At the bottom of this post, you’ll find the Yusp “Build or Buy” questionnaire. Based on the answers, we’ll calculate which option works best for you, and send you the result in an email.
To insource or to outsource? The dilemma is as old as the business functions it concerns. The same applies to SaaS (software-as-a-service) solutions, and to recommendation systems in particular. Ever since the emergence of personalization vendors offering recommendation systems on an SaaS basis, the temptation to build similar solutions in-house has also been present.
In the software business, it’s called the “build or buy” dilemma. It’s complicated: pros and cons abound, and the right choice depends on many factors – some universal, others industry-specific. Your best bet is to find out about your options as much as you can, before making an informed decision.
In this article, we’ll look at the upsides and the flipsides of building or buying a recommendation system. We’ll also examine some other options in between. We’ll tick off a long checklist of factors to consider before taking the leap. We’ll explain why businesses go from buying to building their recommendation systems, or the other way around.
The “Build or Buy” Dilemma in Close-Up
If you want to use a recommendation system for your e-business, you can choose from a variety of SaaS personalization solutions offered by specialized vendors like Yusp. Alternatively, you can opt for building your own recommendation system from scratch. Hybrid solutions also exist, like building the platform in-house but incorporating “prefab” elements provided by personalization providers. (Case in point: Yusp’s standalone deep learning module .)
Buying: Low Cost, Limited Control
In the context of personalization, buying means signing a contract with a provider who will set up, launch, and run a recommendation system based on your business requirements. (Read more about the implementation process here. You don’t actually buy the personalization technology – the software and the infrastructure is owned and operated by the provider. Instead, you subscribe to the service for a monthly fee. In case of Yusp, this fee is calculated based on the added value generated by the system’s recommendations, ensuring return on investment.
Buying is an option that’s easy to love because it has some tangible benefits:
- Speed. Time to market is as short as it gets. After the initial design and testing phase carried out by the vendor, you can start displaying personalized recommendations on your website in a matter of weeks.
- Cost. Buying requires significantly less upfront investment than building, and its running cost is also more manageable. The personalization engine’s monthly fee is lightweight compared to the financial burden of hiring an entire product team and constructing scalable infrastructure.
- Flexibility. The SaaS solution can adapt quickly to the latest technological advances in the personalization industry. For instance, if a new, deep learning-enhanced algorithm is out, there’s no need to ditch your current recommendation system, you can just upgrade it for an adjusted fee.
However, buying also has a distinct set of drawbacks.
- Control limits. Because you don’t actually own the recommendation system, you’re in the back seat. Product development and general strategic direction is going to be in line with the provider’s priorities, which don’t necessarily overlap with yours. For instance, if the personalization vendor of your choice serves clients predominantly in another industry, its optimization efforts may be geared towards issues that are not relevant for your business. (Here at Yusp, though, we have a reputation for being flexible. If a client requests an extra element or a custom feature, we’re ready to work something out – for a fee that stays well below the price of in-house development.)
- Risk. You know what it’s like when you team up with a third party: no amount of careful vetting is going to guarantee that they won’t go unexpectedly bankrupt, or merge with another company and undergo a total, permanent profile change. (One mitigating factor is the abundance of personalization vendors on the market. Selecting a shortlist of potential providers and A/B testing your top choices helps to curb the risk.)
- Redundancy. Since recommendation systems available on an SaaS basis are, to some extent, ready-made, they’re not quite a perfect fit. They might have certain functions you don’t need. (Although, in the case of Yusp, you only have to pay for the elements you actually use.)
- Transparency trade-off. Because you don’t own the IP (intellectual property) of the personalization engine, you can’t look under the bonnet, to access the source code. This way, you may not be aware of the system’s full potential, and you can’t cash in on it – for instance, in the form of upgrades that would further boost your revenue.
Building: Perfect Fit at a Price
In the “build or buy” dichotomy, building means using internal resources (a team of data scientists and machine learning engineers, either created just for this purpose or branching out from other data mining activities) to assess business requirements and to devise the appropriate personalization solutions, while also creating and scaling the necessary infrastructure. Ideally, it includes integrating the recommendation system into the existing tech stack, plus continuous product development and business support.
In light of the “Buy” pros and cons above, it’s plain to see why building your own recommendation system can seem tempting. Its two undeniable advantages:
- Fit. No two businesses are exactly the same, and nobody knows your datasets, your users, your challenges better than you. If you build your own recommendation system, you can be sure that it’s fully customized for your use case.
