You're working on an enterprise software project. How can you make sure it's AI-powered?

As an early adopter of various AI tools, I integrate them into my daily routine and see their expanding role in our business. From aiding with translations to refining marketing content, AI has become an ingredient in our workflow. As someone who values direct communication in emails, I've found rewording messages with AI suggestions beneficial in enhancing tone and effectiveness. Although our business hasn't formalized an AI strategy, we actively engage in experimentation. My perspective is that if AI enhances job efficiency it's a welcome addition. Like any tool, mastery varies among individuals, with some embracing its potential while others may be cynical or struggle to maximize its benefits.

I find it a mistake to start with technology. The business does not care about technology. The business cares about delighting customers, growing revenue, decreasing costs, complying with regulations. Define the business problem and get clear. Of all the things you could do - does this rise to the top? Once you have clarified the problem and determined what is needed to pursue a solution, you can identify and pilot the proper technology - which should consider AI. AI provides unique capabilities when used properly and with adequate controls. In the short term, AI will be a wrapper and buzz word. There will be mistakes made. I'm bullish on AI. But do AI smart.

Pick a focus and learn what works with a specific project. Target an outcome that helps demonstrate that AI provides efficiency. Use that success to foster interest in AI. Take care not to take AI and paint everything with an AI brush as this will lead to disappoint. Instead, choose a project with high ROI and leverage it for learning and demonstrable success.

AI is not a goal for a company put potentially a tool to reach business goals. So one should ask the following questions: 1/ Which of the current company goals could be better(performance, cost, quality, ...) supported by applying AI? 2/What could be additional goals, products, services or features, which haven't been possible without AI and could be part of new company goals? 3/ What pre-requisites do I need to fulfil to successfully apply AI (e.g. quality data available, technology&skills available, policies in place). AI can be a solution, but shouldn't be a goal in itself.

All AI-powered projects live and die in defined use cases. Many discussions during 2023 struggled to move forwards because although there was a perceived need to integrate AI into the business, there was no clear idea on what the use case was and it's relative positive impact to the business. Where there was a clear view on what the goal was, the projects were implemented at pace.

A relevant question you should ask after selecting the goals is: do you really need AI for those tasks? Does AI make it better (or faster...)?

AI is a means to an end and hence should merely be viewed as a tool. The journey that has helped us in accomplishing our goals faster is: 1. Define use cases and goals of what you want to accomplish with them 2. Identify the best team and tools to help accomplish those goals. The skills to use those tools effectively are in short supply. Conversational AI seems simple, but it is prone to it's risks (hallucination, biases, incomplete information etc.). While doing this, make sure that the team is trained/sensitised to not expose sensitive/critical information outside. 3. Rapidly iterate over the outputs till the desired goals are achieved.

Don’t define your AI goal, define goals for overall project or solution and find right technology to solve it. Firstly, you are not moving you business forward by setting goal not alignment with your primary objectives. Secondly, AI has a huge potential but might not be an answer to every issue. It is worth to really take a step back and start with a business issue and think about AI or any other technology as an enabler not a goal as itself.

The use of AI to solve a business problem should be organic in nature. One of the use cases that we solved using AI helped us to improve the efficiency of the business teams in reviewing the documents pertaining to our domain by a few hours. But it was a two-year journey. The challenge lies in creating the training data set and continuously improving the accuracy of the model by adding to the training dataset and retraining the model.

If "Content" is King, "Context" is definitely Queen. The chosen tools must have the contextual awareness of what you are trying to achieve. Everything is subjective

AI technology is constantly evolving and there is a lot of innovation happening in this space. There are many products out there which solves specific needs through AI. Depending on your problem space, you should be able to identify a tool which you can plug-in. I'm a big proponent of reusability rather than trying to reinvent the wheel. Having said that, cost is an important factor to consider and should definitely be weighed in before adopting any tool. Assess the cost with the skill requirements as well and you will have a good model to decide whether to build in house or go for something readily available

Securing your data is P1. Then :: Reliability, Availability, and Serviceability. If the data gathered by this software, generically called AI, is not open to your analytics engine, find one that is. Data silos will make AI a faster way to failure. - Which isn't all bad, but when it is assisting police, fire or healthcare professionals, the decision could be fatal.

