How to Choose an AI Consultancy
What is an AI consultancy?
Artificial Intelligence (AI) is a topic that seemingly within a few short years has transitioned from being in the realms of science fiction to a powerfully transformative real-world technology domain with the potential to reshape nearly every aspect of our lives and society. From commuting with self-driving cars and smart algorithms that can anticipate our needs and send groceries before we order them, to advanced models that can plan and manage complex tasks such as designing houses or provide health insights using data from smart devices. The opportunities seem endless, but the competition is fierce, sparking an arms race amongst the big tech superpowers and Bluechip organisations who compete and offer competitive salaries to attract the high-demand-low-supply pool of data science talent needed to assemble their own internal data science and R&D teams. For businesses who are unable or uninterested in competing for talented data scientists but still want to leverage data and AI to automate and make smarter data-driven decisions, they are increasingly turning to AI consultancy partners.
An AI consultancy is responsible for leading and managing the end-to-end adoption of Artificial Intelligence for their clients. However, this brief explanation does not do justice to a complex and multi-faceted challenge which involves understanding the interrelation between distinct but interrelated factors, all of which influence and determine project feasibility and success. The range of factors that must be considered is extensive, ranging from “strategic challenges” such as:
- clarify business objective(s)
- define the nature of the business problem(s) and understanding the processes
- understand the process in which the problem occurs
- assess constraints or limitations owing to the nature of the available data
- identify factors that may contribute to project risk and possible mitigation tactics
- evaluate time, investment and personnel requirements for build and deployment
As well as “technical challenges” such as:
- identify viable solution approaches
- define metrics to test and validate these approaches
- tweak model parameters to optimise output quality
- deploy and enable interaction with the broader technology ecosystem
- implement data and model governance best practice for data hygiene, auditability and regulatory purposes.
If you’re not a data scientist, odds are many of the above challenges might not mean much to you. They don’t need to. Your responsibility isn’t to be the technical expert but rather the subject matter expert on all things relating to your business and help your AI consultancy partner understand what they need to know to overcome the business challenges.
So how should a business go about choosing an AI consultancy to collaborate with? We provide a few tips on the things you should be mindful of.
Select a consultancy that understands the problem and can fulfill end-to-end project requirements
One of the best ways to mitigate risk is to partner with an AI consultancy with strong experience in developing strategies and solutions to address the types of challenges your business is encountering. The consultancy should be able to provide example case studies on similar projects, communicate the similarities between the cases, articulate their solution approach and the outcome. These don’t need to be a carbon copy in terms of your industry, data or scenario, but should be representative of the nature of the problem. For example, if your aim is to identify and extract information from image-based data then working with a consultancy able to demonstrate a clear competency in the design and training of computer vision and supervised machine learning models is a great starting point.
Another way to mitigate risk is to select a partner who can manage the end-to-end journey. In terms of real business value, an AI strategy isn’t worth the paper its printed on if you are unable to implement the strategy. When assessing potential partners, be sure to clarify the scope of the consultancies’ services across strategy, build and implementation, and project after-care. If the consultancy is only able to advise on strategy without the capacity or competency to support with execution, you’ll either be left responsible for procuring an internal team or another AI consultancy to implement it, who might be unwilling to commit to the strategy, time, resource and investment commitments another consultancy has defined. Rather, look for a consultant partner who is an expert not only in AI, but many of the adjacent areas that connect to AI such as data management, data engineering, cloud and IoT and most importantly, can manage the end-to-end project requirements
Not every problem requires an AI solution.
The hype surrounding AI has made it one of the big business buzz words. However, there’s also a lot of confusion surrounding the term and few business leaders have a clear understanding of what AI is, how it works, how it will affect their business or what’s involved in its adoption.
AI is already capable of performing some incredible tasks and its potential is expanding rapidly. However, AI is not the best solution to every problem. This is because AI excels at performing highly specialised tasks by using large quantities of data to inform the AIs understanding of the problem and enable consistent and accurate response. Because of this, using AI in some situations may introduce unnecessary complexity and be excessive for the task; the technology equivalent of bringing a fire hose to a water fight. Not all tasks need a high degree of intelligence and engineering to perform them, nor justify the extensive requirements to develop this capability. In some instances, alternative solutions such as business process management (BPMS), robotic process automation (RPA), business process automation (BPA) or good old fashion general human intelligence may be the most viable approach.
Once the AI consultant understands your business challenge and blueprint, they should propose the best approach to addressing your needs and should communicate the best approach to solving the problem, not just a solution that uses AI.
Be transparent about the limitations and risk of adopting AI
When undertaking an AI project its important to be aware of risk factors and why they occur.
To understand what creates risk in AI projects it’s helpful to understand the difference between AI and traditional software. Traditional software is rule-based and prescriptive, involving providing instructions on how certain actions in a workflow should be taken under certain conditions, often referred to as ‘if this, then that’ logic. These rules or conditions are specified by humans and involves mapping out an end-to-end process and all potential scenarios and telling the machine how to behave in each: no thought, autonomy or ‘intelligence’ is involved on behalf of the machine.
AI is inherently less prescriptive as it requires machines to detect patterns and trends within a subset of data and then generalise these learnings to another data set to generate insights and/or take some action. Although humans may control the conditions – such as the model selected, data used and training parameters – this process is fundamentally experimental: requiring a hypothesis be formed, tested, and assessed against defined metrics and benchmarks that determine success, then iterating and refining that process to achieve output that is more accurate for the basis of informing decisions. This is in no small part what makes the process of developing and maintaining AI solutions time consuming and expensive; it’s far from straightforward.
These aspects of iterative experimentation and R&D are intrinsic to AI. They are also a big part of what creates risk in AI projects: until you’ve experimented with your data and validated something is possible, you can’t be certain of success. Your AI consultant partner should have tactics for managing this risk, with the simplest being prioritising those business problems/opportunities according to risk profile, based on the maturity of the solution domain and validated use-cases for similar applications. Another is by defining risk triggers and their corresponding responses, including the early termination of a project to avoid unnecessary cost. Finally, ensure effective planning prior to kicking off development or implementation of any AI in your business.
Prioritise planning before action.
Failing to plan is planning to fail. This is true of nearly any goal in life or business and is particularly relevant when it comes to data science and AI projects. Even challenges with well-established techniques or off-the-shelf (OTS) solutions to solve them and many validated use-cases doesn’t mean business leaders should be complacent or cut corners in the planning phase. Just because one or more companies have successfully leveraged AI to tackle a problem doesn’t mean your business will be able to replicate their success, or their approach aligns with your long-term business goals. For example, if your goal is to create intellectual property and subsequently licence it, using OTS solutions is not a viable approach. The determining factor in your success with AI will be tailoring the requirements to your “business blueprint” – the business objectives, data, processes, technology, and ways of working unique to your business.
Any AI consultancy worth their salt will know that the best way to reduce risk is to plan effectively. Although it might be tempting to skip this stage and ‘learn by doing’ or base a strategy on second-hand evidence, resist the temptation leap straight into solution mode. This shotgun approach of jumping straight into experimenting with different algorithms, model architectures and data configurations creates the risk of investing significant time, resources and budget in a project that effective planning may have revealed was not viable from the outset for several reasons: not technically feasible, absence of fundamental pre-requisites such as relevant data or infrastructure, pursuing a sub-standard approach to solving the problem, or does not align to the real business priorities. Finding out you’ve spent weeks or months testing and learning only to discover a project is not viable, which could have been identified early through effective due diligence, will shake confidence and discourage future opportunities from being pursued.
Written by Clayton Black