Explore The Managed Capacity Model for successful AI Solution Development AI Delivery

Nov 6, 2023

85% of AI projects fail due to unclear objectives and poor R&D project management

Just imagine, the adoption of artificial intelligence has increased by 270% within the last four years. The expected result accounts for a $15.7 trillion contribution to the global economy by 2030. 

Orgniazations across all industry domains are seeking to integrate AI (if they haven’t already). They aim for predictive analytics, intelligent vehicles, fraud management, and other process-oriented models with extra layers of automation and insight. Still, there is a lack of understanding regarding the use of artificial intelligence. 

Gartner states that 85% of AI projects fail due to unclear objectives and obscure R&D project management processes. As well, 87% of R&D projects never get to the production phase, while 70% of clients indicated minimal or even no impact from AI.

As one of many examples, the collaboration to improve cancer care coordinated by IBM and The University of Texas M.D. Anderson Cancer Center wasted $62 million due to improper planning and unrepresentative data training for AI solutions.  

While proper R&D project management and development methodology isn’t the be all end all of AI project success, it can have a serious impact on the success (or failure) of an AI project. 

At Neurons Lab we have extensive experience in AI delivery. This article will tell you more about: 

  • Reasons why you may be disappointed with your R&D project
  • Essential differences between AI projects and traditional software 
  • Specifics and benefits explained about managed capacity model
  • 5 key benefits of using a managed capacity model for AI projects
  • Agile project development methodology
  • Successful delivery examples of AI projects by Neurons Lab

Let dive in.


Typical Pain Points Many Face with AI Project Development

Many companies strive to differentiate user experiences, disrupt markets, and create digital products to delight customers. Data and work with analytics both play a huge role here, making artificial intelligence and machine learning the natural step for innovation first companies looking to outperforming other players. 

Options like outstaffing data engineers or allocating a team of AI/ML specialists internally are some existing working models for starting AI projects. However, there are pros and cons to both approaches. 

It is worth noting that it does not matter if the company is a startup or an enterprise with millions for their R&D budgets. Big players like Apple, Facebook, and Amazon frequently fail with AI projects. For instance, the social media giant Facebook appears to be very inefficient at predicting illegal and hateful content and removing it from the platform. 

To continue, Amazon failed with its facial recognition system. The system works with just 53% accuracy. A similar issue arose with Apple’s face ID. This technology failed since a plastic mask was successful at fooling it and made those users concerned with their privacy very disappointed.

You may have great expectations of quick and outstanding results from AI technology while, in reality, you may face pain points like:

  • AI technology overpromises and under delivers
  • Budgets for AI/ML projects are overblown 
  • Research is prolonged over many years
  • Enormous difficulty in forecasting the end-results 
  • A high level of uncertainty is still in place 

So why does AI/ML fail to deliver what is expected? And are there any options or methods that could increase the chances of successful R&D project implementation? 

To start, the most critical factors you need to consider include: 

1. Insufficient data – it is necessary to choose the correct data to train the system, and it should be sufficient to follow real-world patterns and scenarios without bias. 

2.Poor engineering – there is still a lack of capable AI/ML engineers and data scientists since the technology is still evolving. The best mids are hardly affordable. 

3.Complex areas for AI application – areas like healthcare, law, and construction require specific understanding and lack the data for proper analysis. For building an accurate AI system, data scientists that are more than just capable are needed, but also field specialists and enough applicable information.  

Note that innovative startups sponsored by venture capitalists experience around a 75% failure rate even though they are at the forefront of technological development. 

Key takeaway: Failure is inevitable; this is just a part of the process, and the company capable of building and refining internal processes for the ability to experiment will win. Google and other data-driven organizations differ only due to their ability to learn from mistakes and pivot. 

So what is the best way to use AI to create new business models and products and innovate in a way that no one else has done before?

All in all, traditional outstaffing methods are not enough for digital transformation within R&D projects. These days, AI-empowered projects should include not only tech knowledge and capabilities, but also strategic view, agile co-creation flow, automation to change, innovation enabling, and actual result-delivery that is faster and cheaper.


