Altisly
Altis AI — Agentic Intelligence Layer

One AI agent.
Every product.
Out of the box.

Altis AI is not a chatbot. It is a modular agentic intelligence system that perceives data, reasons across context, and executes actions — autonomously or with human-in-the-loop approval — inside every Altisly product.

Born inside Atreasury. Built to plug into any financial operation — reconciliation, compliance, merchant management, forecasting, and settlement — without configuration overhead.

altis-agent · treasury-context · live
01
OBSERVECitibank UK balance: $12.2M → $9.8M (−$2.4M, −19.6%)
02
CONTEXTCross-referencing settlement schedule, pending SWIFT queue, FX positions...
03
DETECTGap source identified: SWIFT MT103 TXN-9182 delayed 4.2h past SLA
HIGH PRIORITYLIQUIDITY RISK
04
REASONModelling 3 resolution paths. Optimal: sweep SC GBP → Citi USD at current FX (GBPUSD 1.2741)
05
COSTSweep cost: $1,200 FX · Overdraft avoided: $4,200 · Net benefit: $3,000
06
ACTIONRouting approval request to treasury@co.com · Confidence: 97%
AWAITING APPROVAL
7awaiting next event
Multi-step reasoningContext memoryExplainable outputsHuman-in-the-loopAutonomous executionReal-time learning

<500ms

Agent Response Time

97%+

Decision Accuracy

6

Products Powered

24/7

Autonomous Monitoring

0

Prompt Engineering Required

What Altis AI does inside the treasury system

Altis AI was built to solve the hardest problems in corporate treasury — where the cost of a wrong decision is measured in millions. These are its core capabilities in native deployment.

Anomaly Detection on Cash Flows

Real-Time

The agent monitors every incoming balance update against 90-day behavioural baselines per account, currency, and time-of-day. It detects unexpected inflows, outflows, missing credits, and unusual sweep patterns — flagging them in real time with explanations.

> ANOMALY: GTBank NGN balance dropped 34% vs 7-day avg > Cause: Missing batch settlement from merchant run #1847 > Severity: HIGH · Recommended action: Investigate + hold sweep

Liquidity Gap Prediction

Predictive

The agent forecasts intraday and multi-day liquidity gaps before they occur by combining real-time balances with the pending settlement queue, known payment obligations, historical inflow patterns, and FX movements.

> FORECAST: Citi USD gap predicted at 14:30 today > Drivers: SWIFT delay (TXN-9182) + payroll run at 15:00 > Recommended: Initiate sweep now · window closes in 47min

Automated Sweep Orchestration

Autonomous

When the agent identifies a liquidity imbalance it evaluates all available funding sources — cross-currency, cross-entity, cross-bank — calculates the cheapest path including FX costs, and either executes automatically or routes an approval request with full workings shown.

> SWEEP PLAN: SC GBP £1.96M → Citi USD $2.5M > FX: GBPUSD 1.2741 · Cost: $1,200 · Net benefit: $3,000 > Execution: T+0 via RTGS · Approval: pending CFO

Intelligent Report Generation

NLP

At 09:00 daily the agent generates the liquidity report in plain English — not just numbers but a narrative explanation of what changed, why, what the risks are today, and what action it has already taken or recommends.

> DAILY BRIEF (09:00): Group position $24.5M, up $1.2M > Key driver: Paystack batch settled overnight (+$2.1M) > Watch: JP Morgan maturity today · FX hedge expiry Thu

Pattern-Based Fraud Detection

ML Model

The agent scores every treasury-level transaction against fraud typologies — structuring, round-trip flows, unusual beneficiary patterns — using ML models trained on financial crime datasets, flagging suspicious activity before funds move.

> FRAUD SIGNAL: 8 transfers · $9,800 each · 2hr window > Pattern match: Structuring (FATF Typology 14) > Action: Transactions held · Compliance notified

Natural Language Querying

Conversational

Treasury teams can ask the agent questions in plain English directly from the dashboard. It queries across all connected data sources, applies context, and responds with precise answers and the underlying data — no SQL, no BI tool required.

> Q: "What drove the NGN shortfall last Tuesday?" > A: Paystack settlement delayed 6h (system outage). Gap > was covered by auto-sweep from Stanbic at 14:12.

Perceive. Reason. Act. Learn.

Altis AI runs a continuous agentic loop — it observes every event across connected systems, builds context, reasons through the right action, and either executes autonomously or routes to a human for approval.

01

Perceive

The agent continuously ingests real-time data streams from every connected system — bank APIs, settlement queues, ledgers, market feeds, and user actions.

  • Real-time balance & transaction feeds
  • Pending settlement queue monitoring
  • Market data (FX, rates, benchmarks)
  • System event streams & webhooks
02

Reason

Multi-step reasoning chains connect observations across data sources, apply domain knowledge, model outcomes, and produce an explainable decision with a confidence score.

  • Cross-source context assembly
  • Domain-specific financial reasoning
  • Multi-path outcome modelling
  • Explainable reasoning trace
03

Act

The agent takes action — autonomously for pre-approved low-risk decisions, or by routing an approval request with full workings to the right human for high-value or novel situations.

  • Autonomous execution within defined limits
  • Structured approval requests with workings
  • Action logging with full audit trail
  • Rollback capability on executed actions
04

Learn

Every approved, rejected, or modified decision feeds back into the agent's models. It continuously recalibrates confidence thresholds, improves pattern recognition, and adapts to your organisation's evolving behaviour.

  • Human feedback loop integration
  • Confidence threshold recalibration
  • New pattern recognition over time
  • Organisation-specific model fine-tuning

Same agent. Every product. Zero setup.

