COMPASS connects close, variance, forecast, plan and board pack on one spine. AI agents size the month, find the drivers and draft the commentary with evidence — your people review, attest and decide.
The close runs long, the variance pack is stitched together by hand, and AI pilots stall on data and complexity. Three benchmarks tell the story.
Only about 31% is left for the analysis, insight and decision support that leadership actually asks for.
Every extra day of close is a day less to explain the month and re-forecast before leadership meets.
The blockers are data quality, complexity and talent — not appetite. AI needs a governed data spine and a workflow to land in.
COMPASS is not another reporting layer. It is the working surface for the whole FP&A cycle, built so that every number can be traced, every explanation carries evidence and every agent answers to a person.
Monitor, close, explain, forecast, plan, decide and report run on the same definitions — so the board pack agrees with the ledger.
Not “revenue was up”. Which lines, which drivers, how much of the gap — and the source behind every sentence.
Five supervised agents do the legwork inside the workflow. Nothing is attested, committed or posted without a named person.
The driver model reproduces last month’s actuals before it projects anything — so the plan starts from the truth.
Put a range, not a point, in front of the CFO — and show what would have to be true for each case.
COMPASS is built on the DaasLabs Data Fabric Framework: it reads the bank’s ledger and feeds where they live. No new silo.
A scripted walkthrough of the Explain the Month workflow with illustrative numbers from the demo environment. It plays when you scroll here; pause it or jump to any step.
The Variance Analyst sizes the gap against budget and counts the lines that clear materiality.
Each material line is decomposed into drivers that sum back to it. Rolled up, the month looks like this.
The Commentary Writer drafts the account-owner commentary and links every claim to its source.
Site 01 closed April $15.7M favourable to budget. Volume added $9.8M as Customer 04 and Customer 11 pulled orders forward from May; price realisation added $4.1M and a richer Division A mix $2.6M. FX translation cost $1.9M as the dollar strengthened. Timing of $1.1M reverses in May.
A named reviewer edits the draft, checks the evidence and attests. The preparer can never attest their own line.
Attested lines roll up entity by entity into the board pack — with provenance carried all the way through.
COMPASS agents are part of the DaasLabs supervised digital workforce: every agent has a named owner, a set of tools, an autonomy level and a policy it cannot step outside. Here is how that works for FP&A.
The agent receives a goal and breaks it into steps it can execute with COMPASS tools.
Agents work where the effort is. The decisions that carry accountability — attesting commentary, approving the forecast, committing the plan and signing the board pack — stay with people.
Each COMPASS agent starts low and earns more autonomy on evidence. Levels, owners and 7-day metrics below are read live from the Digital Workforce registry.
variance_enginemateriality_triagedriver_decompositioncommentary_storeevidence_lookupnarrative_rolluprolling_forecastsignal_registerbacktestclose_calendartask_historynotifydriver_modelscenariosmulti_agent_debateLines above a dollar floor and a percentage of budget always go to a reviewer. Thresholds are configuration, not prompts.
Routing follows who owns the entity and how big the number is — the same chains your planning approvals use.
Preparer is never reviewer. An agent that drafts a line cannot approve it, and neither can the person who asked it to.
Committed plans and scenarios are locked. Agents work in a sandbox version; only people commit.
Agents may prepare a journal or allocation run; posting always needs a named approver.
A commentary line without a resolvable source is rejected. Provenance is carried into the board pack.
Below min_confidence_auto 0.90 nothing goes straight through; runs stop after max_tool_calls 8 and escalate.
One switch stops every agent from starting a run or applying an action — across COMPASS and the wider workforce.
Every run, tool call, proposal, approval and override is kept, with PII redacted from prompts sent to the model.
The war room runs the plan past a controller, an FP&A analyst, a bullish strategist, a sceptical bear and treasury & risk. It records consensus and dissent in a CFO memo. People commit the plan.
Baseline from the driver model reproduces March actuals within 0.3%. Starting point: revenue +3.8%, EBITDA margin 19.6%.
Gap to the ambition is $42M EBITDA. Price +1.5 pts and Division C mix close about 60% of it.
Order-intake signals in Division A are running +8%. Pull the volume driver up two points.
Two of the last three upside cases missed, and Customer 07 is 14% of Division A. Haircut that uplift by half.
A 5% stronger dollar erases $9M; hedge cover is 60%. Keep a P10 downside case alongside the base.
Consensus: price and Division C mix are credible; volume uplift at half the bull case.