Differences Between Traditional & AI-Powered Projects 

Let’s first check out what makes AI projects different from any others. Clearing up the difference between AI and traditional software development can help set the correct expectations for R&D projects and their development approach. 

1. AI projects have a higher level of uncertainty

AI projects imply a high level of innovation compared to traditional software development. Standard apps and websites concentrate on resolving business issues or creating a specific solution. In the case of an AI project, the results are more uncertain, definitely more risky, and work with many unknown factors. This leads to the thought in the minds of stakeholders that AI technology overpromises and underdelivers, making stakeholders unhappy.

It is only possible to make some specific conclusions with the project flow once the hypotheses have been tested. AI projects should be oriented based on their own rules and processes, which implies a tailor-made approach in their development methodology.

2. AI projects have a higher degree of complexity

The most remarkable difference between solutions powered by AI is the technology behind them. The general expectation from AI is the ability to make independent decisions based on the input information. However, the created and learned AI network relies on algorithms and patterns that come up with conclusions. The primary artifact of any traditional software is the code, while it is the data in AI. Thus, the main deliverable of the AI project concentrates on data collection, labeling, and analytics with the help of algorithms and pattern recognition.

For example, it is possible to drive a car with the help of software. The software can collect data from microphones, cameras, radars, sensors, and similar sources and work with this data to understand the consistent patterns. Then, you can educate a vehicle on an immense amount of data and expect that it will navigate in the right direction.

3. AI projects need multi-disciplinary teams

As mentioned, AI-centered projects are incredibly complex and very diverse. These projects need the top specialists from different areas to complete the work. The team may include data scientists, developers, designers, psychologists, and user experience specialists. In the case of working as a multi-disciplinary team, the project is sourced with the necessary combination of skills, knowledge, and proven methodology to release AI projects. 

As explained by JP Baritugo, director at management and IT consultancy Pace Harmon, “Firms need solid, multi-disciplinary talent to drive AI initiatives, especially given the expertise required to extract, clean, model, and analyze the data – not to mention evangelize AI and drive user adoption. AI talent is still scarce, and developing AI talent organically may not be feasible.”

Thus, it is a mistake to think that augmenting your team with one data scientist or a few full-time specialists will be enough to attain great AI product delivery results.  

4. AI projects require many trials & failures

The process of AI product development is non-linear, requires more experimentation, and has to complete a few iterations before providing even preliminary results. This is what makes R&D projects especially exciting to work on. The final results are not always known in advance and could be both challenging and rewarding in the end. Besides, even though AI project incompletions can be irritating, these failures could still be beneficial and become the next pivot toward adopting AI. 

The director of Cognitive Automation and Innovation at ISG, Wayne Butterfield, mentioned: “Finding how not to do something might be a success. It’s relevant in the world of AI and data; so, we need to be careful in broad-brushing failures.”

Therefore, the elaboration phase for R&D projects is more complex and may need more time with a few exploratory steps and many trials before arriving at any conclusions. All these specifics of higher uncertainty, complexity, and failures dictate agile development methodology for the R&D project development. 

To note, AI does not change the nature of creating top products. Rather, it has nuances that make applying the existing software methodologies more complex. 



Existing Misconceptions about Delivering R&D Products

There are a few incorrect ideas about creating R&D projects. In many cases, business owners want to work on an AI-driven project for a fixed price, outstaff an AI/ML engineer, or hire professionals in-house. Let’s shed some light on some ineffective approaches to delivering R&D products and not waste time and money.

1. Choosing staff augmentation for AI projects

You may think that if you hire a capable AI engineer with the necessary skills, you will definitely receive great end results. In reality, many business owners encounter regular failures with hired data scientists and AI professionals. Many spend time on micro-management, experience miscommunication with specialists, and consequently lose thousands of dollars.

So why is that? For starters, let’s clarify one point about the staff augmentation model here. 

Staff augmentation adds talented people to the team, filling in a gap with the necessary expertise. At the same time, the whole management process to release the planned solution relies on your internal company’s resources or your own. 