- Control. The personalization engine you build is yours to experiment with, and to tweak according to your evolving needs. In case you want to extend it, or to diversify its array of functions to serve new purposes, you’re free to do so.
But beware: such freedom comes at a hefty price. The risks and disadvantages of building include
- Flying blind. Unless you’re in the business of personalization (in which case, of course you’re building recommendation solutions), chances are, your expertise lies elsewhere. In order to build your own recommendation system, you need to venture out of your comfort zone and acquire new know-how.
- Time to market. Building a recommendation system from scratch can take anywhere from months to years before it’s ready to be deployed. In terms of speed, building definitely can’t compete with the near-instant launch of an SaaS personalization engine.
- Expense. Initial investment and running cost are both likely to be much higher than when you buy an SaaS solution. Human resources are probably going to be the biggest item on your list of expenses – top-notch data and tech professionals are a luxury few companies can afford.
- Dependence on expertise. As good data scientists and machine learning engineers are in high demand, you run the risk of losing them to a competitor. Without proper replacement for these key figures, your in-house personalization project could get permanently stalled.
- Benchmark issues. Building a recommendation system to help you achieve your personalization goals is only part of the challenge. How can you tell if it meets global industry standards? Without being firmly based in the recsys community, it’s tricky to compare your product to current solutions in order to make sure it’s up to scratch.
- Staying competitive. Keeping abreast of the state-of-the-art, that is, the latest advances in recommendation technology is an ongoing effort, and one that consumes a lot of resources. For a personalization vendor, this constant optimization is a core function, while an in-house team might perceive it as a tedious chore.
Factors to Consider When Facing the “Build or Buy” Dilemma
Weighing up the pros and cons of both options is only the first step when pondering the “build or buy” question. Evaluating your circumstances and formulating your preferences is the key to finding the right direction for your business.
Start off by establishing your stance on the general factors:
- Time to market. When buying, it’s close to zero, which minimizes risk; when building, it’s considerably longer. How much time are you willing to invest in your personalization solution?
- Initial investment. High when building, much lower when buying. In broad terms, what’s your upfront budget for kicking a personalization engine into gear?
- Running cost. In case of building, this includes the salary and overheads of the complete product team and any hardware or infrastructure investment; in case of buying, it’s the vendor’s monthly fee (plus any associated internal costs). The latter is bound to be lower. Are you ready to allocate a bigger budget for running costs in return for a more tailored recommendation system?
- Added value. In the short term, buying SaaS wins hands down. As for mid-term, the jury’s out: maybe your in-house recommendation system will reach the efficiency of an SaaS package, and maybe not. In the long run, however, the benefits of building start to show: depending on the size of your company, the added value of even a minor adjustment can be huge.
Would you prefer to generate added value as soon as possible, or are you taking the long view?
- Risk. Buying keeps the risks associated with time to market and total cost to a minimum. Building is more of a gamble: even if you dedicate a whole lot of time and resources to an internal recommendation system, there’s no guarantee that it will work in an ROI-positive way. How risk-averse are you in this regard?
Next, look at the factors linked to the specifics of your business and your industry.
- Company size. When it comes to the “build or buy” decision, size really does matter. The bigger a company – meaning more users and more data – the more inclined it is to consider building a recommendation system. But realistically, building is a viable option only for a handful of the largest enterprises, those who can afford the necessary human resources. For everyone else, it simply makes more sense to buy a proven SaaS solution.
Which category does your company fall into?
- Technological roadmap. The “build or buy” choice is also shaped by where you are in the process of adopting technology (with data at its core). If you have a long history of collecting data and amassing knowledge about your users, and you’re already working with data scientists or AI experts on various elements of your tech stack, then you could look into building your own recommendation system. If you have other priorities to take care of before upping your personalization game, it’s perfectly OK to go with an SaaS solution.
Where would you place your business on that progress bar?
- Available technology. If your field is well covered by personalization providers, that’s a strong case for buying. If you’re in a niche domain or in an industry that is difficult to serve by standard personalization solutions, then building could be worth considering.
Are SaaS recommendation systems easy to come by in your sector?
