FP&A, controllership and finance business partners
DeepGarden for finance
Finance teams do not lack numbers. They lack agreement about them. A large share of the time between a question and a decision goes on establishing whose version of a figure is correct, and that time is invisible in every plan until it costs a week.

Three systems, three definitions of the same word
Each has a defensible definition of revenue, and none of them agree.
Giving everyone the ability to build their own view of margin without first agreeing what margin means multiplies the number of defensible answers instead of reducing it.
Where the answer is currently spread
- General ledger and consolidation system
- Billing and revenue systems
- CRM, for pipeline and bookings
- Procurement and payables
- Workforce system, for people cost
Who is involved
- Controller
- That anything reported reconciles to the ledger and can be evidenced at audit.
- FP&A lead
- Variance against plan, explained by driver rather than just quantified.
- Finance business partner
- Answering their function's questions without building another extract.
- Internal audit
- That a reported figure can be traced to source months later.
Questions this makes routine
These all cross a system boundary, which is why they currently get asked once a quarter rather than once a week.
- Which cost centres moved most against plan this quarter, and what drove it?
- What does margin look like by product line once support and delivery costs are counted?
- Where is revenue recognition timing distorting a month-on-month comparison?
- Which accounts are expensive to serve relative to what they pay?
- How much of this quarter's variance is volume, price or mix?
What an answer looks like here
A question, the clarification it needed, the answer, and the query behind every figure.
Multi-system · finance
AskWhich accounts are expensive to serve relative to what they pay?
Clarifying question
Cost to serve is defined two ways in your catalog. Which one?
- Direct delivery and support cost only (certified)
- Fully loaded, including allocated overhead (draft)
Eleven accounts have a cost-to-serve ratio above 0.6. The highest is Northwind Logistics at , against annual contract value of .
Calculation
The ratio uses the certified direct definition, which excludes allocated overhead.
cost_to_serve_ratio = (delivery_cost + support_cost) / acv
Query evidence
- Dataset
- Finance ledger + CRM + delivery
- Rows
- 11
- Duration
- 631 ms
- Executed
- 2026-07-14 10:04:19
- Status
- succeeded
- Mode
- read-only
View assembled from the result
- 0.83
- 0.76
- 0.71
- 0.68
- 0.64
What DeepGarden does about it
What changes, specifically, and what lands in front of the person who asked.
One agreed definition, enforced
Margin, recognized revenue, cost to serve, headcount cost and the rest are defined once, certified, and used by every answer. When two departments have historically meant different things by the same word, that gets resolved in one place rather than re-litigated in every report.
Your close calendar, not the Gregorian one
Period logic follows your fiscal calendar, and comparison rules stop a figure computed on one basis being set against a figure computed on another.
A number that can be evidenced at audit
Every figure opens onto the executed query, the tables it touched, the row count and the timestamp — retained for as long as you require, which is the form auditors actually want.
Why the output holds up
The parts of the governed path that matter most in this particular setting.
- Read-only against the ledger. There is no configuration in which a question posts an entry.
- Derived figures show the formula and the assumptions, so a margin number can be checked rather than believed.
- Every figure is re-checked against the returned rows before the answer is shown; anything that does not reconcile is flagged rather than published.
- Comparison guardrails prevent metrics computed on different bases being presented as if they were comparable.
How implementation would work
Roughly six to ten weeks for a first domain
Bring last quarter's most argued-about number.
We will reproduce it, show the query, and you can decide whether the provenance is good enough.