by Johannes Humbert – LinkedIn | Twitter

The basics of programme design for AI-based software development. Don't build white elephants!

What can be done to make purchasing more successful is what we looked at in Part 2. Now it's about what to look out for in order to implement AI projects successfully.

The most effective AI projects are scientific in nature. This hypothesis from Part 1 can be used to explain why it is still the case today that many AI projects (reportedly around 80% according to IBM) do not get beyond the proof of concept (POC) stage.

Two things in particular can be done to ensure that an AI programme is not reminiscent of "Jugend forscht", but embodies the construction of a solution with the help of one of the most relevant technologies of our time.

Do not make the use case mistake

Use cases are vehicles for consultants and salespeople to sell a technology to people who understand nothing or not enough about this technology.

However, a good use case is not a story on a slide that sounds good, but a coherent interplay of individual solution needs, technological architecture and good fit for the target process.

What worked well for another company does not necessarily have to work the same way again. And even if that would be the case, it is not said that a different solution would not have brought a higher ROI.

So take a look behind the curtain!

The easiest way to do this is to define your divisional or corporate goals and overlay them on your existing data in a feasibility analysis. This requires specific experience in building AI solutions. The questions that need to be answered are: "Which use cases pay maximum dividends towards your goals?", "With which algorithms can these be realized?". At the end, you receive an ROI-driven evaluation of use cases and can make well-founded prioritizations.

In addition, concrete expected values and benchmarks can be derived in this way. A fact that will be of decisive importance later in the course of the project when it comes to determining what is successful and what is not.

On the other hand, from a bottom-up driven feasibility analysis of your data, you will get a clear picture of which AI applications will maximize your goals.

As a result, one or two use cases may turn out to be less easily feasible or impactful, while other applications that seem less obvious promise significant added value with manageable effort.

In any case, it is worth questioning the ratio of the business units and ensuring in this relatively low-cost analysis step what the real drivers behind your business are and how an effective AI programme could be designed.

Don’t buy a pig in a poke!

Every AI project harbours technical risks.

Data must be available and of a certain quality, use cases must be chosen correctly, assumptions from preliminary analyses must be confirmed, information must be integrated back into the legacy systems and processes must be adapted. In addition, employees must be ready to interact with the new systems.

These are just a few points, and they cannot be argued away. However, they can be handled quite straightforwardly.

Define project phases with corresponding result expectations and buy them in as individual building blocks one after the other, for example:

Preliminary analysis -> Proof of concept -> Implementation -> Operation.

It is imperative that each use case is underpinned by a corresponding business case and a technical assessment in the sense of a preliminary analysis. It is tempting to let projects run once they have been started. Especially in view of a possibly lacking error culture in the company, which equates every well-founded project termination with failure.

It is therefore important, not only with regard to the result, but also for the careful use of political capital and for acceptance in the company, that a transparent understanding in hard figures is created in advance of what is to be achieved and what happens if these benchmarks are not met.

In this way you eliminate the risk to both your company and your position and create an incentive for your service provider to only do things and pursue project strands that actually pay off in terms of project success.

Conclusion

Many projects stumble in the process because they were not set up properly at the beginning. Success stories can be written with future technologies such as artificial intelligence through a clean definition of goals and feasibility analysis, the distribution of risk and a step-by-step commercial approach.

Previous Post