staff, so Article 50(1) requires that they are informed they are dealing with an AI system unless that is obvious from the context. It is not listed in Annex III and makes no decision about a natural person.
Rules that apply
Guidance
- NIST AI Risk Management Framework (NIST, North America). A structure to map and manage accuracy, access and misuse risks of an internal assistant on sensitive financial data.
Controls to put in place
- Role based access that applies to the assistant as it does to the user
- Audit log of questions, queries and answers
- Model inventory entry with an owner, test results and a change process
- Review of the policy documents the assistant relies on, with named owners
Frequently asked questions
- What can a treasury assistant answer today?
Bandwidth's treasury asks how much sits in euros, what cleared overnight and where the company stands this morning, and gets answers from live bank data. WPP's Treasurer Bot answers from treasury policy, core reporting and past committee papers. Microsoft's Leonard answers business questions about financing programs, tools and policies.
- Does it save measurable time?
The published cases give no figures. Bandwidth's treasurer says it saves hours every week, and Microsoft lists time savings among the benefits of Leonard. ServiceNow's 40 to 60 weekly hours of manual data work is the baseline it started from, not a result.
- Should an assistant be able to move money?
None of the cited deployments describes it doing so. A safer pattern is to keep the assistant read only and have it draft any payment or sweep for a named approver.
How to cite this page
Blits.ai AI Use Case Library, "AI assistant for corporate treasury teams", last verified 13 October 2026, https://www.blits.ai/ai-use-cases/treasury-team-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 13 October 2026: First published