Enable Altis AI on any Altisly product and it immediately understands the context, connects to the relevant data streams, and starts operating. No training data labelling. No prompt engineering. No integration work.

AI-Enabled

AI-Assisted Break Resolution

Reconciliation Suite

The agent analyses every unmatched transaction, identifies the probable cause, suggests the correct match or resolution action, and auto-resolves breaks that fall within configured tolerance bands.

Classifies break type (timing, amount, duplicate, missing)
Suggests matching counterpart with confidence score
Auto-resolves tolerance breaks within defined thresholds
Drafts resolution notes for auditor review
AI-Enabled

AML Risk Intelligence

Compliance Suite

The agent scores every transaction against AML typologies in real time, explains which rules were triggered and why, surfaces the highest-risk cases to analysts first, and drafts SAR narratives for review.

Real-time typology scoring on every transaction
Natural-language explanation of each alert
Analyst queue prioritisation by risk score
Auto-drafts SAR narratives from case evidence
AI-Enabled

Merchant Risk Scoring

Merchant Suite

The agent continuously monitors merchant transaction velocity, chargeback ratios, and behavioural patterns — flagging merchants approaching risk limits and recommending holds, enhanced monitoring, or limit reductions before breaches occur.

Continuous merchant behaviour monitoring
Chargeback ratio trend prediction
Auto-flags merchants exceeding risk thresholds
Recommends reserve adjustments proactively
AI-Enabled

Intelligent Forecasting

Liquidity Forecasting Suite

The agent generates probabilistic cash flow forecasts, explains what is driving each projection, surfaces the top risks to the 30-day position, and proactively recommends hedging or funding actions when it sees developing pressure.

Narrative explanation of forecast drivers
Proactive alert before threshold breaches
Recommends hedging actions on FX exposure
Recalibrates daily as actuals come in
AI-Enabled

Settlement Failure Triage

Settlement System Suite

When a settlement fails the agent immediately diagnoses the cause, evaluates retry eligibility, models the downstream liquidity impact, and either auto-retries within policy or routes an exception with a recommended resolution path.

Instant failure root-cause classification
Retry eligibility assessment within policy
Downstream liquidity impact modelling
Auto-retry or structured escalation routing
AI-Enabled

Cross-Product Intelligence

All Products

Because Altis AI operates across all products simultaneously, it can surface connections that no single product would see — a merchant settlement delay causing a treasury gap, or a compliance hold blocking a reconciliation.

Cross-product event correlation
Unified audit trail across all agent actions
Single natural language interface across products
Organisation-wide risk dashboard

The capabilities that make it work

Multi-Step Chain-of-Thought Reasoning

The agent builds a reasoning chain across multiple data sources before arriving at a conclusion — not pattern matching, but structured inference. Every step is logged and explainable.

Context Memory & Session Continuity

The agent retains context across interactions — it knows what decisions were made yesterday, what was approved last week, and how the organisation tends to respond to specific situations.

Explainable AI Outputs

Every action recommendation includes a full explanation of what was observed, what was inferred, and why this action was chosen — not a black box. Designed for regulated financial environments.

Configurable Autonomy Thresholds

Define exactly which action types, value ranges, and risk levels the agent can execute autonomously versus which require human approval. Full audit trail regardless of mode.

Natural Language Interface

Ask the agent questions in plain English across any connected product. It queries live data, applies domain context, and returns precise answers — no SQL, no dashboards required.

Real-Time Continuous Learning

Every human override, approval, or correction teaches the agent. Confidence thresholds auto-calibrate over time, and new patterns in your data are incorporated into the models continuously.

You set how much the agent operates on its own

Altis AI ships with three configurable autonomy levels. Set them globally, per product, per action type, or per value threshold. The agent never exceeds the trust level you define.

Advisory Mode

Safest

The agent observes, analyses, and recommends — but takes no action without explicit human approval on every decision. Full reasoning shown for every suggestion.

Supervised Mode

Recommended

Low-risk, pre-approved actions execute automatically. High-value or novel decisions route to a human with full workings. Thresholds are fully configurable per action type.

Autonomous Mode

Maximum Speed

The agent executes all decisions within its configured scope without human approval, logging every action in real time. Used for well-understood, time-critical operational tasks.

Altis AI — Approval Request
AGENT ACTION REQUEST

I have identified a liquidity gap of $2,400,000 in your Citibank USD account expected at 14:30 today.

CauseDelayed SWIFT settlement from JP Morgan (Ref: TXN-9182)
RiskIntraday overdraft · est. cost $4,200 in bank fees
ActionInitiate sweep of $2.5M from Standard Chartered GBP → Citibank USD
FX rateGBPUSD 1.2741 · est. cost $1,200 in conversion
Confidence97%

Deploy Altis AI the way that fits your stack

Embedded in Product

Enable Altis AI with a single toggle inside any Altisly product. Immediate activation — no integration work, no data pipelines to configure.

// Dashboard → Settings → Altis AI // Toggle: Enable AI Agent // Autonomy: Supervised // ✓ Active in 30 seconds

REST API

Call the Altis AI agent directly from your own systems. Pass any financial event and receive a structured decision with reasoning, confidence score, and recommended action.

POST /v1/agent/evaluate { "context": "treasury", "event": { ...transaction }, "autonomy": "supervised" }

Webhook Triggers

Subscribe the agent to any event stream. When a defined trigger fires — balance threshold, failed settlement, new alert — the agent activates and responds in under 500ms.

altis.on("balance.anomaly", async (e) => { const action = await agent.evaluate(e) await action.execute() // or route })

Deploy the intelligence layer across your products

Talk to our team about activating Altis AI on your Altisly product stack.