Written by As a product manager, the new buzzword is “do AI”. But what exactly does it mean? You need to trust the data you’ve collected and using. When execs ask – can we say we are doing AI? Well yes we do, but if you’re going to stake your job on the data that’s being used, you want to make sure you’re giving it good quality data. This draws blank stares. If you want to avoid people jumping into “solution” mode, the DASUD model can help.
D = Design
At the crux of every problem, is its definition. When you can define what success means, then you’re better able to vet your defined outcome. Should this change along the way, you can point to the goal you were trying to hit.
But what does success actually look like? Let’s take a bank trying to complete a home loan.
- Is it the number of phone calls the call centre takes to fulfill the loan?
- Or is it the number of loans that are approved, kept for 12 months and not been churned?
Get SPECIFIC. Each area may have different ideas of what success looks like. Often, by mapping the entire value chain, you can collect different data points you need, to show that you’re successful. And just because a metric goes up, doesn’t mean it’s adding value. As always – get specific. Then work out what data you need. As product managers, you already know this part.
A = Acquire
When it comes time to gather your data, make sure you have a specific reason for its collection. Then, you need to ensure that the data you’re using is of high quality. What does that mean exactly?
There are 6 main dimensions of Data Quality:
- Complete
- Accurate
- Consistent
- Timely
- Unique
- Validity
In established organisations, there will be teams completing these checks. This is usually done either through your data management team or your customer success team. If it’s not being done, then you can advocate for such measures to be used.
However, when you’re in a startup or a much smaller team, the responsibility may fall on you. You can use DASUD if you want to avoid compliance issues. It will help you determine what the steps should be taken to strengthen your product launch.
In a phased approach, I always recommend starting with Complete, Accurate and Consistent. By profiling (analysing) your data, you will quickly answer:
- Is a value there in the field? (Complete)
- Does it match what I’m expecting? (Accurate)
- Is it the same every time I get it? Or is it the same as it travels through the various systems? (Consistent)
By asking these simple questions, you’re putting yourself ahead of 40% of most teams that collect “everything”. You will also want to create a way to capture data quality issues. These issues are errors in the data. Then have them resolved as close to the source as possible. For example, don’t ask people to enter their address in a free text field. Instead, make them select it from a drop-down list from the National Global Address File (NGAF, available from Aus Post).
S = Store
Always make sure you store your data safely. Regardless of your company size, storing it on removable media (USB sticks/ portable HDD etc) is not recommended. But if you absolutely must, please apply a robust password. And no, “password123” is not a good combination. Check out some of the worst ones (and make sure you’re not using one of them…)
If you’re in a Microsoft environment, make sure you have it in the right SharePoint folder and lock it down. This will also take care of your version history so you don’t need to name your file as “customer_data_final_FINAL_USETHISONE.xls”
The other concept you need to go through is Role Based Access Control (RBAC). (Note: I didn’t cover this in my talk, but wanted to add value).
When deciding who should have access, you can go through the Create Read Update Delete (CRUD) matrix. By understanding what role can have create access or read access, you can prevent errors. Ensure you also differentiate between update and delete access. To extend upon the SharePoint example, create a user group called “Create.” Then, assign the appropriate users into this bracket. This reduces the administrative overhead of user access maintenance.
U = Use
When it comes to using the data, you may have heard of the “pub” test. But have you heard of the “creep” test? If you can identify something with your data and it makes you feel creeped out, you should probably avoid doing it. There is a (new to me) concept being floated – creating junk data that mimics the attributes of your data. So should it ever get leaked, the actual data isn’t lost, just the fake data. Because you need to remember, depending on your context, a data breach can mean different things. It’s one thing to lose someone’s shopping habits and preferences. Even if they’re into something “odd,” everyone is “odd.” We just don’t want to publicly acknowledge it. But a breach in say, the medical world, means something entirely different. Each entry on that line could represent an actual human being. It could include their entire DNA, for example. Do the right thing. Even if no one is watching. The future thanks you for it.
D = Delete
My gosh, people, when you collect your data, have an end date to it. And document it down to a yes/no condition. In the medical world, you needed to keep any research data associated to radioactive treatments for 50 years. What does that mean? No one on that current project is probably going to be alive when it comes time to delete it. Don’t delay dealing with the data. Write out all the conditions required to get rid of it. And if they’re met, then they need to be removed, make it that plain and simple. Deleting things is a lot easier than expected. If you must keep it – consider whether this is going to add unnecessary noise to your data. Perhaps you can create and utilise junk data. Then you won’t be too concerned if it needs to be deleted.
The overall takeaway
AI is strongly dependent on good quality and trusted data. DASUD is one way to build trust in what you have collected. By applying this quick and simple framework, you can have good governance around your product data.
Nige’s Note: I was surprised and honoured to have won enough people’s confidence to give my talk on the day. The joy was short lived unfortunately. When updating my mentors and supporters about my win, I learned tragic news. A friend of mine, similar in age to my father, lost his short battle to cancer the previous night. His kindness in always asking about my father’s health was remarkable. He also inquired about how I was doing. This happened while he was undergoing his own chemotherapy. His actions gave me immense strength and resilience. Pete, this speech is dedicated to you. I hope it helps make the world better, as you have made mine better, in the short time we got to know each other.
Nigel D'Souza 