B2B · Enterprise Technology Since 2009
Fifteen years of building systems
that organisations depend on.
Coadjutant began with websites, grew through mobile and ERP, matured through data engineering and analytics, and now operates at the frontier of enterprise AI — reconciliation engines, intelligent audit systems, AI agents, and executive intelligence. Every era of work is still in production somewhere.
15+
Years in enterprise technology
Web · Mobile · ERP · Data · AI
9
Active AI deployments
Africa · UK · USA · South Asia
94%
Audit anomaly precision
Smart audit deployments
18M
Records digitised
Government archive pipelines
Our journey
We grew with the technology,
one hard problem at a time
Coadjutant was not founded as an AI company. It was founded to solve real operational problems for businesses — and the tools available to solve those problems have changed dramatically over fifteen years. Here is how that journey looked.
2009–
2013
2013
Foundation
Websites, portals, and digital presence for businesses that had none
Our earliest work was in web development — building corporate sites, e-commerce portals, intranet systems,
and CMS-driven publishing platforms for SMEs and mid-market businesses. We worked with clients in retail,
education, healthcare, and professional services. What distinguished this period was not the technology but
the discipline: every engagement was scoped, documented, and delivered against a defined brief. We built a
reputation for finishing what we started.
2013–
2017
2017
Mobile & ERP
Mobile applications and enterprise resource planning for operations-heavy businesses
As mobile became the primary interaction layer, clients needed applications that connected field teams,
customers, and back-office systems in real time. We built native and cross-platform mobile apps for logistics,
field service, retail, and finance. Simultaneously, we entered the ERP space — implementing and customising
systems for manufacturing, distribution, and multi-branch service businesses. This period taught us how
enterprises actually work: the gap between how processes are documented and how they are actually performed.
2017–
2021
2021
Data Engineering
Data pipelines, warehousing, and analytics for organisations sitting on unstructured operational data
By 2017, many of our clients had years of transaction history across disconnected systems with no way to
interrogate it. We built data engineering practices to address this: ETL pipelines, data warehouses, reporting
layers, and BI dashboards that gave operations and finance teams visibility they had never had. This period
was formative — it revealed consistently that data quality was the limiting factor in every business decision.
It also placed us directly upstream of where AI would eventually sit. In 2019, we formally incorporated as a
limited company registered with ROC Kolkata.
2021–
Now
Now
AI & Agents
Enterprise AI — reconciliation engines, intelligent audit, credit intelligence, agent systems, and executive AI
The shift to AI was not a pivot — it was a natural progression. Our data engineering clients had clean
pipelines and wanted intelligence on top of them. Our ERP clients had process documentation and wanted
automation inside it. Our banking clients had reconciliation backlogs and wanted systems that could close them
without growing headcount. We now operate across the full AI production lifecycle: model development, agent
frameworks, LLM fine-tuning, ethics review, and ongoing reliability management. Clients in East Africa,
West Africa, the UK, North America, and South Asia.
Delivered work — all eras
Case studies across fifteen years
of production systems
Problems, approaches, and outcomes — documented as built. Every case study below is a live or successfully completed production deployment. Client identities are kept confidential across all engagements, by policy.
2009–2013 · Web & Digital
Web Platform
2011South Asia
E-commerce · Retail
A retail group operating across multiple cities needed a unified online storefront connecting
inventory across branches, supporting regional pricing, and handling order fulfilment routing to the nearest
warehouse. No existing off-the-shelf solution handled multi-branch inventory at the price point available.
What Coadjutant built
A custom e-commerce platform with branch-aware inventory management, postcode-to-warehouse routing
logic, regional pricing engine, and an admin dashboard giving central operations visibility across all branches.
Integrated with the client's existing billing system via custom API.
PHP / MySQLCustom inventory engineBilling API integrationMulti-branch admin
Outcomes
3×
Order processing capacity vs manual phone orders
Zero
Cross-branch inventory errors post-deployment
Web Portal
2012East Africa
Healthcare · Appointments
A private healthcare group managing several clinics had no digital system for patient registration,
appointment booking, or medical history access. Receptionists were managing paper registers and phone bookings,
creating scheduling conflicts and lost records.
What Coadjutant built
A web-based patient portal with online appointment booking, digital registration forms, and a
clinic-side scheduling dashboard. Doctor availability management, SMS appointment reminders, and a basic medical
history view for clinical staff. Designed for low-bandwidth environments with graceful degradation.
