To Buy or to Build?
A simple framework to help you understand when to buy 3rd party data software and when should you build it yourself
Working in data is fascinating. Almost every company will claim its data driven, which gives data teams the unenviable task of defining their role in an organization when teams like Growth will use data to run experiments, CRM might want to buy AI tools to help with personalization or where your data insights might at times tell PM’s their product isn’t working.
Data teams can help guide the organizations decision to buy or build the above data products using a simple framework that considers resources, time and quality.
1. A question of resources ($$)
Ultimately, determining whether to buy or build is primarily a question of resources. Big, well-resourced companies, like Google or Facebook can afford to build everything, and it makes sense that they do. Armed with lots of cash, they have the resources to recruit the best talent at everything they do, so they can afford to build everything from scratch.
On the other side of the spectrum, a small 1-person startup, like I was initially as the founder of LMNS, you buy everything (or sign-up to a free trial of every software under the sun). Ultimately, it’s too much to try and do everything yourself, and it’s unlikely that you are an expert in every function of a startup. Additionally, unlike the Facebooks of the world, you likely don’t have access to the same talent pool. This means the decision to buy, or sign up for a free trial is pretty easy. If you aren’t a data scientist, it’s likely that the company that built a lead scoring model already does it better than you.
2. A question of time
But monetary resources aren’t the only thing guiding the decision. The other factor is time. A startup may have hired its first data scientist, and so, technically they have the resources now to build a lead scoring model. But how long will it take them?
Adding another axis to our matrix, when we have low resources, and building would take a long time, the decision to ‘Buy’ is pretty clear. From the previous example, if we don’t have the resources to build something in the first place, then again, ‘Buy’ also seems like the clear decision forward. If you can build something quickly, and can get the resources in place to do it, then the decision to ‘Build’ is pretty easy.
But, what about that last quadrant? What about if you have the resources in the team, but it might take a longer time, do you buy or build then? Also, can 1 individual build a better model than a software built for that purpose?
3. Quality
So far, we have talked about companies on either extreme of the spectrum with respect to size. Tiny vs. large, where it is easy using the 2 factors above to determine whether to buy or build. But this becomes much trickier for your Series A to B companies smack bang in the middle. Strapped with more cash resources than your startup, you can attract both better talent than a small startup, and you can also afford to buy a more expensive version of the software than the small startup. So what do you do now?
These are the companies whose decisions fall into that last quadrant of our matrix where the question mark is.
Let’s revisit the lead scoring model. There are countless tools and softwares out there that offer out of the box lead scoring models. You are now considering if it’s better to purchase the software or get your data scientist to do it. You could use the axis above to do a simple cost benefit analysis between the 2 options. You know how much your data scientist earns (resources $$) and you can ask them how long it would take to build (time). Multiplying those 2 together will give you the total cost to build. You can then compare that cost to the cost of the software (resources $$) for however long you will need them (time).
This is where the last factor, quality comes in. If both solutions can create an output of equal quality, you choose whichever one is cheapest.
If they don’t, you simply add the quality differential in as a cost. And you can understand the parameters of that cost by asking questions. For example, the leads scoring algorithms used by many software companies are logistic regressions. Most data scientist should feel comfortable building and deploying that.
Can I evaluate quality if I am not an expert?
However, just because you lead a team doesn’t mean you know the ins and outs of every algorithm. In that case, engage your team. In many situations, the choice of buying or building can cause tension between teams in a company. The marketing team wants to buy the software, but data feels they could do it themselves. In these situations, engage the teams and make it a joint decision. Grab your data scientist and put them in that meeting so that they become part of the decision.
What do you think? What guides your companies decision to buy vs. build?
