Should You Build or Buy Your Marketing Artificial Intelligence Solutions?

What’s your problem?

That’s the first thing I say when marketers ask me if they should build or buy artificial intelligence and machine learning (AI/ML) solutions.

Yes, I’m aware. I’m like an old-school Marlboro cigarette. Rugged. No filter.

But alas….

Determining whether to rent/buy or create-your-own AI solutions is not like walking into a Build-a-Bear Workshop. In my experience, the amount of money you shell out per minute is about the same however. Incidentally, those salespeople are very clever with their “do you love your bear? Do you want him to have a heart just like yours? It’s only a teensy-bit extra” charm directed to your tiny humans. As if anyone wants to give their little a heartless lovie. But I digress…

In-house vs. outside (vendor) is a big decision, and unlike a teddy bear that may be thrown to the wayside the minute your tiny tyrant sees the mall candy store or ice cream shop, your AI/ML solution will be around for quite some time. It’s essential you dot all your i’s and cross all your t’s. To help you make good choices (yes to hearts, no to pixie sticks), here are some questions that will help you determine the right choice for your business.

IDENTIFY YOUR PROBLEM

What’s the problem you are trying to solve?

How are you currently solving the problem?

Does this problem need artificial intelligence to solve it? Is the issue repetitive? Does it require automation? Does it involve data? How will the use of AI have a positive impact on the outcome(s)?

How valuable will solving the problem with artificial intelligence be to your future success? Are you looking to save money? Make money? Reduce headcount? Improve your customer experience? Something else? Are you counting on the AI to get better over time?

Is this a common problem, or is this unique to your company? If it’s a common problem, then other companies may have solved it. How do other companies solve the problem? Do they do it in-house or use an outside service/vendor? Even if you decide to build something in-house, you should research how your issue has been fixed/solved by others to draw inspiration and/or learn what you should (or shouldn’t) do.

Are there third-party vendors who offer solutions to your problem? Do you need to cobble together multiple outside solutions to solve your problem, or can one vendor do it all?

Is there any other way you can solve the problem?

Will solving this problem be a key differentiator in your success vis-à-vis your competitors?

How will you gauge success?

What are the ramifications of not solving the problem?

DETERMINE YOUR BUDGET

What’s your budget? What’s your budget for the initial build, and how much money can you spend to maintain it? What’s your cushion amount? Just in case the project goes over budget.

Are you able to spend all the money upfront, or are recurring payments better for you? (The answer to this question can be the dealbreaker for many companies.)

What economies of scale do you get by buying vs. building? One of the reasons companies buy AI tools, especially when they’re first starting with AI/ML, is because they want to take advantage of the vendor’s ability to distribute their maintenance costs over multiple clients. (This loosely translates to it can be more affordable, especially upfront.) If you are building, you may also get savings advantages by increased usage of existing resources.

What kind of guarantees are there? Are you able to test drive the solution before you commit? Do the vendors you’re looking at have FREE trials? Many AI vendors will make you deals you can’t refuse. This can be a catch-22 because they often do it because they need the clients/data to build their models. You get the deal, but you become the guinea pig. With that said, free trials and data audits can be BIG benefits, so be sure to identify who has what.

LIST YOUR AVAILABLE RESOURCES

How many humans do you need to build your solution in-house? Developers? Project Managers? Machine learning engineers? Data wizards? Testers/QA? Etc.

What other departments need to be involved in your project? Sales? Customer Service? Legal? How will they be looped in? Vendors often comment that questions like these should come down the road. Poor communication is one of the top reasons why AI projects fail. Figuring out the time and effort it will take to keep everyone in the loop often changes how you see a project. I wouldn’t skip it. Companies seem to be good at identifying the heavy lifters of their projects but not necessarily all the people who work behind the scenes to make things happen. Broadway shows are powered by a whole lot more than just the actors. Marketing AI projects are much the same.

Do you have the talent, experience, and expertise in-house today, or will you need to hire additional people? You’ll want to consider this for the initial project and the ongoing maintenance. Building AI solutions is similar to creating web solutions in the 1990s. The majority of people are spending oodles of money and doing subpar jobs. (I’m being generous with “subpar.”) Plus, the available articles/advice that’s out there is often theoretical and, unfortunately, not road-tested. When reviewing your resources, look at them objectively and plan to fill in the gaps where needed. For some projects, you won’t need full brainpower. For others, you’re going to require oodles of it AND people who have been in the trenches. As an aside, hiring AI talent can be challenging, depending on where you are and your budget/benefits. If you must hire people to make the project work, that’s a-ok, but please make sure to build the hiring process into the timeline.

How many people need to be involved if you use a third-party vendor instead of building something on your own?

When listing your available resources, you should list the titles, the amount of time they’ll need to allocate, and their relative salaries. This applies to buying a solution too. However, you will likely need different talent using an outside vendor.