PHP / LaravelSMS gateway integrationLow-bandwidth designScheduling engine
Outcomes
68%
Reduction in scheduling conflicts and double bookings
4 clinics
Unified on one system within 8 weeks of go-live
2013–2017 · Mobile & ERP
Mobile App
2015West Africa
Logistics · Field Operations
A logistics company running last-mile delivery across a major urban region was managing its driver
network entirely by phone. Dispatch had no real-time visibility of delivery status, drivers had no structured
way to record delivery proof, and customer notifications were manual. The operation was scaling faster than the
coordination method could support.
What Coadjutant built
An Android application for delivery drivers with offline-first architecture — critical for areas
with intermittent connectivity. Features included route assignment, barcode scanning for delivery proof, photo
capture, GPS status sync on reconnection, and automated customer SMS triggers. A dispatcher web dashboard
provided real-time fleet view with exception flagging.
Android nativeOffline-first SQLiteBarcode scanningGPS syncDispatcher dashboardSMS automation
Outcomes
91%
Delivery confirmation rate (digital proof of delivery)
40%
Reduction in customer service calls on delivery status
ERP
2016South Asia
Manufacturing · Operations
A mid-size manufacturer was operating procurement, production planning, inventory, and finance across
four separate software tools with no integration. Month-end reconciliation took over a week. The sales team had
no visibility of production status when promising delivery dates, resulting in chronic over-commitment.
What Coadjutant built
Full Odoo ERP implementation covering procurement-to-pay, production planning, inventory management,
and financial accounting. Custom modules built for the client's specific bill-of-materials structure and
multi-site inventory logic. Data migration from four legacy systems with full historical transaction import.
Staff training programme across three departments.
Odoo ERPCustom BOM moduleMulti-site inventoryLegacy data migrationMRP configuration
Outcomes
9 days → 1
Month-end reconciliation duration
Single system
Replacing four disconnected tools across all departments
Mobile App
2017United Kingdom
Financial Services · Client Ops
A wealth management firm was communicating with clients entirely through PDF reports emailed
quarterly and ad hoc phone calls. Clients were increasingly requesting self-service access to portfolio data
and document signing capabilities. The firm had no mobile presence and no secure digital document workflow.
What Coadjutant built
iOS and Android apps providing clients with a read-only portfolio dashboard, secure document inbox,
and in-app e-signature workflow for standard agreements. Built with end-to-end encryption, biometric
authentication, and a backend that connected to the firm's existing portfolio management system via a secure
API layer Coadjutant designed and documented.
React NativeE2E encryptionBiometric authE-signature workflowPortfolio API layer
Outcomes
78%
Client document signing via app within 6 months of launch
Zero
Security incidents post-launch through full client lifecycle
2017–2021 · Data Engineering & Analytics
Data Engineering
2019East Africa
Banking · Reporting
A banking group with operations across three countries had transaction data distributed across six
core banking instances, two loan management systems, and a treasury platform. Monthly management reporting took
three weeks of analyst time to produce, involved significant manual spreadsheet work, and was frequently
inconsistent between departments using nominally the same numbers.
What Coadjutant built
An end-to-end data warehouse architecture: extraction connectors for each source system, a
transformation layer standardising entity definitions and currency handling across jurisdictions, a dimensional
model built for financial reporting, and a Power BI reporting layer with role-based access by country and
business unit. Automated daily refresh with data quality alerts.
Python ETLApache AirflowPostgreSQL warehousedbt transformationsPower BIData quality monitoring
Outcomes
3 weeks → 4 hrs
Monthly management report generation time
Single version
Of truth across all three country operations
Analytics
2020North America
Retail · Customer Intelligence
A retail business operating across physical stores and an e-commerce platform had no unified view
of customer behaviour. Online and in-store purchase data lived in separate systems, marketing campaigns were
not measurable end-to-end, and the merchandising team was making ranging decisions based on intuition rather
than demand data.
What Coadjutant built
A customer data platform connecting POS, e-commerce, loyalty, and email marketing systems into a
unified customer record. Cohort analysis, basket affinity models, churn prediction scoring, and a self-service
analytics interface for the merchandising and marketing teams. Weekly automated insight summaries delivered to
the leadership team.
Customer data platformIdentity stitchingBasket affinity modelChurn predictionLooker analyticsBigQuery
Outcomes
360°
Unified customer view across all channels for first time
22%
Improvement in marketing campaign conversion rate
2021–Present · Enterprise AI
Banking · AI
2022East Africa
Reconciliation · Settlement
A commercial bank processing high daily transaction volumes across four core banking systems was
experiencing a persistent mismatch rate requiring large analyst teams to resolve manually each day. The legacy
rules-based tool could not learn from past resolutions and generated high volumes of false positives that
consumed analyst time without producing fixes.
What Coadjutant built
A four-layer pipeline: format normalisation across heterogeneous CBS schemas → ML classification
of mismatch typology across 26 defined categories → root-cause isolation → automated correction entry
generation with human-override at each threshold. Trained on seven years of historical resolution data and
retrained weekly on live patterns. Full audit trail on every automated action.
