01
Detect
Every metric on every source, watched against a seasonality-corrected baseline. Your event calendar kills the false alarms.
The platform
Four layers between a metric moving and a decision that's measured.
Signal comes in at the bottom. A scored decision comes out at the top. Everything in between exists so that what reaches you is a verified action — not a chart to interpret.
The loop
Niti's Learning Loop
We give you the missing system of record for your decisions and their impact.
02
Explain
Each anomaly walked from symptom to root cause — labelled Confirmed, High, or Hypothesis. It says when it is unsure.
03
Recommend
One ranked action with an estimated impact range attached. An instruction, not an observation to translate.
04
Gate & approve
Margin, supply and data gates are enforced, not advisory. What fails is suppressed and named. Approval is one tap.
05
Measure
Actuals pulled at 7, 14 and 30 days against the estimate. Accuracy per decision type recalibrates step 03.
Layer 01
One model of how revenue works. Every source extends it.
Sources don't create dashboards — they declare a role in a shared model. A new connection immediately extends your baselines, your attribution paths and the decision types available to you. No code, no build.
Spend
Where budget goes
Ad platforms, bid systems, partner and affiliate spend.
Sales
What actually closed
Your store, your marketplaces, your core system. The ledger.
Supply
What you can deliver
Stock, capacity, capital — the constraint that gates scaling.
Finance
What it costs you
Cost of goods, unit economics, contribution per unit sold.
Layer 02
Root cause, with the confidence stated.
Every anomaly is walked from symptom to root cause across three tiers — and labelled with exactly how sure the engine is. A confident wrong narrative is worse than "we don't know yet," so the engine stops rather than guessing.
Tier 1
ConfirmedA rule engine for known patterns. When the evidence matches, the root cause comes back confirmed — no statistical guesswork.
Tier 2
High · 0.82A structural causal model with an attribution percentage. The engine states how much of the movement it can explain, and how sure it is.
Tier 3
HypothesisRanked hypotheses with the falsification test attached, returned when the evidence genuinely isn't there. It names what would confirm it — never manufactures certainty.
Layer 03
A ranked queue of actions, gated before you see them.
Every recommendation carries an estimated impact range and must clear margin, supply and data-quality gates before it enters the queue. Approval is one tap; execution writes back to the platforms.
The gate cascade — enforced, not advisory
Supply cover > 7 days✓
Margin floor cleared✓
No upstream data fault✓
Scale Meta 20%HELD
P0Budget reallocation
Cut 25% of daily budget from prospecting. Move two-thirds to the channel still below saturation, hold the rest.
True ROAS 1.9× vs 3.1× target · fatigue 0.81
supply 9d · margin cleared · no data fault
+$1.08K / 14 days ($0.8K–$2.6K)
$1.08K
$1.21K
$2.10K
+$1.02K · accuracy 0.86
Layer 04
Every decision scored. The record compounds.
Actuals are pulled 7, 14 and 30 days after each decision and compared to the four weeks before it. Every decision — taken or not, favourable or not — stays in the ledger, and accuracy per decision type recalibrates the next recommendation.
Cut prospecting 15%
YES
$2.10K
0.86
Raise bids · 12 keywords
YES
$0.39K
0.89
Retire fatigued creative
YES
$0.22K
0.37
Pause spend · supply gate
YES
$0.98K
0.97
Shift 3% across channels
NO
—
—
Illustrative. The point is that a recommendation you declined is still on the record.
Cut prospecting 15%
YES$2.10K0.86
Raise bids · 12 keywords
YES$0.39K0.89
Retire fatigued creative
YES$0.22K0.37
Pause spend · supply gate
YES$0.98K0.97
Shift 3% across channels
NO——
Illustrative. The point is that a recommendation you declined is still on the record.
Connections
Connect the tools you already use.
Read-only, minutes to authorise, and each one extends the model rather than adding another dashboard.
30+ sources
'%3e%3cpath%20d='M130%2029v132c0%2014.77%2010.2%2023%2021%2023%2010%200%2021-7%2021-23V30c0-13.54-10-22-21-22s-21%209.33-21%2021z'%20fill='%23f9ab00'/%3e%3cg%20fill='%23e37400'%3e%3cpath%20d='M75%2096v65c0%2014.77%2010.2%2023%2021%2023%2010%200%2021-7%2021-23V97c0-13.54-10-22-21-22s-21%209.33-21%2021z'/%3e%3ccircle%20cx='41'%20cy='163'%20r='21'/%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
+15
Secure & compliant
Read-only by default. Encrypted throughout.
Niti never trains on customer data. Independently audited against SOC 2 and ISO 27001, with the same controls whether you're a ten-person brand or a lending group.
See the four layers running on your own numbers.
One month of spend and sales data is enough. No integration, no credit card.