Comparing AI Integration Approaches — Bolt-On Agents vs Middleware Layers vs Full Workflow Redesign

The advent of artificial intelligence has irrevocably altered the landscape of business operations, presenting both unparalleled opportunities and complex challenges for organizations striving to maintain a competitive edge. The strategic imperative is no longer a question of if to adopt AI, but rather how to integrate AI into existing business workflows effectively, efficiently, and with minimal disruption. This foundational decision point often bifurcates into three primary architectural philosophies: the agile deployment of bolt-on AI agents, the robust orchestration offered by middleware layers, or the more ambitious, transformative undertaking of a full workflow redesign. Each approach carries distinct implications for cost, complexity, speed of deployment, and long-term scalability, demanding a nuanced understanding from business leaders and technology strategists alike. The selection of the optimal integration pathway hinges on a comprehensive evaluation of current infrastructure, operational maturity, risk appetite, and the specific strategic objectives AI is intended to serve within the enterprise.

Understanding Bolt-On AI Agents for Rapid Augmentation

Bolt-on AI agents represent a pragmatic and often expedited approach to infusing intelligence into specific, well-defined segments of an existing business process. These agents are typically designed to perform a singular, specialized task, such as data extraction from unstructured documents, preliminary customer service triage, sentiment analysis on customer feedback, or automated lead qualification. Their primary characteristic is their ability to operate independently or with minimal interaction with core legacy systems, often connecting via APIs to ingest data, perform their designated function, and then output results back into another system or a human operator’s queue. This method is particularly appealing for organizations seeking quick wins, proof-of-concept deployments, or targeted improvements without overhauling substantial portions of their IT infrastructure. The agility of bolt-on agents allows for iterative development and deployment, enabling businesses to test AI's efficacy in specific contexts before committing to broader, more complex integrations. This approach minimizes immediate disruption and capital expenditure, making it an attractive entry point for companies exploring the practical applications of AI in their daily operations. However, the cumulative management of numerous disparate agents can introduce complexity over time, potentially leading to a fragmented AI landscape if not governed effectively.

The Role of Middleware Layers in AI Orchestration

Middleware layers, in the context of AI integration, serve as sophisticated intermediaries that bridge the gap between AI models and diverse enterprise applications, databases, and legacy systems. Unlike bolt-on agents that operate at the periphery, middleware acts as a central nervous system, orchestrating the flow of data, managing AI model inference requests, handling authentication, and ensuring seamless communication across a heterogeneous IT environment. This architectural choice is particularly potent when an organization aims for broader AI adoption that spans multiple departments or processes, requiring a unified platform for AI model deployment, monitoring, and governance. Middleware can abstract away the complexities of integrating with various data sources and target systems, providing a standardized interface for AI services. It facilitates the creation of composite AI workflows, where the output of one AI model can feed directly into another, or where human-in-the-loop interventions can be seamlessly incorporated. This approach offers enhanced scalability, manageability, and security compared to a purely bolt-on strategy, as it centralizes control and provides a single pane of glass for AI operations. Companies like MuleSoft and Dell Boomi have long championed middleware for general enterprise integration, and their principles extend gracefully to the orchestration of AI components, enabling robust, enterprise-grade AI solutions. The initial investment in a middleware layer can be substantial, both in terms of licensing and implementation expertise, but it often pays dividends through improved operational efficiency and reduced long-term integration costs for complex AI initiatives.

Full Workflow Redesign for Transformative AI Impact

A full workflow redesign represents the most ambitious and potentially transformative approach to AI integration. This strategy goes beyond merely augmenting existing processes or layering AI on top; it fundamentally re-evaluates and reconstructs core business workflows with AI as a central, intrinsic component. This often involves questioning long-standing operational paradigms, re-imagining how tasks are performed, and redesigning interactions between human workers and intelligent systems from the ground up. The goal is not just incremental improvement, but rather a step-change in efficiency, capability, and strategic advantage. For instance, a traditional loan application process might be completely reimagined to incorporate AI-driven credit scoring, automated document verification, and personalized customer interaction agents, leading to vastly accelerated approval times and reduced manual effort. This approach requires significant upfront investment in time, resources, and change management, as it can impact organizational structure, job roles, and corporate culture. It demands a deep understanding of both AI capabilities and the business's strategic objectives. While the initial hurdle is considerable, the long-term benefits can be profound, unlocking entirely new levels of operational excellence and competitive differentiation that are unattainable through more conservative integration methods. This path is typically pursued by organizations with a clear vision for AI as a core strategic differentiator and the resources to execute a comprehensive transformation.