The Gartner research states that most respondents who adopt AI face many challenges. 56% of organizations experience a deficit of the necessary skills, 42% lack an understanding of the use cases of AI, and 34% experience concerns with data quality and scope.

Therefore, the issue could arise not from the specialists themselves, but from the process and inability to manage AI development. An expert may have the necessary knowledge, experience, and capabilities to perform well, but may still fail without a well-established AI product delivery process.

Any AI-powered project should include a solid data management foundation to deliver accurate insight, build trust, and reduce bias. This is the key challenge within AI projects – to prepare the correct data. 

Mr. Hare, distinguished VP Analyst at Gartner, mentions that finding the right staff is the biggest concern with any innovative technology involvement. The increasing number of AI projects implies that businesses need to reorganize and ensure that AI-based projects are staffed and funded in the right way. Companies may use managed service providers, create training programs for employees, or partner with universities. 

2. Hiring a team in-house and opting for increased quality

But what if you are thinking of hiring the necessary specialist in-house for AI project implementation?

Of course, this is a good option. This way, you will have direct access to the team, but can expect to face a lengthy and burdensome hiring and onboarding processes. This is because the market is experiencing a supply shortage of AI specialists, and real experts are hard to find. Additionally, you will need to pay extra for local taxes, vacation, sick leave, office space and equipment, and similar expenses associated with employee retention. 

As well, even if you complete the whole process, you may still lack the necessary expertise and proven AI project management methodology. 

The benefits of outsourcing AI projects to managed service providers allow you to save up to 50-70% on these IT expenditures. Many studies, including one by Capital Counselor, have mentioned that cost reduction is the main reason for IT outsourcing. This means that you pay only for the hours each specialist has dedicated to your R&D project while being flexible with strategic business tasks.

Any digital transformation or software innovation requires a set of activities, including the creation of new experiences, scalable backends in the cloud, and powerful data analytics. This workload cannot be completed just by one specialist and assembling such a team in-house is a lengthy and costly process. In addition, it is necessary to create a disruptive, agile, and high-performing engineering culture. 

3. Expecting a fixed price for an AI project

Another critical misunderstanding about AI product delivery lies in the fixed cost. Would you agree that projects with the highest levels of uncertainty, a greater number of unknown factors, and added innovation cannot possibly be estimated precisely?If we speak about any software development, generally there are two main options based on which the clients are charged: Time & Material and Fixed scope. 

a) Fixed scope – a payment for the services provided for a specific timeline and scope of work. The pros include that the project budget is stated prior to the project’s start. This model is suitable only for short-term and typical projects. 

In a fixed payment scheme, deliverables, success criteria, and their order are planned and set before the start of the project in the contract agreement. This model needs comprehensive requirements upfront, which is unrealistic for R&D projects.

b) Time & Material – a payment process determined by the hours (time) required of a professional (material) to complete the necessary scope of work and, thus, is calculated based on an hourly rate. The advantage includes higher flexibility during the implementation process which fits more complex and ongoing projects. 

It is crucial to understand the benefits of using the time and material payment option for AI-driven projects, despite some possible anxiety.

These are some of the main factors behind AI project costs making it impossible to fix the price, and this list could be expanded: 

  • Type of software and its complexity
  • The domain of AI adoption 
  • Level of the product’s intelligence 
  • Quality and amount of data you will feed into the system to learn from
  • The level of accuracy you aim to achieve 

All in all, AI-based projects dictate the need for flexibility and adaptation within the process and intermediate results, and work well with time & materials payment model constraints. Initially, the client has a rough project estimate and the ability to adjust during the development process and create a solution according to the findings and requirements.

4.Having an incorrect expectation of team extension

The next misunderstanding has similar connections to the fixed scope. Many think that once there is a need to add an expert, then it is necessary to pay the full-time cost of the specialists added, similar to the outstaff model. Within the managed capacity model, the client pays for the project’s delivery and its management. In this regard, the managed service adds value by covering processes, scenarios, predicting numbers, and applying practices where new team members can be amplified in the best way.

You may agree that in different processes and environments, the same professional, even the most capable and committed one, can work with a different level of productivity. 

