By Zain ul Abideen | CompassPoint Consulting | June 2026
There is a lot of noise about AI in finance right now. Most of it is marketing. A lot of it is being pushed by IT vendors and technology consultants who have something to sell.
The result is a feeling I see in every second founder I speak to. The fear of missing out. Everyone knows what AI is by now. Very few have any clear idea what to do with it. Least of all in SMEs, where leadership is buried in internal problems and has no time to sit and think through an AI finance strategy.
At CompassPoint, we have spent the past few years bringing CFO thinking and modern technology into SME finance functions. I decided to write this article without the hype, the marketing language, or the LinkedIn-friendly promises.
Just the reality. And the work it takes to get there.
Every SME finance function I have walked into shares the same set of pain points. Management accounts that arrive on day 15, not day 5. A cash position the founder never quite trusts. Variance commentary written in a panic on the morning of the board meeting. Errors that slip through the cracks. A founder who wants an answer to a numbers question now, and a finance team that needs three days to give it.
The ten use cases below are ordered to match those pains. The first few solve the daily grind. The middle ones build governance and responsiveness. The last few are about the strategic moments. The board. The founder’s hardest questions. The raise.
The 10 use cases
1. Fully automated management reporting, consolidation and 3-statement model
When the chart of accounts is clean and the systems are mapped, the consolidation that used to consume the first ten days of every month runs on its own. The 3-statement model rebuilds automatically. Variance flags surface before anyone opens the file. Your finance team stops being formatting clerks and starts telling leadership what the numbers mean.
What used to happen in weeks happens in days.
2. 13-week rolling cash flow, refreshed every Monday
Most founders ask for a 13-week cash flow. Few finance functions deliver it consistently because the manual lift is too high. AI changes that equation, when AR ageing, AP commitments, payroll cycles, VAT timing and debt service all flow in from clean source systems.
The rolling view refreshes every Monday morning. The cash position is current, not week-old. A founder who knows their cash position weekly makes different decisions to one who finds out monthly.
3. Variance analysis and commentary at month-end
Budget vs actual is the most expected output of any management pack. The narrative explaining the variances is where most teams run out of time.
With AI drafting the first version, your finance function delivers consistent variance commentary every month. Sales below budget by 8%, driven by a delayed renewal. Marketing spends over by 12%, attributable to the May campaign push. The CFO reviews, sharpens, signs off. The pack closes faster. The quality is higher because the system never gets tired and never skips a line.
4. Board pack narrative writing
The most time-consuming task in any finance month is not the numbers. It is the words around the numbers.
Your finance function can have AI draft the first version of every board pack narrative, pulled from actuals, budget, prior period and operational data. The CFO edits, sharpens, signs off. The result lands faster, with less staff cost, and reads with the consistency the board appreciates.
5. Anomaly detection on management accounts
Errors in management accounts are usually not catastrophic. They are small, regular, and consistently miss the human eye. A revenue accrual booked twice. A cost centre tagged to the wrong department. An intercompany balance that does not eliminate cleanly.
AI scans every month-end pack against a learned baseline and surfaces what looks unusual. The CFO decides what is real and what is noise. This catches things the team missed because the team did not have time to look that closely.
6. Ad-hoc analysis and scenario modelling, in minutes
“What happens to our gross margin if Client X reduces their retainer by 20%?”
That used to be a two-day exercise. With the model connected and the data loaded, scenarios run on demand. The founder gets the answer in the same meeting they asked the question. Decision velocity goes up. The finance function is no longer the bottleneck.
7. Live dashboard chat: ask your numbers a question in plain English
When accounting, CRM, project and HR systems are connected into a single data layer, an AI agent can sit on top of it.
A founder asks: “What is my net margin on retainer clients vs project clients this quarter?” The answer comes back with the source data attached. No SQL. No Excel rabbit hole. No “let me come back to you tomorrow.” This is the feature founders want to keep once they have seen it. It changes the relationship between leadership and the numbers.
8. Client and cohort profitability analysis
This is the use case that surfaces material findings most often. A founder asks “which clients are actually profitable once we include time, overheads and indirect costs?” When the data supports it, AI runs the analysis across project, time-tracking and cost data. The answer is rarely what the founder expected.
But this is also where the foundation matters most. If revenue, costs and time are not tagged at source, no AI in the world can build a profitability view that holds up.
Profitability is a data structure problem first. AI is just the speed multiplier.
9. Building a fundraising-ready financial model
Most founder-built fundraising models are two things. Optimistic. And fragile. Investors pull the thread on the unit economics and the whole thing unravels.
This is not work a founder should be doing alone. Your finance function, with AI on the structural heavy lifting, can deliver an investor-grade model in days. 36-month P&L, balance sheet, cash flow. Unit economics by tier. Defensible pricing logic. Three scenarios. A sensitivity layer. Stronger, not weaker, because it has been stress-tested more in the same time.
10. Investor diligence Q&A prep
A raise generates roughly 20 questions every investor will ask. Some predictable, some sector-specific. All require a defensible answer that ties back to the model.