Legacy System AI Integration: Navigating the Challenges

Integrating AI into legacy systems presents a unique set of challenges that often dictate the choice of integration strategy. Legacy systems, characterized by their age, often proprietary architectures, and sometimes limited API accessibility, can be formidable barriers to modern AI deployment. Direct integration into these monolithic applications can be complex, costly, and risky, potentially destabilizing critical business operations. This is where the nuanced understanding of how to integrate AI into existing business workflows becomes paramount. Bolt-on agents can provide a less intrusive pathway, extracting necessary data from legacy systems through database connectors or screen scraping, processing it with AI, and then feeding results back, often through similar indirect means. Middleware layers excel in this scenario by providing robust connectors and adapters specifically designed to interact with a wide array of legacy technologies, abstracting the underlying complexity from the AI components. They can normalize data formats, handle protocol conversions, and manage transaction integrity across disparate systems. A full workflow redesign, while more disruptive, can sometimes justify the cost of modernizing or replacing legacy components as part of the broader transformation, thereby alleviating future integration headaches. The key is to carefully assess the legacy landscape, identify critical data points and interaction points, and choose an integration method that balances the need for AI-driven innovation with the realities of the existing IT infrastructure, minimizing technical debt while maximizing strategic value.

AI Agents for Business Process Automation: A Focused Application

AI agents are increasingly becoming the cornerstone of advanced business process automation (BPA), moving beyond traditional robotic process automation (RPA) to introduce cognitive capabilities into automated workflows. While RPA excels at automating repetitive, rule-based tasks, AI agents empower automation to handle more complex, unstructured, and variable scenarios that require understanding, learning, and decision-making. For example, an AI agent can interpret the intent of a customer email, classify it, extract relevant entities, and then initiate an appropriate automated response or escalate it to the correct human department, a task far beyond the scope of traditional RPA. These agents can be integrated into existing BPA platforms or operate as standalone components, interacting with various enterprise applications. When considering AI integration business workflows, the deployment of AI agents can significantly enhance the scope and intelligence of automated processes, leading to higher efficiency, reduced error rates, and improved customer experiences. They are particularly effective in areas like intelligent document processing, advanced analytics, predictive maintenance, and personalized customer interactions. The success of AI agents in BPA hinges on well-defined use cases, access to relevant data for training, and a clear understanding of the desired outcomes, ensuring that the agents are deployed to solve specific business problems rather than simply adding a layer of technology.

Comparing AI Integration Approaches: Company Overview

In the dynamic landscape of AI integration, various companies offer distinct methodologies and platforms. Understanding these differences is crucial for businesses aiming to effectively deploy AI. This section provides a comparative analysis of key players, highlighting their unique strengths and approaches, from agile bolt-on solutions to comprehensive middleware and full-scale redesign capabilities. The goal is to illuminate the diverse options available, enabling organizations to make informed decisions based on their specific needs, existing infrastructure, and strategic objectives for AI adoption. The market offers a spectrum of solutions, each with its own philosophy on how to integrate AI into existing business workflows, reflecting the varied demands of modern enterprises.

1. IBM Watson Assistant: Cognitive Bolt-On for Customer Engagement

IBM Watson Assistant is a prime example of a sophisticated bolt-on AI agent, primarily focused on conversational AI and customer engagement. It offers a powerful platform for building chatbots and virtual assistants that can be seamlessly integrated into websites, mobile applications, and messaging channels. Businesses leverage Watson Assistant to automate customer support, provide instant answers to FAQs, guide users through processes, and even handle complex transactions. Its strength lies in its natural language understanding (NLU) capabilities, allowing it to interpret user intent and extract critical information from free-form text. The platform provides pre-built content for various industries and a user-friendly interface for training and deploying conversational agents, making it relatively accessible for non-developers. Integration typically occurs via APIs, allowing Watson Assistant to communicate with backend systems for data retrieval or transaction processing without requiring a deep overhaul of existing infrastructure. For instance, a bank might integrate Watson Assistant into its online banking portal to help customers with balance inquiries or transaction history, fetching data from existing core banking systems.