Python reconcile engineXGBoost classifierEvent-driven architectureSWIFT MT940 parsingHuman-in-the-loop UIAudit trail generation
Outcomes
2.1% → 0.09%
Daily mismatch rate post-deployment
97.4%
Transactions auto-resolved without analyst intervention
82%
Reduction in analyst hours on reconciliation
Microfinance · AI
2022West Africa
Fraud Detection · Origination
A microfinance institution was experiencing material portfolio losses from ghost borrower fraud and
loan stacking. The fraud was invisible to individual application review because patterns only emerged at network
level — across borrowers, addresses, phone clusters, and identity documents shared across multiple applicants.
What Coadjutant built
A graph neural network model resolving hidden entity relationships across seven data sources —
national identity, phone networks, address clusters, bureau records, and repayment history. Fraud signals scored
at origination in under 400ms. Designed with a low false-positive rate to minimise friction for legitimate
borrowers while blocking fraud rings at first application attempt.
Graph neural networkIdentity resolution engineNational ID integrationReal-time scoring APIExplainability layer
Outcomes
71%
Reduction in fraud-related portfolio losses
0.4%
False positive rate — minimal customer friction
9 in 10
Fraud attempts caught at origination stage
Insurance · AI
2023United Kingdom
Smart Audit · Compliance
A regulated insurance firm needed to evidence fair customer outcomes at individual transaction level
under new FCA guidance. Existing sampling-based manual audit covered less than 1% of transactions — inadequate
for regulatory purposes and operationally impossible to scale manually to full coverage.
What Coadjutant built
Continuous monitoring pipeline ingesting claims, pricing decisions, and customer communications.
NLP-based outcome classifier flagging transactions in real time against a regulatory outcome taxonomy.
Auto-generated evidence packs structured to FCA submission requirements, eliminating the manual collation
process and dramatically reducing reporting preparation time.
NLP outcome classifierRegulatory taxonomy mappingEvidence pack generationAudit trail pipelineFCA reporting format
Outcomes
100%
Transaction audit coverage — from sub-1% sampling
94%
Anomaly detection precision
3 days
Regulatory evidence pack (was 6 weeks manual)
Government · AI
2023South Asia
Digitisation · Records
A public authority held decades of land, procurement, and contract records exclusively in physical
form across multiple regional archives. Manual retrieval for legal and audit purposes took weeks. A statutory
digitisation mandate required fully searchable, structured digital output — with multilingual document handling
across Hindi, Marathi, and English records of varying quality.
What Coadjutant built
End-to-end digitisation pipeline: scan intake → page orientation correction → OCR fine-tuned on
Hindi and Marathi official document formats → named entity extraction → document classification → structured
ingest into a government-approved document management system. Human verification queues triggered automatically
for items scoring below confidence threshold.
Custom multilingual OCRHindi / Marathi NLPEntity extractionDocument classificationISO 15489 compliantHuman verification queue
Outcomes
18M
Records processed and fully structured
98.7%
OCR accuracy across multilingual document types
~6 hours
Average retrieval time — was 4 to 22 working days
Fintech · AI
2023North America
Credit Intelligence · Decisioning
An alternative lending platform was operating on a legacy scorecard with a high manual-review rate
and approval cycles measured in days. The scorecard was incorrectly rejecting a meaningful segment of
creditworthy applicants — the feature set did not reflect alternative repayment signals available through open
banking, and there was no mechanism to learn from resolution patterns over time.
What Coadjutant built
An ensemble model incorporating 140 features drawn from bank feed, open banking, bureau, and
behavioural signals. Monthly automated retraining pipeline with drift monitoring. SHAP-based explainability
layer generating decision rationale for every application — required for fair lending documentation, adverse
action notices, and regulatory examination readiness.
Gradient boosting ensembleOpen banking integrationSHAP explainabilityAutomated retrainingFair lending audit layerAdverse action generator
Outcomes
6×
Faster end-to-end credit decisioning cycle
−23pp
Reduction in manual review rate
Compliant
Fair lending and adverse action documentation audit-ready
Executive AI
2024North America
AI Agents · CFO Suite
A finance leadership team was spending several working days each week aggregating data from ERP,
CRM, treasury, and market data sources into a board reporting pack. The process was manual, error-prone, and
produced a document that was often days stale by the time it reached the board. Strategic decisions were being
made on data with no clear freshness guarantee.
