A major issue in claiming an R and D tax deduction is to mapping the language that the government defines as r and d to your project. 

1. What is the science?
2. What is the technical risk?
3. Is it innovative? 
4. What testing have you done? 

And the list goes on 

The reality is that most software projects and ongoing developments of software and apps should be defined as R and D . It all about the interpretation and justification – and the key is to be able to document and justify your project as an r and d one….because I believe that anything can be justified, and the intention is that the government wants to encourage companies to continue to innovate – 

Because innovation is the food for growth! 

Everyone is talking AI and machine learning….. is an app just a microcosm of a larger machine learning AI model? 

So, what makes AI or an app useful?  It’s all about how data in the model is used.

An AI model is a way of looking at data – as data changes, the AI model nnneds to adapt accordingly – 

An AI system needs to be built with five attributes in mind says Dinesh Nirmal – vice president of analytics development at IBM. 

1. Managed

The AI and machine learning model needs to have  thoughtful, durable, and transparent infrastructure. 

That starts with identifying the data pipelines and correcting issues with bad or missing data. There needs to be a methodology of integrating  data governance and version control for models. The version of each model — and their might be thousands of them concurrently needs to clearly  indicates its inputs – 

where the data came from needs to be known 

2. Resilient

Being fluid means accepting that models will fall out of sync. That “drift” can happen quickly or slowly depending on what’s changing in the real world. Regression testing needs to be done on a regular basis. .

Accuracy thresholds need to be defined and and automatic alerts to let you know when the model need attention. 

Will you need to retrain the model on old data, acquire new data, or re-engineer your features from scratch? The answer depends on the data and the model.

Before trying to find the problem, one needs to look at defining the problem.

3. Performance 

The AI model needs to compute the transactions in milliseconds, not minutes, to gain a competitive advantage and make the system work. 

Optimum performance is key 

The AI model needs to run fast and error-free regardless of where you deploy it on premises , or in the cloud.

4. Measurable

The results and outputs need to be clearly measured and have adequate reports. 

When starting the project , visualize how you are going to report what you’re learning and how it changes.

What you can measure you can manage – think about how you can easily report on short , medium and long term goals 

Some Kpis 

1. improvements in data access and data volume,
2. improvements in model accuracy, and ultimately
3. improvements to the bottom line.

5. Continuous

The AI model needs to change and continuously learn as the world changes. The Ai model needs to be continuously evaluated and retrained to adapt to a changing world. 

Jupyter and Zeppelin notebooks that can plug into processes for scheduling evaluations and retrain models are useful tools to use 

You will gain an understanding of absorbing  the advantages and limitations of the algorithms, languages, datasets, and tools that are being used. 

Fluid AI demands continuous improvement for data, tools, and systems, but also continuous improvement from the team. 

Data science is a journey. Pay attention to these five attributes and you’ll bring focus to each moment and force yourself to find clarity about the future.

The data will never sit still, but would you really want it any other way?