With the managed capacity model, the added value the client pays for is the framework and the process that allows for the completion of the AI project according to its needs and with the highest effectiveness of the resources, costs, and timeline.


Managed Capacity Model for AI Projects: Details Explained

The definition of a managed capacity model (sometimes called managed delivery) is one of the IT outsourcing engagement models. Managed service providers (MSPs) are defined as strategic outsourcing partners that manage and deliver IT projects remotely. The model is composed of two words explaining everything – “managed” and ‘’capacity.“

The word “managed” means that the full project development and management process lies on the IT services provider’s shoulders. The managed service provider carries complete responsibility for the project’s delivery based on the client priorities defined. 

An integral part of the model is allocating a unique pool of talents in the form of agile teams with cross-functional skills and experience. The mix of team members is based on project needs, specifics, technology preferences, domain expertise, etc. 

The word “capacity” stands for the quantity of the services provided. The capacity factor is based on the monthly team cost and not on the project’s scope. This is because the project’s scope may fluctuate along with the project’s progress. Once the scope or project needs change, the team set adjusts its capacity accordingly.

The full-time equivalent factor (FTE) is used for calculating the working capacity of one team member, measuring the specialist’s involvement in a project. For example, 1.0 FTE equals one full-time person employment, while an FTE of 0.5 is equivalent to half of a working day.

What does this mean precisely in some examples?

In simple words, the team-set within a managed capacity model is composed of the expert’s hours necessary to complete the work rather than the need to hire 3-4 engineers on a full-time basis.

Basically, you spend just as much as is necessary for the top experts to perform well within 1 development sprint (usually lasting 2 weeks) and save some costs on unused hours compared with the example of a simple outstaff model. 

Therefore, the number of team members, their workload, and their expertise can be adjusted from sprint to sprint. As well, managed service providers foresee and anticipate the needs and business demands that cannot be seen at the very start of the complex R&D projects.

To finalize, the key benefits of the managed capacity model include: 

  • Quicker start of the project with only initial client involvement
  • Minimal project risks since the full responsibility is taken on by the selected managed service provider
  • Greater cost-efficiency compared to hiring an R&D dev team in-house
  • Improved project management, transparency, and accountability
  • Higher team motivation and commitment directed towards receiving measurable outcomes
  • Complete access to the project’s progress via clear tech documentation and knowledge

Essential Benefits of the Managed Capacity Model

So let’s clear up what benefits are provided once you choose a managed capacity model rather than opt for outstaffing an AI engineer or data scientist, namely:

1. Cost savings & easy budgeting

Many organizations, especially startups, find it enormously expensive to hire top experts like AI engineers, especially in-house. At the same time, the news states that lots of AI startups and projects fail due to inefficient management, making entrepreneurs and innovators unsure about the end result.

In the case of entrusting AI project implementation to an experienced team with a well-established project delivery process, you not only cut down on costs due to the effective workflow, but receive predictable monthly cost estimates based on the project’s scope and the team-set required to cover all the aspects. 

Thus, you will understand the project cost within the project’s flow and have a clear understanding of the project budget. This way, a MCM is the most optimal solution for the delivery of AI-powered projects available on the market. You can receive a solid procurement to cover all the services involved in the product’s development.

2. Access to innovative and unique expertise

This is the most critical benefit that you can only receive from a MCM. Once you contact a reputable and experienced managed service provider like Neurons Lab for an AI-driven project, you access the top resources and domain expertise. In one shot, you hit a few targets. 

With the managed capacity model, you get access to the top experts in their niches who can deliver the best business and technical services available on the market worldwide. As well, you have an established AI project delivery process in this package!

3. Proactive support & quick response time

The top managed IT vendors are 100% dedicated to your project. This means that the communication flow will be transparent and easy. You can receive regular updates on the project’s processes from the internal product delivery manager. 

4. Better business focus

With a managed capacity model, you will receive product management services. This frees you from the necessity to guide the project, set tasks, handle potential issues, and follow the deadlines. Business owners and entrepreneurs can focus on their direct business activities, core competencies, and strategy instead.