Does your team have deep expertise in this area, or will they require outside support? If you’re buying or using a hybrid model, do you have access to any learnings/knowledge share?

Are the key technology players jazzed about doing this? Too often, I get “it doesn’t matter, it’s their job” to this question. It’s a worthy one, however. When you’re working with AI in almost any capacity, you’re basically adding a child to your family. Your infant will grow up and become self-sufficient at some point but in the beginning? They can be quite the hassle, and even if you are all love-and-light about your newest bambino, they’re still going to be a time-suck. I see too many good projects fail because the tech players are just not into it and/or they have too many other things to do.

If your founding development team and/or key players walk(s) out the door during the build process, will you be able to finish the project? What happens if there’s a mass mutiny down the road? Will you be able to handle it?  If you are buying instead of building, what happens if your vendor is sold? How will this impact you? Although this may sound like a Chicken Little exercise, it’s not. Right now, there is limited AI talent. Doing a “what happens if” worst-case scenario gut check is a good idea.

How important is quality to you? I know. I know. Folks like this question about as much as “what’s your problem?” and it’s important to answer anyway. Marketing AI projects are rarely perfect. However, in most cases, they don’t need to be. After establishing what you’re trying to build/buy, you need to figure out what’s acceptable to you and what’s not. I’ve seen many projects where everyone did the best they could, but they still couldn’t deliver to the level that was expected/required. Some AI projects need a 99% quality level; others can get by with less than 20% because even 1 out of 5 being right/useful is still making a material difference. You need to know your acceptable range.

EVALUATE YOUR DATE AND REVIEW YOUR INTEGRATION

What tools do you need to help you? In-house? Outside? (For example, hyperscale cloud providers.)

What in-house systems do you need to integrate with?

Whether your build or buy, your data is the key to your success in AI projects. That’s why it’s important to comprehensively review your data before deciding which path to go down. Do you have enough data for the project? Will using an outside solution help you or not? As you get more into AI, you’ll see that some companies pay nothing/peanuts for their AI solutions because their outside vendor wants/needs the company’s data to help build/grow their model. Remember, it’s still early in the AI Gold Rhodium Rush, so there are a lot of horse trades. It’s worth exploring how you’d benefit from this. Or not. Read more about data here.

Do you have enough data to do the project? Test it? Train it? Validate it? If not, how will you supplement it? How much will the data cost? Is it available? Has your vendor worked with your outside data vendor(s) before? Depending on what type of AI solution you’re creating, you may need more data than you have. Knowing what data you’ll need and how you can provide it before diving into your project is important. You’ll also need to know who has ownership of the data. If you’re using an outside service/vendor to enhance your data, get all this in writing upfront. You’ll want them to list the restrictions, if there are any. If there are known biases within the data, it’s also best to have those spelled out.

FIGURE OUT YOUR TIMELINE

When do you need this project to be up and running?

Are you replacing an existing solution where you have a specific end-by date, or is your timeline flexible?

Will you need to be able to maintain your current process while you build a new solution? If yes, do you have the resources to do so?

How fast will you start benefitting from this solution? I’ve learned the hard way that nobody tells you to ask this question, but your answer may change your whole build-vs-buy perspective. (This is project specific so your mileage will vary based on the project.)

FLESH OUT YOUR PROJECT

How will your new solution be integrated with your existing software/hardware? If you buy, does the vendor have previous integration experience with your software/hardware? How long will the integration take? If you are building something in-house, how do you plan to accommodate the new solution? How long will this take you? For the record, data and integration issues are one of the biggest roadblocks in new AI projects. You can avoid them by solid planning upfront.

Who will be responsible for training, testing, validation, and ongoing maintenance? No matter what anyone tells you, artificial intelligence is not a set-it-and-forget-it thing. Once it’s up and running, you’ll need to confirm that the algorithms are performing as (or better) than expected. This takes time and human oversight. This is perhaps the most important of all the questions on this list. Second only to “does your project really need AI?”

Testing, training, and validation are key to building a successful foundation. Maintenance and planned disruptions (when necessary) are critical to your ongoing success. Too many folks get super excited about building marketing AI solutions but lose their enthusiasm to maintain it. If this sounds like your company, you should probably take the “build in-house” option off the table. For most projects, ongoing maintenance is a HUGE part of AI, and anyone who tells you otherwise probably isn’t using AI.

How will it need to be updated, and by whom? How frequent are the updates? How much time/money needs to be allocated for them? These questions are applicable whether you are doing things in-house or with an outside vendor, but the amount of effort will vary greatly.

How much customization will you need to do the job? If you’re using an outside package/service, what will you need to make it work best for you? How much time and money will that cost you? Looking at a package with all the bells and whistles? Be sure to ask yourself if it’s overkill for the problem you’re trying to solve. Scope creep is a HUGE problem for a lot of AI projects. (Many folks think this only applies to vendor solutions, but it doesn’t. You should run through the same exercise if you’re building things in-house.)