Like others mentioned here...would suggest not starting with what problems can AI solve, rather start with what problems your customers have and see if AI as a technology can help solve those. Sometimes, AI may not even be appropriate. If you as a company want to embrace AI and get teams excited about it, would suggest arranging for some talks, brownbags in the company and then follow it up with a hackathon for engineers to apply AI to your product use cases. I tell you amazing business ideas come out of this..then form a small team to go after couple of promising ideas. At the end of the day, AI or any other technology should be in service of business and not other way around.

Depending on your AI Goal, there are many Cloud Services available that are easy to use to begin quickly experimenting. As mentioned by the others, depending on the AI domain, there maybe existing SaaS-style solutions to support your requirements. However, I would highly encourage developing some technical skills around AI to deepen your understanding of the underlying technologies. This will help you unlock additional possibilities you many not have originally considered.

Respectfully, I would suggest that choosing your AI architecture first, will help with the tools decision. Once you understand how you want to approach your customers, and meet their needs, the required tools ought to become self-evident. Acknowledging that some tools with compete with similar capability, the base toolset that matches your requirements should be short. I think the another critical consideration is available skills in the toolset deployment. You may choose a tool that meets all your needs, but the skillset in the tool may be more rare. You might think you are saving time with this tool, but you might end up spending more money. It's a balance b/t a multi-vendor approach and less complexity, but I would prioritize skills.

You don't have to reinvent the wheels! use an AI platform that servs the purpose with no code / low code capabilities, strong connectors to data sources and generate result as an app or integrate seamlessly with main application.

A key consideration is AI tools that are low code and business user enabled. If these tools require data scientists and Lots of learning behind them whilst valuable they may not meet the timeframes and user experience people are expecting; so understand what these tools need in terms of infrastructure, people and time to deliver the required value and insights

After you've determined what you want to use AI for, you need to think of MVP and future ideal state. You need to realistically assess what you can spend on AI infrastructure. Current indications are that cost will not go down any time soon. Depending on your budget and goals and expected flow, you might want to consider a hosted model where a larger AI infrastructure vendor provides API and development platform for you to build the AI-based portion of your solution.

Modular, flexible, API based… that’s sound advice. Goal is to be agile, innovate, iterate, and fail fast when needed. The industry is evolving so quickly, with many thriving communities driving innovation, ongoing, and iterative will be necessary. To get started, I think of architecture as a component-based pipeline of use cases/ features that will ultimately comprise a platform of business services to be woven into the operating model, a combination of shared APIs and interactive end user components. You can do in-house, PaaS via Cloud, SaaS… many options with benefits and implications to each.

Designing an AI architecture is about striking a balance. It’s crucial to ensure that AI components harmoniously integrate with existing infrastructure like databases, APIs, and cloud services. We focus on creating modular, flexible workflows encompassing data ingestion, processing, and deployment. This approach enhances robustness and adaptability. Remember, AI isn’t just about complex algorithms; it’s about how these technologies seamlessly fit into your overall system, adding value without overwhelming complexity.

To AI-power your enterprise software project in implementing AI features, start by defining specific AI goals aligned with project objectives. Identify areas for automation, enhanced analytics, or improved user experience. Carefully choose AI tools such as machine learning algorithms, natural language processing, or computer vision that seamlessly integrate with project requirements. Ensure the selected tools enhance overall software functionality without unnecessary complexity. Adopt an agile approach to implementation, breaking down AI features into manageable increments for continuous testing and refinement. Regularly gather feedback and iterate to fine-tune AI capabilities.🚀✨

A critical aspect of designing your AI architecture is determining which cloud services will fit your business use case the best. Typically in a large enterprise environment, you want to design architecture that’s vendor agnostic and allows you to operate a hybrid cloud environment. That way you can ensure that you’re controlling your cloud costs while also offering the most extensive and reliable infrastructure platform to your development teams.

This depends on your target market, operating at Enterprise scale is very different to smaller more targeted architectures. High concurrency, efficiency and scale become increasingly difficult to course correct if the fundamental architectural foundations are not correct.

What surprises many people is the cost per query for generative AI. This is many, many times greater than semantic search. When designing the system think about scale from the outset as this can influence the overall viability of the service. It is not always necessary to deliver a generative answer, sometimes a semantic result instead is enough. The right index should make the process of integrating systems far quicker with pre-built OOB connectors to most main systems like; SAP, Microsoft, Salesforce, Adobe etc. Once relevant datasources (that contain content answers) are connected these are retrieved by semantic search and feed into the LLMs in chunks. Work with vendors who have demonstrable record in chunking data for this purpose.

