Choosing FP&A Software: Make the Demo Follow Your Workflow
A workflow-based method for FP&A software demos: what to prepare, what to ask vendors to show, and how to record what was demonstrated, what depends on services and what still needs proof.
Finance Workroom editorial · 7 min read
A vendor’s standard demo runs on prepared data, along a path the vendor knows well. That is a reasonable way to screen options. On its own, it does not establish whether the software fits the way your team closes, forecasts and reports.
To find that out, ask each shortlisted vendor to demonstrate the same workflow against acceptance criteria you set in advance, and record what you actually saw. For this evaluation, distinguish three stages: an introductory demo to screen options, a tailored workflow demonstration for the shortlist, and a scoped proof of concept only where a consequential requirement is still unresolved. This article covers the middle stage. Even a well-run tailored demo shows only that the workflow can work in an agreed setup. It does not prove production readiness.
Hypothetical example. The finance team, workflow and table entries below are invented to illustrate the method. They do not describe any product.
1. Define one workflow and the result you need
Pick one recurring process the team wants to improve. In our example, an Excel-heavy finance team runs a monthly cycle with five steps: refresh actuals from the general ledger, update one departmental forecast assumption (Marketing’s planned Q4 program spend), review the impact, approve the change and produce the management reporting pack.
Before any demo, write a few acceptance criteria you could observe on screen. Ours:
- Refreshed actuals reconcile in total to the agreed trial balance or control totals.
- Department-level figures reconcile to an agreed departmental general ledger report or another appropriate source.
- The changed Marketing assumption flows through the forecast and into the reporting pack without manual rebuilding.
- A reviewer can see which version they are looking at, what changed, who changed it and whether it is approved.
- The team can name who will maintain account and department mappings and model logic after go-live.
These are examples, not universal requirements. What matters is agreeing on them internally and sharing the same set with every shortlisted vendor.
2. Prepare a small, representative demo pack
Give vendors enough to run the workflow, and no more:
- Synthetic or properly anonymized data with your real structure: a few departments, a few months.
- Your account and department mappings.
- A small existing model, or a written description of its logic.
- One reporting output you want reproduced.
- One realistic exception (see step 4).
- The acceptance criteria, and who from your team will attend in which role.
Don’t send confidential company files without approved handling under your company’s policies. Stripping names may not be enough. NIST describes how researchers re-identified individuals in released datasets that had names and similar identifiers removed, by combining the data that remained (NIST SP 800-188). In finance data, a one-person department’s salary line or a customer recognizable by its revenue can reveal who is behind the numbers.
If two of your own reports disagree on figures you plan to use, settle which is authoritative before the demo. Our guide to checking finance data when reports disagree walks through how.
A small sample tests the workflow. It doesn’t validate production data volumes, performance or every integration.
3. Make the demo follow the workflow
Ask the vendor to show each stage rather than describe it:
Stage 1: Refresh actuals
Load or refresh the sample actuals. Tie the totals to the trial balance or agreed control totals, and the department figures to the agreed departmental source. Show where the mapping lives.
Agreed exception: unmapped account (step 4)
Stage 2: Change an assumption
Change Marketing’s Q4 program spend using the permissions intended for an ordinary finance user. The vendor can operate the demo; the point is to establish which permissions the change requires. Show who is allowed to change assumptions, formulas, mappings and permissions.
Stage 3: Update the forecast
Show the change flowing through. Pick a forecast number and trace it back to its inputs and logic on screen.
Stage 4: Review and approve
Show the version, the change, who made it and its approval status, and what happens if the reviewer rejects it.
Stage 5: Produce the output
Show the reporting pack updating. Does it stay linked to the model, or does someone rebuild or paste it?
At each stage, note where Excel remains in the workflow, if anywhere, and whether that matches your requirements. Some teams want to keep Excel for analysis or outputs; others want it out of the core process. Neither is right by default. For every step, also ask who does this after go-live: a finance user, an administrator, IT, the vendor or a partner.
4. Add one agreed exception
Continue the example: September actuals include a new general ledger account that isn’t in the mapping yet. Agree this exercise with the vendor in advance. The goal is to see the process, not to catch anyone out.
Ask them to show how the unmapped account is detected, who is notified or assigned to fix it, how the mapping is corrected, and how the correction appears in the reconciliation and the reporting pack.
Then record what kind of work you saw: ordinary finance-user work, administrator configuration, specialist implementation, custom code, or a capability that was only described.