Your finance function can run the deck and the model through AI, pre-empt the questions, draft the answers, and stress-test them. The founder walks into the diligence meeting with the answers ready. Better still, the founder spots the weak spots in their own thinking before the investor does.
Why most SMEs cannot have these yet
Some of our clients have several of these use cases running today. Other clients are on the way to achieve most use cases. We got there by doing the unglamorous work first. Cleaning the data. Redesigning the chart of accounts. Building a single layer of truth. Documenting the close calendar. Only then did the AI agents go on top.
Other clients we work with are still on that path. They will get there too. The foundation work has to happen before any AI agent becomes useful.
The technology is not the problem. The foundation is.
Here is what gets in the way.
Month-end closes that arrive late. Or arrive on time but with the wrong numbers.
If your management accounts close on day 15 instead of day 5, AI does not help. It just delivers your slow report faster. If your numbers are wrong when they arrive, AI confidently presents incorrect commentary on incorrect figures. Garbage in. Confident garbage out.
Data scattered across disconnected systems.
Accounting in Xero. CRM in HubSpot. Project data in Notion. Time tracking in a spreadsheet. Operational data in someone’s head. AI cannot reconcile a business it cannot see. The dashboards described above only work when these systems feed into a single, structured layer.
A chart of accounts that does not match how the business actually runs.
I have lost count of how many SMEs have a chart of accounts that looks tidy on paper but does not reflect their commercial reality. Revenue lumped into one line when there are five distinct streams. Overheads mixed with direct costs. Project codes that nobody uses consistently. AI inherits whatever structure it is given. Bad structure produces bad analysis.
No documented processes. Every cycle rebuilt from scratch.
Most SME finance functions run on the memory of one or two people. There is no SOP. The reporting works because Sarah remembers how the consolidation goes together. When Sarah leaves, the system breaks. AI cannot codify a process that was never documented in the first place.
No single source of truth.
This is the one that quietly kills more AI projects than anything else. Three versions of revenue. Two versions of gross margin. The board pack disagrees with the budget tracker which disagrees with the investor update. AI sitting on top of contradictory data does not produce insight. It produces confident confusion.
Fix any of these in isolation and you have an improvement. Fix all of them and you have a finance function that can actually use AI.
The data infrastructure redesign that has to happen first
The foundation work is not a list of small fixes. It is a redesign of the data infrastructure underneath the business.
What we do, in practice, is bring every relevant system into one place. Accounting from Xero, QuickBooks, NetSuite or Zoho. CRM from HubSpot or Salesforce. Project management. Time tracking. Payroll. All of it flows into a modern data layer, built on Azure SQL and Microsoft Fabric, structured into clean reconciled layers and reshaped to match how the business actually operates.
Only then does the reporting layer go on top. Power BI dashboards built on logic a CFO has signed off, not a developer has guessed at.
Only then does the AI layer go on top of that. Claude reading from a single source of truth, not from five conflicting spreadsheets.
There is no shortcut. Every business I have seen try to put AI on top of broken data has produced the same outcome. Confident output that nobody trusts.
Here is what most SMEs have not yet processed. This redesign is not optional. It is what AI demands. And the businesses doing it now will have a structural advantage over the ones still consolidating in Excel three years from now. Not because they will have better AI. Because they will have better foundations for AI to work on.
The early adopters will not be the ones with the cleverest AI. They will be the ones who fixed the data first.
The pilot, not the autopilot
Even with the foundation in place, AI does not run a finance function on its own.
Modern airplanes fly themselves for most of the journey. The autopilot is genuinely capable. But every commercial flight still has a pilot in the seat. Not because the autopilot might fail. Because a human is accountable for the outcome.
AI in finance is the same. The model will run. The narrative will write itself. The variance commentary will land in the inbox. None of that means a CFO is no longer needed. It means the CFO’s job is different now. Less typing, more judgement. Less consolidation, more challenge. Less commentary drafting, more decision quality.
AI has no accountability. If the model is wrong, AI will not lose its job. If the variance commentary misses the real story, AI will not face the board. If the cash flow forecast is off by 20%, AI will not be the one explaining it to the founder.
The qualified human is accountable. That is not a limitation of AI. It is the entire point.
What this means for your business
If you are a founder or a CFO sitting on the question of “how do we use AI in our finance function?” the order matters.
Fix the data first. Map your systems into one layer. Rebuild the chart of accounts so it matches your operating model. Document the close calendar. Build a single source of truth. None of this is glamorous. All of it has to happen first.
Then automate. Once the foundation is right, AI sits on top of it cleanly. The 10 use cases above become reachable. Some in weeks, some in months.
Then deliver insight. Once the work is flowing automatically, the team has time to do the thinking AI cannot do. Strategic decisions. Commercial challenge. Real partnership with leadership.
Most businesses skip step one and go straight to step three. The result is a dashboard that looks impressive and tells leadership nothing reliable.