What Coadjutant built
A seven-agent orchestration system with domain-specialist agents covering treasury, revenue, cost,
market signals, risk, compliance, and narrative synthesis. Each agent maintains domain memory and flags material
changes with confidence scores. A synthesis agent assembles variance commentary and forward projections. Output
delivered as a structured executive briefing on a defined automated schedule with full source traceability.
Multi-agent orchestrationERP / CRM connectorsVariance analysis engineNarrative synthesis LLMSource traceabilityStructured report delivery
Outcomes
~4 minutes
Board pack generation — was 3+ working days
Real-time
Data freshness at point of distribution
22 hrs
Weekly analyst capacity reclaimed for higher-value work
What we build today
Current capabilities — built on fifteen
years of production experience
Our current AI practice stands on the engineering discipline developed across four eras of delivered work. These are not theoretical offerings — each capability has live deployments behind it.
AI Reconciliation & Fix Models
Reconcile-and-fix pipelines for interbank, intra-system, and multi-ledger environments. Mismatch classification, root-cause isolation, and automated correction entry generation with human-in-the-loop at every defined threshold.
AI Agent Development
Multi-agent orchestration for back-office automation, document processing, and executive intelligence systems. Agents that operate within defined boundaries, escalate correctly, and produce auditable outputs.
Smart System Audit
Continuous monitoring of transaction logs, system behaviour, and access events. Anomaly scoring, audit trail generation, and regulatory evidence packaging — automated end-to-end, 100% coverage.
AI in Finance & Credit
Credit risk modelling, fraud detection, and liquidity forecasting across retail banking, microfinance, and wholesale credit. Explainable outputs built for regulatory scrutiny from the first sprint.
Digitisation of Records
Conversion of physical archives into structured, searchable knowledge bases. Custom OCR, multilingual NLP, entity extraction, and classification — built for government-grade accuracy and compliance standards.
Quality LLM Outsourcing
Fine-tuned, domain-adapted language model services for regulated industries. Model selection, alignment, evaluation, and ongoing reliability management — so internal teams focus on product, not infrastructure.
AI Ethics & Reliability
Independent ethics review, bias auditing, model risk assessment, and explainability frameworks. Embedded at design, not applied retrospectively. Our ethics practice is independent of our delivery practice.
Data Engineering & Analytics
We still build the data foundations that AI requires — ETL pipelines, warehouses, and analytics layers. Every AI engagement begins with an honest assessment of whether the data is ready for it.
Predictive to Executive Intelligence
Moving organisations from dashboards that report history to AI systems that advise on strategy — synthesising internal data, market signals, and risk indicators into decision-ready executive briefings.
AI Ethics & Reliability
We build AI that can be questioned.
Enterprise AI without accountability infrastructure is a liability. Coadjutant embeds ethics, explainability, and reliability principles as structural requirements — not post-launch compliance activities.
Our ethics practice is independent of our delivery practice. We will decline to ship a system that fails our reliability standards, and we document that position in every engagement contract before work begins.
Explainability by design
Every credit, risk, and audit model produces human-readable decision rationale. No black-box outputs in regulated contexts.
Bias audit standard
Pre-deployment bias testing across protected characteristics with documented disparate impact analysis for every scoring model.
Human-in-the-loop protocols
Defined escalation thresholds at which autonomous decisions route to human review. Logged, timestamped, auditable at every step.
Model risk governance
SR 11-7 model risk management guidance applied as the baseline standard across all client deployments, regardless of jurisdiction.
Drift & reliability monitoring
All production models include automated drift detection with alert thresholds and documented retraining triggers. No silent degradation.
Data sovereignty
Client data stays in agreed jurisdictions. In-country deployment supported for African and South Asian regulatory requirements.
How we engage
Our five-phase delivery process
We do not begin development until a diagnostic is complete. Building the wrong system faster is not progress — and we will say that directly at the first meeting.
01
Technical diagnostic
2–3 week assessment of data quality, system architecture, and operational processes. Deliverable: a frank report — including whether AI is the right solution.
02
Problem definition
The problem defined in measurable terms: what changes, by how much, and how we will verify it. Signed by both parties before model work begins.
03
Prototype & test
Working prototype on real data within 6–8 weeks. Human-readable outputs from day one. No black-box demonstrations at any stage.
04
Production deployment
Staged rollout with shadow-mode validation. Systems go live only when performance matches prototype benchmark within agreed tolerances.
05
Ongoing reliability
Continuous monitoring, quarterly model reviews, and documented retraining schedules. SLA-backed uptime and anomaly response commitments.
Start here
Ready to discuss a specific problem?
Our engagements begin with a 60-minute technical conversation — no sales deck, no discovery form. Bring your data environment, your business problem, and your honest view of what has already been tried.
Africa
Nairobi · Lagos
Europe
London
North America
New York
South Asia
Mumbai · Delhi