If AI features appear, such as a suggested mapping for the new account or drafted commentary, ask which outputs are generated suggestions and which are deterministic calculations. Also ask whether a suggestion passes through the same approval control as a manual change. NIST notes that generative AI can produce confidently stated but erroneous content (NIST AI 600-1). A suggested mapping is a proposal for someone to approve, not a correction.
5. Record evidence and decide what to verify next
Record each requirement with a descriptive status, not a score:
- Demonstrated in the agreed setup
- Described, not demonstrated
- Requires configuration or services
- Requires further validation
A requirement can carry more than one status. For example, exception handling may be demonstrated and still require an administrator to configure it.
Actuals reconcile: totals to trial balance, departments to agreed departmental source
- Evidence to request
- Refresh of the sample; total and department tie-outs on screen
- Delivery and maintenance owner
- Setup: implementation team. Ongoing: to confirm
- Status / unresolved question
- Demonstrated with file import. Live ERP connection requires further validation
Marketing assumption flows to forecast and pack
- Evidence to request
- Q4 spend changed under finance-user permissions; pack updates; number traced to inputs
- Delivery and maintenance owner
- Model logic built at implementation
- Status / unresolved question
- Demonstrated. Open: can FP&A change model logic without services?
Version, change and approval visible
- Evidence to request
- Reviewer view with version, author, status; one rejection
- Delivery and maintenance owner
- Approval rules: administrator
- Status / unresolved question
- Described, not demonstrated. Request a follow-up
Unmapped account detected and corrected
- Evidence to request
- Exception flagged, assigned, fixed and reflected in pack
- Delivery and maintenance owner
- Mapping fix: administrator
- Status / unresolved question
- Demonstrated; requires configuration. Can the finance team do it?
Ownership of mappings and logic after go-live
- Evidence to request
- Written responsibilities; admin screens
- Delivery and maintenance owner
- Finance, vendor or partner: to confirm
- Status / unresolved question
- Described, not demonstrated. Get written confirmation
| Workflow requirement | Evidence to request | Delivery and maintenance owner | Status / unresolved question |
|---|---|---|---|
| Actuals reconcile: totals to trial balance, departments to agreed departmental source | Refresh of the sample; total and department tie-outs on screen | Setup: implementation team. Ongoing: to confirm | Demonstrated with file import. Live ERP connection requires further validation |
| Marketing assumption flows to forecast and pack | Q4 spend changed under finance-user permissions; pack updates; number traced to inputs | Model logic built at implementation | Demonstrated. Open: can FP&A change model logic without services? |
| Version, change and approval visible | Reviewer view with version, author, status; one rejection | Approval rules: administrator | Described, not demonstrated. Request a follow-up |
| Unmapped account detected and corrected | Exception flagged, assigned, fixed and reflected in pack | Mapping fix: administrator | Demonstrated; requires configuration. Can the finance team do it? |
| Ownership of mappings and logic after go-live | Written responsibilities; admin screens | Finance, vendor or partner: to confirm | Described, not demonstrated. Get written confirmation |
Then sort open items into three kinds:
- A genuine mismatch with a non-negotiable requirement. For example, reviewers can’t see approval status, and the vendor confirms there’s no supported way to provide it. Either the requirement changes or the option leaves the shortlist.
- An acceptable dependency. Configuration or services you can accept, provided the owner, scope and ongoing maintenance are clear.
- An unresolved question. Ask for written confirmation, or run a proof of concept scoped to that requirement alone. The more consequential the requirement, the stronger the case for testing over taking someone’s word.
Finally, check what you saw against what you’re being offered. Confirm that the modules or edition used in the demo are the ones in the proposal. Also confirm that the configuration and services you noted are assigned to someone: your team, the vendor or a partner.
If you’re still forming a shortlist, our FP&A software page can help you explore fit before you script the first demo.
Keep reading
Six Finance Reporting Tasks Worth Automating
From recurring exports to first-draft variance commentary: six reporting tasks where spreadsheet improvements, rules or AI assistance can help, and what your team still owns.
11 min readBefore AI: Five Checks When Finance Reports Disagree
A worked example of conflicting revenue figures, five checks to resolve them, and how to decide whether to proceed, narrow the task or pause before using the data with AI.
9 min readYour First Finance AI Use Case: Start with a Report You Can Check
Turn "use AI in finance" into one task you can check: a completed task card, an independent reference result, and five acceptance tests, including a missing-budget exception.
7 min read