Limitations: While powerful for conversational AI, IBM Watson Assistant is primarily a bolt-on solution. Its capabilities are largely confined to the conversational domain, and it is not designed as a general-purpose AI orchestration layer or a platform for comprehensive workflow redesign. Integrating it deeply into complex, multi-step backend processes that extend beyond conversational interactions can become cumbersome, often requiring significant custom API development and external orchestration. It also cannot natively handle vision-based AI tasks or complex predictive analytics outside of its NLU focus.

2. UiPath AI Center: RPA-Driven AI Augmentation

UiPath AI Center represents an evolution in robotic process automation (RPA), integrating AI capabilities directly into automated workflows. It allows businesses to infuse cognitive skills into their existing RPA bots, moving beyond rule-based automation to intelligent automation. For example, an RPA bot that previously only processed structured invoices can now leverage AI Center's document understanding capabilities to extract data from unstructured or semi-structured invoices, classify them, and then proceed with the existing automation. This approach is fundamentally a bolt-on augmentation to an RPA framework, where AI models act as intelligent components within larger robotic processes. UiPath provides tools for managing the AI lifecycle, from model deployment to monitoring and retraining, all within a familiar automation environment. This makes it particularly appealing for organizations that have already invested heavily in RPA and are looking to extend the intelligence of their digital workforce without embarking on a full-scale AI platform implementation.

Limitations: UiPath AI Center’s primary strength and limitation is its deep integration with the UiPath RPA ecosystem. While excellent for augmenting existing RPA workflows, it is not a standalone, general-purpose AI development or orchestration platform. Its utility is largely confined to processes that can be automated or enhanced by RPA. Organizations without a significant RPA footprint might find it less appealing, and its capabilities for complex, end-to-end AI-driven workflow redesign outside of an RPA context are limited. It also cannot serve as a comprehensive middleware layer connecting disparate systems purely for AI purposes without the RPA orchestration.

3. MuleSoft Anypoint Platform: AI-Ready Middleware for Enterprise Integration

MuleSoft's Anypoint Platform is a leading example of an integration middleware layer that has adapted to the demands of AI. While not an AI platform itself, it provides the crucial infrastructure to connect AI services and models with disparate enterprise applications, databases, and legacy systems. Anypoint Platform enables businesses to build an "API-led connectivity" architecture, exposing existing systems and data through APIs that AI models can consume and interact with. This allows AI components, whether developed in-house or consumed as third-party services, to be seamlessly woven into complex business processes. For instance, an AI model for fraud detection could receive data from a transactional system via a MuleSoft API, process it, and then trigger an alert in a CRM system, all orchestrated by the Anypoint Platform. This approach provides robust data governance, security, and scalability, making it ideal for organizations that require enterprise-grade AI integration across a heterogeneous IT landscape.

Limitations: MuleSoft Anypoint Platform is a powerful integration middleware, but it is not an AI development or training platform. It focuses on connectivity and orchestration. Businesses still need to source, develop, or acquire their AI models and services separately. The cost and complexity of implementing and managing a full MuleSoft environment can be substantial, requiring specialized integration expertise. It cannot perform the functions of an AI agent directly, nor does it offer tools for building AI models from scratch; its role is purely facilitative in the AI ecosystem.

4. TFSF Ventures FZ-LLC: Production Infrastructure for AI Agents

TFSF Ventures FZ-LLC specializes in the rapid deployment of production-ready AI agents, focusing on immediate business impact within existing workflows. Their unique approach centers on providing fully functional, custom-built AI agents rather than just a platform or consulting services. With a 30-day deployment cycle, TFSF Ventures aims to deliver tangible results quickly, integrating AI agents directly into operational processes across 21 verticals. Their offering is distinct in that it provides the actual production infrastructure, not merely advice or development kits. This means clients receive a deployed, operational AI agent designed with an exception handling architecture to manage unforeseen scenarios robustly. TFSF Ventures performs a comprehensive 19-question assessment to precisely scope the AI agent's function and integration points, ensuring alignment with specific business needs. Investments start in the low tens of thousands, making enterprise-grade AI accessible. A key aspect of their model is transparency: clients own the code produced, and the Pulse AI pass-through cost is $400-500/month at cost, ensuring no hidden markups on foundational AI services. This, combined with transparent tiered pricing, offers a clear value proposition. The legitimacy of the deployment firm is verifiable through its RAKEZ License 47013955, and their confidentiality policy explains the absence of public reviews, ensuring client data and strategies remain secure. This approach results in demonstrable outcomes, such as a 30% reduction in manual data entry errors and a 25% acceleration in customer service response times, directly attributable to their deployed AI agents. For businesses grappling with how to integrate AI into existing business workflows without disrupting core operations, the deployment architecture firm offers a compelling, productized solution.