Why Use Agile Processes in AI Project Development?


You would surely agree that change management is an integral part of any development process, including AI projects. Priorities and success criteria of an R&D project are selected dynamically during the development, testing, and customer feedback stages. Therefore, the AI project development flow should be very flexible and exploratory to avoid bigger failures and monetary losses in the later development stages.

Agile methodology focuses on short development iterations and continuous testing and releases opting for faster value delivery, which occurs as an alternative to the traditional waterfall approach. The methodology aims to deliver some intermediate results, make some adjustments to the plan if needed, collect real user feedback, and move toward the next small phase. 

This possibility to adjust within every interaction allows for an increase in velocity and improves the adaptability of the product to the real market and the company’s needs. According to Agile Manifesto, interactions, the delivery of a working product, response to changes, collaboration with customers, individuals, and interactions are more important than tools, adherence to the plan, documentation, defined processes, and contracts.

Being based on two core principles – iterative and incremental, agile allows project development to unlock the following opportunities compared to the traditional method:

  • Budget estimate is 20-50% less, as it is planned on a monthly basis according to the team capacity, precise project needs, and management. 
  • Time to market is significantly faster due to quicker project kick-off, easier negotiations, faster development lifecycle, and adaptation to changing priorities. 
  • Product meets the market’s needs since the success criteria, project priorities, and deliverables are selected dynamically during the development testing stages, and from  feedback from clients and stakeholders. 
  • Feedback cycles are quicker, derived from agile development, and allow the creation of user-friendly and market-fit products.  
  • Improved flexibility and reaction to change requests derived from new discoveries, information from customers, market and business feedback, etc.
  • Development teams are motivated and committed since the process is clear, and changes and feedback are adjusted to the plan more quickly. 
  • Communication is strong and easier among team members, making the workflow effective.
  • Client is involved in every development sprint, can access all the intermediate deliverables, and has a complete understanding of the project’s flow. 
  • Project exit is simpler and faster within 1 sprint notice (usually 2 weeks); the client receives all project deliverables and can pause the project.

Next, we explain the managed capacity model as a dedicated team to work on the project and why it is the best way to proceed with R&D projects.


Managed Capacity Model & Agile Methodology for R&D Projects

A managed capacity model in line with agile development methodology is the way to deliver the great results that so many expect from AI and ML. The model ensures that companies receive the most value from AI/ML technologies via agile development processes and access to dedicated teams holding the top expertise and skills.

Neurons Lab provides a well-established process that has been trialed and completed on 50+ AI-powered projects. This means that we offer an actual working solution to adopt AI properly, rather than separate engineers or data scientists. This approach is a combination of technical expertise and business knowledge, as well as the latest understanding for managing R&D projects according to the agile methodology. 

Why is this approach super effective? 

Our clients receive a unique, proprietary delivery framework that is tailored to the innovation environment, even under fierce competition, tight timelines, little-to-no datasets, and the necessity to generate novel solutions.

At Neurons Lab, we use Scrum as a development process and split the process into bi-weekly iterations with regular feedback and intermediate project deliveries. From the high-level perspective, there are three major stages of AI project delivery:

  1. Scoping phase during which we collect your requirements for working on the solution and SoW (scope of work) 
  2. AI feasibility analysis known as our AI Design Sprint service that consists of the business discovery and technical feasibility
  3. AI engineering phase during which we implement a working AI product

Our company utilizes a managed capacity model where it allocates a team of 3-5 specialists to deliver AI projects. The usual team-set includes 1 specialist who translates business requirements into technical language(Principal Applied Scientist), 1-2 specialists who actually implement the solution (ML Engineer and/or Data Scientist), and 1 specialist who manages the whole process (ML Delivery Manager). 

As a result of this kind of managed capacity teamwork, the client will receive the desired  solutions, as well as: 

  • Tangible results in weeks, not years 
  • Minimum necessary domain expertise to complete the project 
  • Become educated throughout from both the biz and tech sides
  • Free of management and, especially, micromanagement 
  • Great price-quality ratio 

As we mentioned, this does not mean that you receive 3 to 5 specialists working full-time or that you should spend your budget on all these specialists full-time. This means that our team allocates the required hours of a particular specialist to complete the task. We plan the work according to your objectives and dedicate the time required to complete the set solution. 