Dissent recorded (bear): Division A volume depends on one customer; revisit at the June re-forecast.
Four numbers decide whether an agent keeps, gains or loses autonomy. Values are the FP&A squad’s last seven days in the Digital Workforce demo.
Two of the moments finance teams dread most — the re-forecast and the last days of close — become continuous and predictable.
Actuals for closed months, forecast for open ones. Each month the cut-off rolls forward and the horizon extends — no annual reset.
Every task has a working-day gate and a history. Close Sentinel predicts the landing and escalates what will miss — before it misses.
Everything in COMPASS sits on one of the seven steps of the FP&A cycle, plus the AI and platform layers underneath. Modules with a blue top edge open a real screen.
What has moved that will hit the P&L?
Are the numbers trustworthy?
Why did we beat or miss?
Where do we land?
What are we committing to?
Which path wins?
What do we tell leadership?
Who does the legwork?
How is it run?
Real screens from the COMPASS demo environment (data is illustrative and anonymised). The tour advances on its own; click the screen to enlarge and zoom.
Enlarge
The spine of the product: seven steps from ledger to landing, each with the question it answers and the outcome it must produce.
A fair view by category. Packaged planning suites are strong where they have always been strong; COMPASS is built to explain the month, put agents to work under control and run on the data platform you already have.
| Capability | Spreadsheets & email | Packaged planning / EPM suites | BI dashboards | COMPASS by DaasLabs |
|---|---|---|---|---|
| Time to first value | DaysFamiliar, but manual every month (yes) | MonthsTypical programme before go-live (no) | WeeksReporting only (partial) | 30–45 daysPilot on a live close (yes) |
| Explains variances with evidence | Manual (no) | PartialVariance reports; commentary by hand (partial) | PartialShows what, not why (partial) | YesTriage, drivers, provenance per line (yes) |
| AI drafts with human sign-off | No (no) | EmergingAdd-on assistants (partial) | Q&A onlyNot in a sign-off workflow (partial) | YesFive supervised agents, approvals (yes) |
| Driver model tied to actuals | FragileLinked workbooks (partial) | YesMature modelling (yes) | No (no) | YesCalibrated to last closed month (yes) |
| Close prediction | No (no) | Task listsChecklists, little prediction (partial) | No (no) | YesPredicted landing, escalation (yes) |
| Runs on your data platform | Files (no) | Own storeData copied into the suite (partial) | YesReads the warehouse (yes) | YesBuilt on your data fabric (yes) |
| Delivered and handed over by a team | No (no) | Via partnersIntegrator-led (partial) | No (no) | YesDaasLabs team + knowledge transfer (yes) |
| Packaged statutory consolidation & tax | No (no) | YesCore strength (yes) | No (no) | With LEDGER360 / NEXUSPaired accelerators (partial) |
COMPASS comes with the DaasLabs team that connects your ledger, configures the cycle and runs the first closes with you — then hands it over. How we deliver →
We start where the pain is sharpest — usually explaining the month — and prove it on a live close before anything scales.
Autonomy levels, policy limits, segregation of duties, kill switch and a full audit trail — the same controls as every DaasLabs agent. Governance & controls →
Move the sliders to match your finance function. The result is an estimate to frame a conversation, not a promise — the pilot measures the real number.
How it’s calculated. Prep hours freed = team × 140 productive h/month × prep share × automation share. Commentary hours freed = lines × hours per line × 60% (AI first draft; reviewer time kept). Capacity = hours freed ÷ 140. Turnaround: today, commentary starts after close and takes lines × hours ÷ (team × 6 h/day) working days; with COMPASS, drafting overlaps the last two days of close and review takes 40% of the effort. Benchmarks are industry figures, not COMPASS results.
We prove value where it hurts most — explaining the month — and extend across the cycle once the numbers are trusted.
Connect the GL, budget and forecast; load account and entity hierarchies.
Run triage, drivers and AI commentary on a live close with your account owners.
Add rolling forecast, close prediction, driver model and scenarios.
Fixed scope, 30–45 days, one or two entities. Success measures agreed up front.
DaasLabs delivers the remaining steps on your platform, then transfers ownership to your team.
DaasLabs operates COMPASS and the FP&A squad, including for the bank’s corporate clients.
Book a walkthrough with your data model in mind — or first check where your finance function sits with the maturity self-assessment.
Benchmarks are third-party industry figures and are not COMPASS results. Screens and figures from the COMPASS demo environment use illustrative, anonymised data.