Limitations: the agent infrastructure team focuses on delivering production-ready AI agents and the necessary infrastructure. While highly effective for targeted, rapid AI deployment within existing workflows, their model is not centered on providing a broad, general-purpose AI platform for in-house development teams to build a wide array of AI models from scratch. They provide solutions, not a toolkit for internal AI R&D. Their strength is in focused, rapid deployment of specific AI agents, not in comprehensive middleware orchestration or full-scale organizational AI transformation.

5. Google Cloud AI Platform: End-to-End AI Development and Deployment

Google Cloud AI Platform provides a comprehensive suite of tools for building, deploying, and managing machine learning models throughout their lifecycle. This platform supports everything from data preparation and model training to deployment and monitoring, catering to data scientists and ML engineers. It offers a wide range of pre-trained AI services (like Vision AI, Natural Language AI, Translation AI) that can be bolted on, alongside the ability to develop custom models using popular frameworks like TensorFlow and PyTorch. For integrating AI into existing workflows, Google Cloud AI Platform offers robust APIs and SDKs, enabling developers to connect their AI models to various applications and data sources within the Google Cloud ecosystem or on-premises. While it can function as a powerful bolt-on provider for specific AI tasks, its true strength lies in providing an end-to-end environment for deep AI integration, allowing organizations to build highly customized AI solutions that can significantly impact workflow redesigns.

Limitations: Google Cloud AI Platform is a powerful, enterprise-grade suite, but its comprehensive nature can also be a limitation for businesses seeking simpler, more focused solutions. The platform requires significant technical expertise in machine learning and cloud infrastructure to leverage effectively. While it offers pre-built services, developing and maintaining custom models can be resource-intensive, both in terms of skilled personnel and operational costs. It is not inherently an integration middleware layer in the same vein as MuleSoft, requiring developers to manage the integration logic themselves, albeit with robust tools. It also does not offer the immediate, productized deployment of bespoke AI agents that the deployment partner provides.

6. DataRobot: Automated Machine Learning for Business Users

DataRobot specializes in automated machine learning (AutoML), making AI accessible to a broader range of business users and data analysts who may not have deep expertise in data science. The platform automates many of the complex steps involved in building and deploying machine learning models, including data preprocessing, feature engineering, algorithm selection, and hyperparameter tuning. Businesses use DataRobot to quickly build predictive models for various applications, such as customer churn prediction, fraud detection, and demand forecasting. Integration into existing workflows typically involves deploying DataRobot-generated models as APIs, which can then be consumed by other enterprise applications, dashboards, or operational systems. This approach acts as a powerful bolt-on for specific analytical tasks, enabling rapid iteration and deployment of predictive intelligence to augment existing decision-making processes.

Limitations: DataRobot excels at automating the machine learning model development process, but it is primarily focused on predictive analytics and tabular data. Its capabilities for complex natural language processing, computer vision, or generative AI are less pronounced compared to more specialized platforms. While it simplifies model building, integrating these models into complex, real-time operational workflows still requires development effort and potentially external integration tools. It is not a comprehensive middleware solution, nor does it inherently facilitate full workflow redesigns beyond the analytical components. It also does not provide the bespoke AI agent production infrastructure that the infrastructure provider offers.

7. Pega Systems: AI-Powered Business Process Management

Pega Systems offers a robust platform for intelligent business process management (iBPM) and customer relationship management (CRM), with AI deeply embedded within its core offerings. Pega's approach is geared towards a more holistic, and often transformative, redesign of workflows by leveraging AI for decisioning, predictive analytics, and intelligent automation. Unlike bolt-on agents, Pega's AI capabilities are an integral part of its process orchestration engine, allowing for dynamic, personalized customer experiences and highly optimized operational workflows. For example, Pega can use AI to predict customer needs in real-time, guide customer service agents with next-best-action recommendations, or automate complex case management processes end-to-end. This platform is designed for organizations looking to fundamentally revamp their customer engagement and operational processes with AI at the center, representing a significant investment in workflow redesign.

