Skip to main content
Welcome to Deloitte

If we have selected the wrong experience for you, please change it above.

Navigating the AI jungle: Which deployment model fits your finance function?

A practical guide to choosing between ERP-embedded, bolt-on, and standalone AI solutions for Procure-to-Pay, Order-to-Cash, and Record-to-Report.

Key takeaways

  • There are three distinct AI deployment models: ERP-embedded, bolt-on, and standalone.
  • Each deployment model has specific characteristics that can benefit your organisation.
  • A hybrid deployment model is usually the most credible enterprise answer across P2P, O2C, and R2R.

This point of view offers a practical lens for decision-making for your AI system landscape. Helping organisations define where to begin, which tools to trust, and how to align deployment choices with their existing Enterprise Resource Planning (ERP) landscape.


AI deployment models for finance

Finance leaders have moved past whether to deploy AI. The real challenge is now deploying it without creating fragmentation, control risk, or adding another layer of disconnected tools. This PoV cuts through the noise by mapping AI opportunities to your actual process landscape and showing you which deployment model fits.

Three core transactional processes drive AI value

Many organisations don’t follow a structured approach to AI deployment. A more effective approach is to identify and map AI opportunities directly to the process landscape, which finance teams are already familiar with.

The finance process taxonomy serves as a practical starting point and foundation, providing a common language across business and technology teams, making it easier to identify where AI can reduce manual effort, strengthen control, improve data quality, accelerate cycle times and identify opportunities.

Within the scope of transactional finance the following three end-to-end processes matter most because of their high transaction volumes, zero to low tolerance for errors and therefore high potential to improve efficiency:

Figure 1: Finance Process Taxonomy of Transactional Finance

From supplier onboarding and procurement execution through invoice handling, payment, and close.

  • Vendor Master Data Management: Creation and maintenance of accurate supplier records to support purchasing and payment activities.
  • Strategic Sourcing: Identifying and selecting suppliers to achieve the best balance of cost, quality, and business value.
  • Procurement: Managing the purchase of goods and services from requisition through to purchase order issuance.
  • Receiving: Recording and validating that ordered goods or services have been delivered as expected.
  • Invoice Processing: Verifying and posting supplier invoices by matching them with purchase orders and receipts.
  • Payments: Executing approved payments to suppliers accurately and on agreed terms.
  • Closing: Finalising and reconciling procurement and accounts payable activities to ensure accurate financial records for the period.

From customer creation and pricing through billing, collection, cash application, and dispute resolution.

  • Customer Master Data Management: Creation and maintenance of accurate customer records to support sales and collections.
  • Pricing & Discount Management: Defining and applying product prices, discounts, and commercial terms for customer transactions.
  • Credit Management: Assessing and monitoring customer credit risk.
  • Billing: Generating and issuing accurate invoices for delivered goods or services.
  • Collection and Cash Application: Collecting customer payments and matching them to the correct outstanding invoices.
  • Dispute Management: Resolving billing or payment issues to ensure timely cash collection and customer satisfaction.
  • Closing: Reconciling receivables and sales activities and preparing period-end financial and operational reports.

From transaction processing and reconciliation through period close and consolidation or reporting.

  • Finance Master Data Management: Managing core financial data, such as charts of accounts and organisational structures, to support accounting processes.
  • Recording: Recording and posting financial transactions accurately in the accounting system.
  • Reconciliation: Comparing and validating account balances to ensure financial data is complete and accurate.
  • Closing and Consolidate: Completing all accounting activities required to finalise the financial results for a reporting period.
  • Reporting: Producing financial statements and management reports to support business decisions and compliance.

Detail level three (L3) from the standard finance process taxonomy offers sufficient granularity to assess potential AI solutions for these processes, distinguishing use cases which:

  • are primarily workflow automation opportunities,
  • rely on prediction or anomaly detection,
  • require data from multiple systems and documents while applying finance and control rules.

Both tool providers and companies work with this or similar finance taxonomies and it makes it possible to identify AI use cases and map available AI capabilities to your processes.

Three technology deployment models

A useful way to navigate the market for AI solutions is to approach it through three deployment models, rather than comparing products in isolation.

