Is Your Business Ready For Data Science? Five Questions To
Source – https://minutehack.com/
How to make data work for your small business.
There’s an appetite for data science amongst businesses of all sizes. While buzz is one thing, the fact remains that very few – especially SMEs – have actually employed it to capitalise on the competitive advantage it brings.
Some that do make the leap end up disillusioned if it doesn’t deliver what they’d hoped. There can be a number of reasons for this, but the main one is that the business simply isn’t ready for data science.
Let’s first be clear on where the value in data science lies. While reporting processes are great, they are generally limited to revealing only what has already happened. The real value in data science lies in being able to shift focus to what will or should happen and crucially, being able to take action to change the outcome.
Data science can help businesses build strategies that factor in revenue growth over time, rather than chasing short-term gains. However, before making any investment, it’s worth asking some key questions and replying honestly. Otherwise, any efforts can end up being a flirtation rather than a lasting relationship.
So, consider the following questions; each one that you can answer ‘Yes’ to puts you one step closer to being data science ready:
Do you have sufficient historical data across each of your data sources?
Much like the real world, we need to have enough context to predict whether our decisions will have a successful outcome or not.
Training a data science model is somewhat like training a puppy. ‘Good behaviour’ – in this case accuracy and performance – should be rewarded to encourage repetition.
Historical behavioural data and outcome data are the only treats a machine learning model needs, the more it has and by learning the relationship between the two, it gets better at making predictions.
However, the data used for training the algorithm must be representative over time in order that it can discern ‘business as usual’. So, data gathered over the course of the pandemic won’t offer a meaningful basis for data science. If all you have is anomalous data, the goalposts are constantly changing.
Also, the devil is in the detail so ensure your CRM system doesn’t overwrite point-in-time customer data to provide just the current snapshot of your customer database.
The same is true for stock data – you need to be aware of stock-outs if you want to distinguish between when there was ‘no stock’ versus ‘no demand’ as one example.
Is your data infrastructure solid?
Not all data is created equal. Making sense of it across all sources – both from the cloud and data that’s held in internal systems – means ingesting them all into a centralised data warehouse as a single point of truth to be used by all internal departments.
The expression ‘garbage in, garbage out’ holds true when we talk about data science models. Ensuring they can decipher data successfully requires ‘cleaning’ it by standardising formats, de-duplicating it and so-on.
Most importantly, there needs to be a common ID that makes it possible to join up data across different sources. For example, if we want to combine a customer’s transaction behaviour, browsing patterns and engagement with email marketing campaigns, we need an identifier.
Do you have access to expertise?
We often hear data is the new oil, but oil has limited use until it is refined. Extracting data’s true potential requires a willingness to invest in data scientists and technologies. Since data science is highly specialised, it cannot be palmed off on the marketing, or the IT team. It’s not fair on them and is very likely to doom any project to failure.
Any sources of inconsistency or friction points in the data will surface false positives and can lead to ‘data drift’, a scenario in which fundamental changes in the way customers behave over time can render models ‘stale’ and the performance degrades.
Consequently, both the data and the models need to be constantly monitored and occasionally retrained. This takes time, expertise and engineering behind the scenes.
Do you know what specific challenges you need to solve?
Advances in data science and machine learning are still a long way from ‘artificial general intelligence’. Their most successful use cases will have quite specific applications, where decision making processes are manual and available data is being under used in that process. In the simplest terms, you have to ask the right question(s).
Most typically, these applications centre around factors like lead generation and conversion, customer acquisition and retention, and stock control. So, data science has to be grounded in strategic thinking and ideally should be cumulative to build across departments and agreed business-wide goals. So, consider how you’ll be able to keep expanding on each set of results by asking the right follow-up questions.
It helps to adopt a ‘walk before you can run’ approach by focusing on the most pertinent problems or use cases first. If you have enough pointed use cases, these naturally come together to form a solution which delivers tangible value. Otherwise, you’re just doing data science for the sake of data science.
Do you have a data culture?
A data culture is one in which all key stakeholders within the business are comfortable enough to trust in the data and outputs from predictive algorithms. The stumbling block is typically internal politics.
Some teams jealously guard ‘their’ data. It’s perhaps ironic that data democracy requires everyone to get over the politics and recognise shared goals over individual or departmental ones.
Doing data science right requires time, investment and ongoing management, it often needs a change in perspective. A data culture must be led from the top and requires the Senior Leadership Team to recognise the benefits of full transparency from a shared data source. If you’re serious about data science. strong leadership is a prerequisite.
Data science is by no means a fad nor a nice-to-have, it will become increasingly business critical in a crowded market. If you are new to data science, the good news is that there is usually a lot of low-hanging fruit, you just need to know where to look for it.