How much flexibility do you want? This can be tricky to answer, but you should try anyway. If the solution you’re aiming for may change or halt your existing plans, knowing how easily you can pivot and/or modify your systems is essential.

Are you planning to expand this solution in the future? In other words, are you starting with one task (SEO tagging, for example) but hoping to grow this into a much bigger project (ex: SEO tagging + labeling + optimizing + content creation) all stacked together like a Lego house? It’s helpful to identify this upfront, especially if you think you’ll need to move from an in-house solution to a vendor-based solution. (This isn’t impossible, but it can get tricky.)

One of the things that shocks many clients is how many learning curves some of the most reputable vendors still have when working on projects. This shouldn’t be a surprise, but it often is. Even though artificial intelligence has been around since the 1950s, it’s only become helpful for Marketing in the past few years. (Yes, some companies sold “AI” long before that, but frankly, if/then statements that one can build in Google Docs are not considered artificial intelligence.) It’s often difficult to determine every specific challenge that faces your vendor(s) upfront, but make the attempt. Ask them which of your clients are like you. Which projects have you completed that are similar to yours? What would you have done differently? Which of your former clients left for this or similar reason(s)?

LIST YOUR NON-NEGOTIABLES

Are you okay with other people/companies having access to the data/information that will be involved with this project? (Answering “no” to this question is why some companies’ only choice is to build something in-house.)

If you are buying, you’ll want to know who owns the model(s) and what access you have to it/them if you leave. You’ll also want to know how often the models are updated and if you have any control over them. Whether or not this becomes an issue is usually due to the complexity of the model(s) you are building. However, it’s worth some thought, primarily if your vendor works with your competitors and/or companies who sell to the same market.

Can you keep up with this solution’s security and legal requirements? If you build it? If you buy it? Several industries (financial, for example) are so heavily regulated that it makes buying solutions incredibly difficult. Not to mention when a vendor finally gets their tool into such an organization, their #1 goal becomes to get into ALL similar organizations (read: ALL your competitors.)

How would IP (Intellectual Property) Ownership benefit you? Will having a patent give you a competitive advantage? You’re not likely to own the IP if you buy a vendor solution, but you could have it if you build it. Is that important to you? It’s important to remember that many AI models can quickly become the most potent weapon in your arsenal. You may not want to share them.

How will you account for bias? So, here’s the thing, bias is a complex subject. It would be irresponsible to pretend that a paragraph in a 3,000-word article will solve anything. It won’t. With that said, when you’re looking to develop an AI solution, you need to have hard discussions about bias, how it will impact your company, and how you will accommodate for it. If nothing else, you’ll want to determine where you stand on personal/human bias, model bias, and data bias. Then, you’ll want to look at how you’d manage those things if you built the AI in-house vs. bought it from an outside vendor. (As an aside, my opinions on bias have greatly changed since my first AI project. Model bias gets the most press, but for many marketers, data bias is the thing that can make or break you when it comes to Marketing projects. Both are incredibly important, and the latter is often overlooked.)

MISCELLANEOUS QUESTIONS YOU’LL PROBABLY WANT THE ANSWERS TO….

Would you consider a hybrid? Buy/license some parts of the project and develop the rest. There are so many solid base solutions out there right now; this is worth noodling.

What analytics/measurement tools will you need if you do this in-house? With a vendor?

How will this influence your current tech stack? Are there efficiencies to be gained/lost by doing it one way or the other?

How often will your solution need to be updated, and by whom? This is a repeat question from above, but it should be said twice. Marketers, especially eCommerce marketers, are used to postponing upgrades to the last possible minute or buying their way out of specific problems. With AI projects, it doesn’t work quite the same. It’s a good idea to have a plan of attack upfront.

What is the opportunity cost of building? Buying? Hybrid (DIY + outsourced tools, for example?) For some reason, this is often a controversial topic between marketers and developers, but it’s something you’ll want to consider. For example, if you decide to build, it may take time and resources away from other equally/more important projects that don’t have alternative/outsourced solutions.

I don’t have skin in the game when it comes to you developing your AI solutions. (If I did, I’d disclose it.) You must do what’s right for your business now and in the future. (Captain Obvious reporting for duty.) I’ve heard from many marketers that they haven’t gotten good advice regarding building/buying and made the wrong choices. Hopefully, the information above will help you make a solid decision. Before answering the questions above, you may also want to read Why Most AI Projects Fail.

If you buy an AI solution, there’s a whole article about selecting an AI vendor here. You may want to reference this as well.

Have questions about building or buying AI marketing solutions? Have a tip you’d like to share? Tweet @amyafrica or write info@eightbyeight.com.