For instance, a Principal Applied Scientist will work only 4 hours a day (0.5 FTE), and you will not need to pay for a full-time worker.

Neurons Lab offers an effective team composition that could consist of the following experts:

1.Principal Applied Scientist – a professional with a related PhD in healthcare, physics, biology, or a similar field, and 5-7 years of experience. This is a key expert that guides the team, possesses a deep knowledge of the industry domain and technical expertise, and knows how to manage an R&D project. Usually, it takes 0.33- 0.5 FTE, or 2.5 to 4 hours per day. 

2.ML Engineer – an expert in the specific domain that knows the tech side of the task and works on coding a solution. 

3.Data Scientist  – a specialist who knows how to work with data in general, namely collecting, analyzing, and interpreting extremely large amounts of data.

4.ML Ops Engineer – an engineer, whose title stands for machine learning operations, to streamline ML models into production, then maintain and monitor them. 

5.ML Delivery Manager – a  professional who knows R&D project development methodology (Agile), is responsible for managing the team, and makes sure that all artifacts are completed as stated. 

For example, to succeed with an AI project in HealthTech, you would need to hire a PhD in healthcare science with scientific knowledge and an understanding of the business context, a data scientist to work with the necessary datasets for machine learning, and a delivery manager to manage the AI project development flow.

One of the successful adoptions of AI in the marine domain was related to a vessel’s predictive maintenance. ChordX is a Singapore maritime data analytics company that focuses on the energy management of significant naval assets. Initially, statistical models for root caused fault analysis and made visualizations of the engine combustion process for insight generation. 

The AI product goal was to transform their work with asset data onboard their vessels. Our team helped them achieve the following objectives: 

  • Establish alternative models and algorithms to identify anomalies across different vessel types
  • Identify a set of solutions and provide insight that allows for early alerts and notifications to prevent failure modes
  • Create new models and algorithms that help reveal inefficiencies in the engine combustion process and generate new valuable insight into the possible root causes
  • Reduce fuel consumption and emissions

As a result, the AI solution created predicted vessel maintenance needs via advanced physics-aware deep learning models for more accurate fault detection.


Final Word: Successful AI Project Implementation

Projects involving artificial intelligence and machine learning are considered the most complex, expensive, multi-disciplinary, and time-consuming among all other IT projects and traditional software development.  Successful AI implementation requires the right team-set of multiple specialists (e.g., data engineers, software developers, designers, scientists, etc.), capabilities, and management flow.

A managed capacity model is the way to deliver outstanding results that so many expect from AI and ML technologies. The model ensures that companies receive the most value from the advanced technologies in line with agility, top expertise, and skill access. With this model, you will receive the necessary combination of skills, knowledge, and proven methodology to release AI projects and not get stuck with a prototype.

  • Managed team saves money and time on implementing the project
  • Client receives access to a pool of experts to complete the work
  • Client lowers IT expenditures
  • Managed service provider holds responsibility for project outcomes in line with the client 
  • Managed capacity teams align with agile principles and value collaboration with clients over bureaucracy

Having implemented many AI projects, Neurons Lab can help you with proper AI project delivery and help you with ongoing project growth and improvement. Note that Neurons Lab hires only the top 5% of specialists within AI/ML, data science, engineering specializations.

About Neurons Lab

Neurons Lab is an AI consultancy that provides end-to-end services – from identifying high-impact AI applications to integrating and scaling the technology. We empower companies to capitalize on AI’s capabilities.

As 1 of 18 certified AWS partners in Applied AI globally, our goal is to help businesses to unlock the full potential of AI technologies with support from our diverse and highly skilled team, made up of applied scientists and PhDs, industry experts, data scientists, AI developers, cloud specialists, user design experts and business strategists with international expertise from across a variety of industries.

Get in touch with the team to discuss your next AI initiative.

Get in touch Leverage the power of AI for your business