How to enable a vision AI project with Alteia?

If you plan to leverage vision AI in your organization, consider running a project with Alteia to prove the feasibility of your idea, generate initial results, and plan for full deployment.

When you plan to undertake a vision AI project, you may feel unsure of your path forward.

To secure approval for the project and make sure it runs smoothly, you’ll likely prepare an internal plan that covers these four elements – and maybe more:

A business objective
A process or structure for the project
A budget
A timeline

This article aims to help you with all four. It describes the role of a Proof of Concept (PoC) in ensuring the success of your early efforts with vision AI. You’ll also understand how much time and money to budget for running a PoC with Alteia. The recommendations here are based on Alteia’s experience with more than 50 organizations that have completed successful pilot projects with us.

Identify what a successful vision AI project looks like

Your top goal for a pilot project is to anticipate the questions it must answer. What will it take to secure approval for the next step beyond your pilot?

Before going further in your planning, ask yourself:

What business results or outcomes do we expect our project to deliver?
What hypotheses do we want to test? What questions are we trying to answer? What problems are we trying to solve? What goals are we trying to reach?
What are the essential criteria our project must achieve to be approved for rollout on a broader scale?
What stakeholders need to be involved ?
For this project to be successful, what would have to be true?
What will the follow-on project involve, after the pilot is successful?

Start as simply as possible

We suggest an initial project of limited scope. It should have just enough outputs to be usable by a given set of users who can then provide feedback for future product development, and low complexity so your company can execute it fast, at reasonable cost and with minimal risk:

Consider limiting the scope of your project to a single business function, business unit, or location
Limit your pilot to a narrow use case
Use data sets that are easy and inexpensive to obtain, or use existing ones
Avoid plans that require complex or resource-intensive integrations of data or system

Define well your objectives and budget accordingly

You can design your proof of concept to test a variety of hypotheses, assumptions, or approaches. The success of your PoC will depend on how clearly you define these in advance.

Do you want to test the validity of deploying an algorithm or model to perform a specific task?
Do you want to test the architecture of a new data model or software application?
Do you want to test the end-to-end workflow of your process, as it would occur in a full deployment?

Your budget should vary with your objectives

Let’s say you want to test generic object-recognition or segmentation models on a given set of data. For that you might invest in basic model development and processing. Your PoC budget might be in the range of $50 kUSD. What if you want to integrate a computer vision workflow in your IT infrastructure? And you want to validate that the approach creates value for your entire organization?

In that case, you should plan for a PoC that tests all of these elements:

Customized data or software architecture
Development of custom models
Deployment and metrics to prove the solidity of a planned implementation

In this case, your PoC budget for a project is likely to be around 150 KUSD.

Budget more to prove more

We recommend that you plan and budget for a PoC that’s solid enough so managers can make a confident decision after it ends.

A proper PoC creates evidence about the practical feasibility and deployment of an idea or an approach:

Can you execute it properly?
What effort, skills, and resources will it require?
Will it meet the needs and expectations of your users?
How do I measure its success?
What challenges and costs are likely to arise in deploying the application at scale?
How will it integrate with the rest of the organization?

Some companies undertake PoCs we believe are too limited to assess the feasibility of a vision AI project. Instead of trying to understand how the technology can help their organization, they focus on proving the technology.

How does Alteia contribute to your success ?

Our work generally include the following topics:

Data audit, ingestion, consolidation, and contextualization

We identify what data can be leveraged (existing or new data) in order to solve the business problem that needs to be addressed, and recommend ad hoc data acquisition methods when relevant.

Model building and training

We prepare the training datasets, build the models with the right framework, test the models, and deliver accuracy reports.

Data processing and analysis

We integrate the model into a data processing workflow including the generation of outputs and the validation that the model accuracy matches the expected business outcomes.

Visualization of the outputs

If relevant, we build a specific UI to interact more easily with the data and outputs in order to integrate human-based validation processes for example.

Definition of an architecture for full deployment

We help design the software architecture matching the IT constraints of your organization and support you along all the internal validation steps.

At the end of an Alteia PoC, you have access to these elements:

A test not only of key technical elements of your application, but also of its operational feasibility at scale
A good sense of the software, data, processes, skills, and IT infrastructure you will need to deploy your application after the pilot
A list of recommendations for additional activities to prepare for deployment and scale-up
A clearer vision of the roadmap to production
A list of identified risks and mitigation plan for scaling your project
A set of tools and AI models tailored to your business requirements