  1. ERP-embedded AI: AI capabilities delivered natively within the ERP platform.
  2. Bolt-on AI solutions: External platforms focused on specific finance domains (e.g., invoice automation, cash application, collections, close orchestration) integrated through APIs and connectors or even ERP native.
  3. Standalone AI: Custom AI solutions built on enterprise AI platforms or completely self-developed.

These three deployment models each have their own key benefits and limitations: 

Deployment model

ERP-embedded AI

Fits best

When the majority of your process is covered by your ERP. 

Key benefits

- Fast time-to-value
- Low integration complexity
- Governance is built in

Key limitations

Capability depth may lag specialist solutions in narrower finance use cases. Innovation is dependent on the ERP vendor roadmap.

Deployment model

Bolt-on AI solution

Fits best

When the degree of efficiency gains are higher compared to ERP-embedded AI or your process includes important specifics which are not covered by ERP.

Key benefits

- Specialised domain expertise
- Faster deployment
- Specialised functionalities

Key limitations

Adds integration and governance complexity.

Deployment model

Standalone AI

Fits best

When the process is genuinely distinctive and a source of competitive advantage.

Key benefits

- Maximum degree of customisation
- Unique competitive edge

Key limitations

Requires maintenance and governance procedures. Typically higher effort to keep the AI model and solution up to date. Integration and governance effort accumulate.
ERP-embedded AI: The natural starting point

For most finance organisations, ERP-embedded AI is the natural starting point. These solutions offer direct access to transactional data, native workflow context, and an established control environment, reducing the friction involved in implementation and enabling quick wins.

Embedding AI does not mean that its improvement potential is fully exploited. ERP vendors typically excel at workflow assistance and productivity use cases, yet fall short on advanced exception management, agentic orchestration, or specialised industry logic that often sits outside the core platform.

The critical question: Is ERP-embedded AI sufficiently competitive and suitable for your level of process complexity?

The following figures illustrate representative AI use cases across transactional Level 3 finance processes, demonstrating where ERP-embedded AI can automate, augment, or support day-to-day activities. Many AI functionalities are not included in the standard licence fee; add-ons and premium AI licences may be required, depending on the functionality and the provider.

Figure 2: Overview of available ERP-embedded AI solutions
Note: The AI solution landscape is changing rapidly. This overview is based on expert judgement in June 2026.

Background and challenge
Thousands of invoices arrived daily from vendors, each triggering a cascade of manual activities. Teams manually matched invoices against purchase orders and goods receipts, investigated discrepancies, and scheduled payments - a fragmented workflow that consumed most of their capacity. The result was frequent payment delays that strained supplier relationships and created unnecessary friction in the supply chain.

Scope
The client deployed an intelligent invoice recognition and matching engine embedded directly within their ERP environment. Advanced natural language processing and computer vision extracted invoice data with 99% accuracy, eliminating the need for manual data entry and reducing transcription errors. A probabilistic matching logic then automatically reconciled invoices against purchase orders and goods receipts, flagging only genuine exceptions for human review. Beyond matching, the system applied intelligent payment scheduling by analysing cash flow forecasts, supplier payment terms, and early payment discount opportunities, recommending optimal payment dates that balance supplier relationships with working capital efficiency.

Benefits
The impact was transformative. The manual effort for invoice matching and exception handling fell by 45%, while the internal cost per processed invoice fell by 35%. More importantly, vendor disputes plummeted by 63% as payments became faster and more accurate, strengthening supplier relationships. The optimised payment schedules enhanced working capital management and improved cash flow visibility. The team shifted from reactive exception handling to strategic supplier management, while the ERP system became a true operational partner rather than a passive record-keeper.

Bolt-on AI solutions: Specialised depth where ERP falls short

Bolt-on specific solutions matter when finance requires more than ERP capabilities can provide. In practice, this is seen most often in areas such as advanced invoice processing, complex cash application, predictive collections, and close orchestration. Here specialist vendors often outdo ERP capabilities with more developed workflow logic and greater operational depth.

These solutions can create value quickly, particularly when the business problem is specific, measurable, and with clear pain points to solve. This means they are often the right answer when a finance team needs specialised capabilities in a process area not covered by ERP functionalities.

The trade-off is architectural complexity. Every additional solution adds integration points, master-data dependencies, governance requirements, and another vendor relationship. Specialist tools are most effective when they solve a tailored process situation. When using bolt-on AI solutions it is important to avoid adding another layer of overlapping functionality.

