At Pull we’re on a mission. Our clients are typically mid-size businesses (£10m-£100m t/o) for whom AI is not on their radar as a ‘must do’ any time soon. However, the chances are these are the same size companies that didn’t have the internet down as a must-do at the turn of the millennium. So, we feel we owe it to our clients to acquaint them with the potential of all emerging technologies, and especially AI.
However, sometimes it can feel like pushing on a rope. Why?
Pull runs AI Workshops for clients and friends of Pull with our data science partners. It’s always a great day out and we all leave buzzing with ideas. The sessions help demystify a complex subject matter and focuses on solving clients’ business problems, not technology for technology’s sake. So why are the AI opportunities so difficult to see for businesses of this size? I think there are three key reasons:
- Little or no internal knowledge of the field, so there’s no one to even identify the opportunities for AI. The danger here is forward-thinking brands in the sector will quickly gain competitive edge.
- Most of the case studies available are for global brands with deep pockets. They have the budgets, time, resource and partners to experiment.
- AI projects have a high failure rate.
Regarding the failure rate, my hunch is this is caused by a mixture of lack of experience within the teams implementing the projects, the complexity of projects chosen, and the related high expectations.
Perhaps we just need to be realistic. Hence, I have coined the phrase ‘Practical AI’ and come up with the ‘A Practical Guide to AI for Mid-Size Companies’.
Step 1 - Get your company up to speed with AI
Get key decision makers on board early from across the business. AI should not just be considered as an IT project. Most likely all areas of the business could benefit from being involved in the process. In the near future, AI will be fundamental to many complete operations, don’t get left behind.
Task someone to be your data and AI ‘Project Owner’. They need to lead the process and will be key in accessing the skillsets and expertise you need to be successful, be this internally, or most likely to start with, as a blended approach with partners. The Project Owner needs to ensure everything is measured and things don’t get out of control. Remember that high failure rate, track and measure everything!
Step 2 - Build your AI team or find a ‘right-size’ partner
For a mid-sized brand dipping their toe into AI this might not be the right time to build what will be a multi-disciplinary team of potentially 2-3 people. The business probably needs to get a base understanding of AI and the opportunities before it builds out its own capability, especially when most likely initially you are looking to implement a relatively simple solution.
So, find a ‘practical’ partner(s) who have experience of implementing data and AI projects for similar size brands to yours. There are multiple companies who now have real-world experience of implementing successful AI initiatives, and will understand the challenges of this for mid-sized brands. Bringing the partners in early on in the process would be useful, but realistically all of this process could be done independently.
Step 3 - Make sure you are solving the right problems
Understand what you want to do with AI and why. Start with key business challenges, not with the technology. Build the business case.
But if AI can’t fix key challenges, don’t give up, look at areas such as operations for opportunities to gain business efficiencies using automation. These smaller use cases may actually be more suitable for your first initiative.
In fact, going back to my point around overcomplexity of projects potentially leading to failure, perhaps your first AI project should be looking to automate a simple process rather than trying to re-invent your whole business!
Step 4 – Identify and access your data sources in preparation for AI
This can actually be one of the most challenging pieces of the jigsaw, AI is only as good as the data it’s fed. Based on the problems you want to solve, you need to identify the data you need, make sure it’s clean, organized and in good shape for AI to succeed and that it can be easily accessed. You might get a bit of push-back from IT here, but data is owned by the business, as long as the data is respected and secure, access should be flexible to maximise its value. Just a thought here, data also means images, PDFs, Word documents, or sound and video files.
Step 5 – Do something with AI
If you have followed the four points above it is straightforward now. You have identified a problem AI can solve, you can get to the data, and that data is in a reasonable state, and you have the team or partners in place, so you can now start to activate using AI.
Typically you would build a proof of concept first to ensure the scope and scale of your ambitions are achievable. If the proof of concept goes well, you would build-out and implement your model in a similar way to how any technology project would work with system and user testing.
Before long, you should hopefully be deriving results and value from the work, be that in more efficient process delivery, or analysis of customer data to predict churn or personalise an experience.
My number one message for mid-sized brands is to not ignore AI and start your journey today. As well as the relatively complex custom projects the above process outlines, there are a tonne of free out-of-the box tools available from the likes of Google, Microsoft and Facebook for anything from facial recognition and anomaly detection to sentiment analysis, and even an automated Q&A maker for quick and easy knowledge-base creation.
For Pull, we chose the Microsoft route less than two years ago. Helped along the way by the leading agency Cloud Services Provider, Wirehive, we haven’t looked back. Initially we used Microsoft pre-built Cognitive Services for sentiment and video analysis, before building our first conversational assistant on the Bot Framework. We are now partnered with data scientists The Data Analysis Bureau working on more complex projects using greater amounts of data and custom machine learning models.
All of this culminated in a recently launched project for bespoke online suitmakers The Drop. In partnership with Microsoft, we developed an AI solution specifically to detect anomalies in measurements.
So, if a relatively small creative agency can get this far in two years, what’s stopping you?