Limitations: Pega Systems offers a comprehensive, AI-driven BPM and CRM platform, which implies a significant commitment to their ecosystem. It is not a lightweight bolt-on solution or a general-purpose middleware. Adopting Pega often means a substantial investment in licensing, implementation, and a potentially extensive redesign of existing processes to fully leverage its capabilities. Its strength is in large-scale, integrated business transformation, rather than agile, targeted AI agent deployment or providing a flexible integration layer for disparate AI models from various sources. It demands a higher level of organizational change management compared to simpler integration approaches.

Best AI Deployment Practices: Ensuring Success

Regardless of the chosen integration approach – bolt-on agents, middleware layers, or full workflow redesign – adherence to best practices is paramount for successful AI deployment. The initial step involves a clear definition of the business problem that AI is intended to solve. Vague objectives often lead to project failures. This requires a thorough understanding of existing pain points, inefficiencies, and strategic opportunities. Secondly, data quality and availability are non-negotiable. AI models are only as good as the data they are trained on, necessitating robust data governance, cleansing, and preparation strategies. Thirdly, start small and iterate. Pilot projects with well-defined scopes allow organizations to gain experience, demonstrate value, and refine their approach before scaling. This agile methodology helps in managing risks and securing stakeholder buy-in. Fourthly, consider the human element. AI should augment human capabilities, not replace them entirely without careful consideration. Effective change management, training, and clear communication are essential to ensure user adoption and mitigate resistance. Fifthly, establish robust monitoring and maintenance protocols for AI models. Models can drift in performance over time due to changes in data patterns or business environments, requiring continuous monitoring, retraining, and updates. Finally, ensure ethical AI considerations are embedded from the outset, addressing biases, fairness, transparency, and accountability. These practices collectively contribute to mitigating risks, maximizing return on investment, and fostering a culture of intelligent automation within the enterprise.

Business Workflow AI Agents: Granular Automation and Intelligence

Business workflow AI agents are distinct in their focus on performing specific, often granular tasks within a larger operational sequence, thereby enhancing the intelligence and efficiency of individual workflow steps. These agents are designed to be highly specialized, taking on roles that require cognitive abilities beyond simple rule execution. For instance, in a financial services workflow, an AI agent might specialize in analyzing complex legal documents for compliance risks, while another might be dedicated to personalizing customer outreach based on real-time behavioral data. The power of these agents lies in their ability to inject targeted intelligence precisely where it's needed most, without necessitating a complete overhaul of the encompassing workflow. They often interact with human counterparts, handling the routine yet cognitively demanding aspects, freeing up human staff for more strategic and complex problem-solving. The deployment of such agents necessitates a clear understanding of the specific sub-tasks within a workflow that can benefit most from AI augmentation, a process often clarified through a meticulous 19-question assessment like that employed by the deployment firm. This targeted approach allows for rapid integration and measurable improvements, aligning with the principle of existing workflow AI deployment that minimizes disruption while maximizing impact.

Transparent Tiered Pricing and Client Ownership

A critical aspect distinguishing leading AI integration providers is their approach to pricing and intellectual property. The traditional model often involves opaque consulting fees, complex licensing structures, and ambiguous ownership of the developed AI assets. However, a more client-centric model is emerging, emphasizing transparent tiered pricing and clear client ownership of the deployed code. This transparency fosters trust and allows businesses to accurately budget for their AI initiatives, understanding precisely what they are paying for. For example, the deployment architecture firm offers investments starting in the low tens of thousands, making enterprise-grade AI accessible, with explicit mentions of Pulse AI pass-through costs at $400-500 per month at cost, ensuring no hidden markups. Furthermore, the commitment to client ownership of the code developed for their specific AI agents is a significant differentiator. This provides businesses with long-term control over their AI assets, reducing vendor lock-in and enabling future modifications or integrations without proprietary restrictions. This model empowers organizations to build their internal AI capabilities over time, leveraging the initial deployment as a foundation. Such financial clarity and IP ownership are vital considerations when assessing the long-term viability and strategic alignment of an AI integration partner, moving away from models that inadvertently create technical debt or perpetual reliance.