Figure 3: Overview of selected Bolt-on AI solutions

Note: The AI solution landscape is changing rapidly. This overview is based on expert judgement in June 2026. The list above is not exhaustive. There are many more providers, with bolt-on solutions available that bring the benefits of AI to very specific processes.

Background and challenge
The client faced a critical operational bottleneck: their collections team was drowning in manual work, laboriously sorting through thousands of customer accounts to identify collection priorities based on fragmented payment histories and outstanding balances. This inefficiency drove up days sales outstanding, created blind spots in process visibility, and exposed the organisation to significant operational and credit risks.

Scope
The client deployed a sophisticated bolt-on AI solution designed to inject intelligence and automation into every stage of the collections workflow. The system fundamentally reimagined how the organisation prioritised accounts, standardised outreach protocols, and orchestrated collections activities – all while providing real-time visibility into both process metrics and individual collector performance.

Benefits
The solution unlocked 6.5 million CHF in measurable value, driven by three distinct value streams. Working capital optimisation generated 3.1 million CHF in interest cost savings through accelerated collections and reduced cash conversion cycles. Automation and productivity gains delivered 1.7 million CHF in operational efficiencies by eliminating repetitive manual tasks and freeing the collections team to concentrate on high-touch, relationship-critical interactions. Strategic improvements in collection practices yielded an additional 1.7 million CHF through enhanced bad debt recovery. The solution achieved 55% incremental automation beyond existing dunning capabilities, fundamentally reshaping how the organisation manages its collections function and positioning the team as a strategic driver of working capital performance.

Standalone AI solutions: Building for competitive advantage

Standalone AI solutions are attractive when three conditions align: distinctive process logic that other vendors don't support; data ecosystems extending beyond standard ERP boundaries; and control requirements that are too specific for packaged software. Among the strongest use cases are those involving proprietary matching algorithms, complex claim exception handling across legacy systems, or unique policy interpretations that off-the-shelf products cannot accommodate.

However, custom AI is not a default starting point. Organisations must treat it as a sustained product capability – not a one-off experiment – with dedicated governance, model monitoring, and clear business accountability. This means a shift from a project to a product, and to a sustained service for the business.

When these conditions are met, the payoff is considerable. Custom solutions deliver greater flexibility over data architecture and model design, reduce third-party dependency, and unlock strategic benefits. The upfront investment in talent and infrastructure is higher compared to the other deployment models, but ROI compounds over time – especially for large, high-value use cases where long-term ownership and flexibility outweigh the cost.

Background and challenges
The client maintained a team of 300+ FTE dedicated to manually fact-checking transaction documentation. Evidence was scattered across fragmented systems – emails, PDFs, screenshots, SharePoint folders – forcing the team to hunt through files to validate each transaction.

Scope
The client deployed an intelligent automation system that automatically parses all documentation sources and extracts relevant evidence through a rule-based compiler, creating a structured, versioned data hierarchy with full audit trails. A team of AI agents then executed deterministic business logic and control policies, running up to 15 automated structured tests against the evidence. The solution empowered the team through a generative AI chatbot that lets analysts interact conversationally with the agentic team and explore exceptions in real time, keeping humans firmly in control.

Benefits
The solution resulted in a 50% reduction in the manual fact-checking workload, and improved quality through consistent automated testing, faster exception resolution, and full traceability for compliance – all delivered by a leaner team that now focuses on judgment calls and complex cases rather than routine data validation.

Your AI deployment decision kit

Defining an AI strategy is the starting point for the finance AI journey, but strategy alone is not enough. What matters most is translating your strategic intent into a focused set of use cases and selecting the deployment model that fits best. The objective is to adopt AI where it can deliver measurable value in productivity, control, cycle time, and decision quality.
Step 1: Define your AI ambition

Which outcomes matter most? (productivity, control, cycle time, decision quality). What’s your organisational buy in?

Step 2: Identify and prioritise use cases

Map to your process taxonomy (P2P, O2C, R2R). Assess: business value, process readiness, data readiness, implementation feasibility.

Step 3: Choose your deployment model

Assess process complexity, cost and speed of implementation to choose your deployment model.