- Specific datasets. Whether to build or buy a recommendation system also depends on the variety and specificity of the data your business generates. If you’re in e-retail, for example, the clicks, views, and add-to-carts your users produce are fairly straightforward to work with, so most SaaS personalization engines can do the trick. But if you’re, say, a travel aggregator, you’re churning out atypical data – endless combinations of flight bookings, hotel reservations, car rentals, each with their own metrics. In this case, a standard personalization solution is unlikely to work for you, and building a customized recommendation system could yield greater benefits.
Does your line of business lend itself to personalization?
- Personalization use cases. The majority of e-businesses can implement a well-rounded personalization strategy using the standard offerings of an SaaS solution – recommendation widgets, personalized search, on-site personalization, and so on. Trickier business rules call for a custom-built solution.
How unique are your personalization use cases?
For a more hands-on approach to reaching a decision, fill out our “Build or Buy” questionnaire. Based on your responses, we’ll calculate which option works best for you, and email you the result.
Why “Build or Buy” Is Not a Permanent Choice
As we’ve seen, the “build or buy” dilemma is a complex topic that requires careful judgment. The implications of choosing either option can have a long-lasting impact on your bottom line. This is not to say that “build or buy” is a once-in-a-lifetime decision. On the contrary: it’s quite common that businesses change course, some several times, going from buying a personalization solution to building their own, and back again.
In a typical “buy to build” scenario, a company, in contract with a personalization provider, starts to experiment with building a recommendation system in-house. They might be curious about tweaking algorithms, or reluctant to part with sensitive data, or in some way encumbered by having to adapt to a third-party vendor. Once the management is confident enough, they pull the plug and switch from the SaaS solution to using their own personalization engine.
The opposite case, “build to buy”, is perhaps less well-known, but – at least in our experience – it occurs far more often. Typically, A/B tests are the pivotal points, when a company decides to compare its home-grown personalization solution to a SaaS product, and comes to the conclusion that the latter performs better. According to our internal log of A/B tests, there have been 22 instances when Gravity R&D’s personalization engine, Yusp outperformed in-house solutions, and not a single time it was defeated.
The main reason why in-house recommendation systems are abandoned is their failure to scale. Another possible explanation may be that their algorithm portfolio features “mainstream” elements like collaborative filtering or rule-based algorithms, but not the latest technology, like deep learning, which gives personalization specialists a competitive edge. In either case, failed A/B tests convert builders back into buyers.
Alternative Solutions on the “Build or Buy” Spectrum
Not only is “build or buy” far from being a final decision; it’s not black or white, either. It doesn’t have to be one option or the other; there are alternative solutions between building and buying a personalization platform. This makes it possible for you to finetune your preferences in terms of investment, ownership, and control.
Let’s look at these alternatives in more detail.
- On-Premise. Just like with SaaS, you don’t actually buy the personalization software. However, you own the hardware, and you do the heavy lifting: installing and running the software, as well as its maintenance. This arrangement is worth considering if your data is of the sensitive kind and you’d rather not let it leave your premises.
- Managed appliance. In this hybrid structure, the hardware is yours, the software is owned by the personalization vendor, and they are in charge of installing, operation, and maintenance. This option is also geared towards data security, while being less labor-intensive for the client.
- BYOA (Bring Your Own Algorithm). In this case, the framework of the recommendation system is provided by the vendor, but you get to build the moving parts, that is, the actual algorithms (or some of them). This is a great midway solution if you’re confident in your knowledge of your customers and data, and you feel that only a customized solution will do them justice. You don’t have to worry about integration; your vendor will take care of that. The one caveat is that you share responsibility with the vendor for the seamless functioning of the system. If there’s a hitch, figuring out whether the framework or the algorithm is to blame might lead to some tension.
- Buy the code. If you already have a personalization framework in place and all that’s missing is a cutting-edge algorithm, you can buy that as a standalone product, complete with source code. This means that the algorithm will be yours to use or to tweak to better suit your objectives. This approach combines the pros of buying (speed, cost, flexibility) with those of building (fit and control).
“Buy the code” offerings are quite rare because algorithms are the “secret sauce” of personalization vendors and they’re reluctant to part with them. However, Yusp has made its deep learning algorithms available this way.
Whether You Choose to Build or to Buy, We Can Help
We’ve looked at the build or buy dilemma in detail: all the pros and cons; all the factors to consider before making a decision; the landscape of industry trends; and the various nuances of the build-buy spectrum. All this should provide a good base for an informed decision. Remember: when it comes to recommendation systems, whether you choose to buy one or to build one, or something in between, there are ways we can support you. Get in touch for more information.