Use an AI platform with strong built in connectors, flows capabilities to build the machine and learning process with Experts in the Loop for GenAI and also capabilities to render as an app or connect to existing app solution.

Decouple your GPU compute from other concerns. The market is going to continue to move fast with serving machine learning models. Traditional architectures should work fine otherwise. Queues, lambdas, OAuth, SPAs or SSR, whatever your engineering force knows well can be leveraged. Invest in CI/CD and IaC. Investigate and mitigate bandwidth and storage charges that may arise from your plan. Models and datasets can be enormous, much more so than many other more traditional engineering projects.

A Product Mindset combined with Agile works well here as in the vast majority of software endeavors in the vast majority of cases. Implementing AI in particular with dated waterfall methodologies that view things through a “project” lens will make it harder to get to success. So much experimentation as well as trial-and-error is required with AI that trying to preplan everything against FIXED scope, budget and timeline is at odds with the learning and tuning inherent to AI.

I think that also in this field probability the agile methodology is the best one. I think that that the water fool is to heavy to implement. What is necessary is to teach to the teams the best ways to implement the new tools 🧰

Consider which prompts you intend AI to answer. Engineer these into priorities and variants. Ensure that the LLM's are grounded only in these or else do not answer. Citation is a critical feature for any enterprise to ensure the text answers are truthful and verifiable. User roles should be respected to ensure that content is shown to those who are entitled to view it. More advanced indexes are capable of carrying over entitlements/permission into search results used to feed the LLM - thereby ensuring privacy and governance.

To AI-power your enterprise software project, define specific AI goals in line with project objectives. Identify areas for automation, analytics enhancement, or improved user experience. Choose AI tools like machine learning algorithms or natural language processing that align with project requirements. Ensure seamless integration without unnecessary complexity. Adopt an agile implementation, breaking down features for continuous testing and refinement. Regularly gather feedback for iterative improvements. This strategic approach ensures your enterprise software effectively leverages AI features, optimizing performance and user satisfaction. 🚀✨

What if you created great features that no one knew about?? More than code and testing, articulate value and commitment to AI feature delivery that matters most! Implement features that solve a specific problem, make a center of excellence on the business agile capabilities. Involve the business was and end users in the continuous improvement test, implement life cycle.

And, don't forget to monitor and tune your AI/ML models for their continued accuracy after they are deployed in production and used on new data.

Nothing changed in terms of software engineering, component implementation, but AI allows us to create, test and deploy features faster, providing value by itself. Also, using AI powered tools for SW development provides a leverage, simplifying and speeding up the process, creating a value and efficiency by the fact of adoption of such tools.

This seems like a no brainer but most people use AI as a Google search or like machine learning. It needs to learn not accept what everyone else adds to it. The better a company is at creating the prompts that works with how a real human interacts, then integrates with their software will have massive success. But the results are where it’s at. If it’s not proven. If it’s not easy (from the outside), etc these are all results driven.

I find measuring results to be easier the better job that is done establishing the initial goal/objective. Make sure you clearly understand the business objective you are trying to solve, how AI helps achieve those results, and how you can measure each contributing factor. Evaluating results should start with, did you achieve the desired business result. If so, how did AI contribute? If not, was AI part of the problem? If you have clearly defined the objectives and how you will measure contributing factors, the evaluation becomes much easier.

In order to evaluate the AI outcomes we need to consider different horizontal perspectives: the short and the long. As I told before probability now it is a little premature

Starting with outcomes and KPIs - and having a record of "Before" AI is implemented as a baseline and then measuring the same metrics for the function "After" AI-powering the function is something that helps maintain discipline in my view.

The goals you set out the outset, business-outcome oriented should form the basis of what you evaluate against. Are you making a positive impact or not. Does the benefit outweigh the cost?

How do we know we are successfully implementing AI with results? Tell the story , show the results. A smart approach to evaluating and standardizing success oriented AI. We had this problem. Our end users wanted seamless efficiency. But recent growth made this a challenge. Describe the solution. Implemented ChatBot, predictive intelligence and self help for in the fly fixes. Specific makeover results: “before it took x days”. “Now it takes minutes” Get better at telling stories where people relate and ideate to evolve the best service solutions.

Train - Monitor - Feedback and loop is the best way of making any AI application to get better results and become robust. Evaluating AI results is very important to understand the outcome of it on the business and ensure that it's not hallucinating with its model data and producing wrong results/data to the users.