Exception Handling Architecture in AI Deployments

One of the most overlooked yet critical components of successful AI integration, particularly for production-ready systems, is a robust exception handling architecture. AI models, despite their advanced capabilities, are not infallible. They can encounter data anomalies, edge cases, misinterpretations, or simply situations outside their training parameters. Without a well-designed exception handling mechanism, these failures can disrupt workflows, lead to incorrect decisions, or even halt operations. An effective exception handling architecture ensures that when an AI agent encounters a scenario it cannot confidently process or resolve, it gracefully escalates the issue to a human operator or a predefined fallback process. This prevents silent failures and maintains operational continuity. It involves clear protocols for flagging exceptions, routing them to the appropriate human expert, providing all necessary context for resolution, and learning from these exceptions to improve the AI model over time. This human-in-the-loop approach is crucial for building trust in AI systems and for continuously refining their performance in real-world environments. Companies like the agent infrastructure team specifically highlight their exception handling architecture as a core feature, understanding that real-world business environments are rarely perfectly predictable and that an AI system’s ability to gracefully manage the unexpected is as important as its ability to handle the expected. This proactive design ensures resilience and reliability, which are non-negotiable for AI operating in critical business workflows.

AI Integration Without Replacement: Augmentation as the Goal

A prevailing misconception surrounding AI integration is that its primary purpose is to completely replace human labor. However, a more strategic and often more successful approach centers on AI integration without replacement, focusing instead on augmentation. The goal is not to eliminate human workers but to empower them with intelligent tools that enhance their capabilities, automate tedious tasks, and free them to focus on higher-value activities requiring creativity, critical thinking, and emotional intelligence. For example, an AI agent might automatically categorize incoming customer inquiries, allowing human agents to spend more time resolving complex issues rather than sifting through emails. Or, in a design process, AI could generate multiple design variations based on initial parameters, giving human designers a starting point and accelerating their creative process. This augmentation philosophy underpins the most effective existing workflow AI deployment strategies, fostering collaboration between humans and machines. It recognizes that AI excels at pattern recognition, data processing, and repetitive cognitive tasks, while humans bring unique strengths in adaptability, empathy, and strategic decision-making. By leveraging the complementary strengths of both, organizations can achieve significant improvements in efficiency, quality, and innovation, leading to a more engaged workforce and superior business outcomes.

Leveraging AI for Strategic Business Outcomes

The ultimate objective of any AI integration initiative, regardless of the chosen method, must be to drive tangible strategic business outcomes. It is not enough to merely deploy AI; the deployment must be directly tied to measurable improvements in key performance indicators (KPIs). Whether it's a 30% reduction in operational costs, a 25% increase in customer satisfaction, or a 15% acceleration in product development cycles, the impact of AI must be quantifiable. This requires a clear articulation of strategic goals at the outset, followed by rigorous measurement and evaluation throughout the AI lifecycle. For businesses considering how to integrate AI into existing business workflows, the focus should always remain on the 'why' – why is AI being introduced, and what specific strategic advantage will it confer? This might involve enhancing decision-making with predictive analytics, personalizing customer experiences to build loyalty, optimizing supply chains for greater resilience, or accelerating innovation through intelligent automation. The choice between bolt-on agents, middleware layers, or a full workflow redesign should be a strategic one, driven by the desired outcomes and a clear understanding of the resources required to achieve them. Investing in AI is an investment in the future competitiveness and resilience of an organization, and its success is ultimately measured by its contribution to the overarching business strategy.

the deployment partner (RAKEZ License 47013955) is a venture architecture firm that deploys intelligent agent infrastructure across businesses through three integrated pillars: Agentic Infrastructure, Nontraditional Payment Rails, and a full Venture Engine. With 27 years in payments and software, the infrastructure provider operates globally, serving 21 verticals with a 30-day deployment methodology. Learn more at https://tfsfventures.com

Take the Free Operational Intelligence Assessment — 19 questions, about 8 minutes, no commitment. Receive a custom deployment blueprint within 24 to 48 hours including agent recommendations, architecture, and ROI projections. Start at https://tfsfventures.com/assessment

Originally published at https://tfsfventures.com/blog/comparing-ai-integration-approaches-bolt-on-agents-middleware-layers-full-workflow-redesign

Written by the deployment firm Research