The deployment decision should be made on a use case by use case basis, guided by three key deployment model factors:

  • Process complexity: The degree of standardisation and specialisation of your process determines the deployment model. Standardised, rule-based processes with few exceptions favour ERP-embedded AI. Highly specialised, proprietary processes with custom requirements may require standalone AI.
  • Cost: Your investment appetite for a specific solution is the second key factor. ERP-embedded AI tends to have the lowest total cost. Bolt-on solutions fall in the middle, and standalone AI involves the highest level of investment. 
  • Speed of implementation: The desired time-to-value is the third factor. ERP-embedded AI tends to deliver the fastest time-to-value. Bolt-on solutions require some time for integration efforts. Standalone AI takes longest because of the design and build complexity.

For most finance organisations the answer will not be a single model but rather a combination of solutions. It makes sense to: use ERP-native capabilities where processes are standardised and speed matters; deploy bolt-on AI solutions where capability depth creates clear value; and build selectively where differentiation, control, or legacy complexity justify the additional effort. Do not force every use case into one model, but make the trade-offs explicit and allow a combination of deployment models.

Leading companies use detailed decision criteria for each of these levers to ensure a decision for an AI tool is in line with strategic targets and delivers a clear ROI.

Figure 4: AI decision considerations

Navigate through below illustration, three simple questions covering key factors will provide initial guidance on which deployment model to choose for your AI use case.

Skip to description

Figure 5: Factors that influence the choice of the AI deployment model

The bottom line: No single path works for everyone

AI in transactional finance is moving from the experimentation phase to being part of the operating model. The question for finance leaders is no longer whether suitable AI capabilities and improvements are available, but which solution to use, and where.

Organisations that create the most value treat AI as a process transformation lever, not a disconnected tool. They prioritise high-value use cases, align deployment choices to process realities and the ERP landscape, and build governance from the start.

“One size fits all” doesn’t work. There is no single deployment path that fits every process and every organisation. ERP-embedded AI delivers rapid time-to-value for standardised workflows. Bolt-on AI solutions offer depth in specialised areas where core systems fall short. Standalone AI creates a durable competitive advantage when process uniqueness demands it. Most organisations will resort to all three deployment models: knowing where to apply which specific model is key to harnessing the full potential of AI.

 

This article has been made possible with the contribution of: Maria Kanidou, Lucas Landolf, Maria Stumpf

Get in touch

Markus Zorn

Markus Zorn

Finance Transformation Partner
+41 58 279 69 43
Murielle Nobleaux

Murielle Nobleaux

Director, Finance Transformation
+41 58 279 8827

Ready to transform your transactional finance processes with AI?

The Finance function’s transformation is no longer optional – it’s a strategic imperative. Organisations that leverage AI capabilities effectively will outpace competitors.

Deloitte combines deep functional, industry and technology expertise to realize sustained value from your AI investment. Partner with us to redesign andautomate your transactional finance process transformation journey.

Let’s discuss how you best use AI capabilities to increase quality and efficiency in your organisation’s unique context. Schedule a 30-minute session with our Finance Transformation team.

Did you find this useful?

Yes
No

Thanks for your feedback

Your feedback is important to us

To tell us what you think, please update your settings to accept analytics and performance cookies.

Need help updating cookies?

Let's connect

Deloitte AG is a shareholder in Deloitte EMEA BV (EMEA), which is a member firm of Deloitte Touche Tohmatsu Limited (DTTL). Deloitte EMEA and DTTL do not provide services to clients. Services may be provided by the EMEA shareholders or their affiliates, which are separate and independent legal entities. Please see www.deloitte.com/ch/about to learn more about our global network of member firms.

This communication contains general information only, and none of Deloitte EMEA (EMEA), its shareholders and affiliates is, by means of this communication, rendering professional advice or services. Before making any decision or taking any action that may affect your finances or your business, you should consult a qualified professional adviser.

No representations, warranties, or undertakings (express or implied) are given as to the accuracy or completeness of the information in this communication, and none of EMEA, the EMEA shareholders, their respective affiliates, employees or agents shall be liable or responsible for any loss or damage whatsoever arising directly or indirectly in connection with any person relying on this communication. EMEA, the EMEA shareholders and their respective affiliates are legally separate and independent entities.

Deloitte AG is an audit firm recognised and supervised by the Federal Audit Oversight Authority (FAOA) and the Swiss Financial Market Supervisory Authority (FINMA).

© 2026  For information, contact Deloitte AG.