To AI-power your enterprise software project in evaluating AI results, establish clear goals aligning with project objectives. Implement AI features such as automation, analytics enhancement, or improved user experience. Choose fitting AI tools like machine learning algorithms or natural language processing. Ensure seamless integration and avoid unnecessary complexity. Adopt an agile approach, breaking down features for iterative testing and refinement. Regularly assess AI results, gathering feedback for continuous improvement. This strategic process guarantees your enterprise software effectively leverages AI, optimizing outcomes and user satisfaction.

Evaluation is key. But keep in mind, everything needs its time. You might not see the desired results immediately. Patience, work, perseverance and you might get the expected results

A powerful search platform used to underpin generative content should provide ability to measure, test new ideas and examine results. When it comes to text based service, enterprise search and Generative AI are inseparable relatives. By separating the two one stands the risk of delivering silos with conflicting messages as well as potentially encountering duplicate costs. LLM behaviour should be directed by semantic results from first party data.

Eat sleep AI repeat.... You have to learn and correct bifurcating foundational AI and Gen AI Automate ----> Differentiate ====> Deliver

Interesting article, but the journey to adopting AI will be very different for every business domain. AI is new, and rapidly changing. Agile is the way to go. The "water fool" approach. Is not a typo. Only a fool, or a government would adopt a waterfall approach. My advice is to embrace workflow and data management to verify process and data integrity. Automation is only useful when it truly adds value. How will automation move a KPI. Can you prove it? Would you bet your job on it?

Get lots of feedback and ask everyone for advice and improvement suggestions. You should also get feedback from people outside of the project. Objective opinion is worth the gold nuggets

Once you have implemented your AI features, the next step is to evaluate the results they produce. This involves measuring, analyzing, and improving the outcomes against various metrics and methods. It is important to collect feedback from software users, customers, stakeholders, and partners to gain insights into how well the AI is delivering value and whether it aligns with the original goals. By continuously evaluating and refining AI results, you can enhance the overall effectiveness of your AI-powered software project.

I see too often companies repeatedly making the same mistakes of focusing on the next new technology or platform, like A.I., instead of accurately describing the "As-Is" architecture of the organization's enterprise business operational architecture and supporting technology resources. Often, what they think comprises enterprise is shown to be wildly inaccurate after an objective and proper enterprise architecture assessment is accomplished. It exposes and identifies the gaps and the opportunities where A.I. or any technology can bring value, whether it be changes to operation processes or the technology that supports the business processes.

AI relies on data. So think through your data collection approach including ongoing costs and quality assurance. Pay attention to ownership - of data and of the trained AI.

Even at a slow space, I think the world is moving towards to AI. So it depends on few business aspects such as, the use of software, the budgeting and maintenance, understanding of the AI technology. Or else AI also has a negative impact on the businesses.

No tool is a silver bullet. Start with People (customers, internal employees) and Process and decide where AI can help you scale!

AI solutions will only be as good as the data it has access to and make sure you create an AI solution in your own environment to make sure you own the data or solution the AI solution creates. Remember, AI is a new very powerful tool, so make sure you perform up-front research to select the correct AI platform to support your project.

Don’t fall in the trap of believing you need AI capabilities in every IT or business strategy. Assess your objectives for the project and desired outcomes, if AI powered insights make sense, then do it, but don’t try to force AI everywhere. AI/ML models and insights have their benefits, but are not magic pills for everything. Not even needed all the time and won’t give you a competitive or comparative advantage always.

Testing your results is critical. When AI tools go off track, which they can easily do, results that looks good on the surface can be completely incorrect, missing context, etc.

Business Change Management if the AI tool requires adoption is critical. Answering FAQs upfront such as how does this change my role, how can I provide feedback, what does success look like and so on will be key to a successful implementation

AI is a powerful tool, but it's just that – a tool. By strategically embedding AI where it adds genuine value, you'll create software that's not just intelligent, but truly transformative for your business. Identify Opportunities: Analyze core processes for tasks ripe for automation or augmentation. Customer service bots, predictive maintenance, data-driven insights are areas for AI's magic touch. Choose the Right Tool: Assess existing AI tools and libraries like NLP frameworks, data analysis platforms, or pre-trained models that solve your specific problems. This saves time and resources. Build with Data in Mind: Data is the fuel for AI engines. Ensure clean, organized, and accessible data pipelines to nourish your models.

In my experience, I have noted that utilizing an agile approach in project implementation is quite beneficial. It enables continuous testing and improvement. The AI goals should also align with the project goals