While it’s true that the best data science is done by those who know their organisation very well, there’s a lot about data science that lends itself well to consulting style engagements. I’ve worked as a freelancer in data science (and analytics more generally) for the better part of a decade and in this post I’ll be showing how you can freelance using your data science skills.

The only requirement

There’s really only one thing you need to do in order to become a freelance data scientist - you need to get clients to pay you money to do data science work. That’s it.

Other posts on this topic will discuss things like getting a business license, saving up 6 months of expenses. They’ll list all the skills you should be proficient in before seeking your first client. But that’s how you do things the wrong way around. You could have all the skills in the world, a business license, and a nest egg. But without paying clients you’re not a freelancer, you’re just taking a career break.

With that said, let’s get started on the important stuff.

How to get clients

There’s an adage in the freelance world that goes something like this - anybody can get a first client but only a true business person can get a second. What this means is that anybody can find a client through their network. They can pester friends, cousins, successful uncles. They can contact previous employers. They can talk to enough people that they’ll eventually stumble into some work.

Leveraging your network like this is both a blessing and a curse. When you land that first client the day after starting your freelancing experiment, you’ll think that it will always be that easy. You’ll rely on your network to get you the next project. And you’ll cruise through your relationship with your first client without ever putting in the sales and marketing work needed to ensure that there’s another one waiting for you when you’re finished. Only to find that the network has dried up.

Let me be clear. You should tell everybody you know that you’re going freelance. A network never hurts. But if you want to have a serious go at being a freelance data scientist you absolutely can’t rely on your network to find you projects every time you’re out of work.

Data Science is different to most freelance disciplines - very few people need it (unlike hairdressing services, for example), success is objective (unlike logo design), and it’s very high-value work that hasn’t been commoditised yet (unlike basic coding). It’s important to keep these distinctions in mind when planning your marketing and lead generation efforts.

Know you customer

There are usually only a handful of people in every organisation that have the power to purchase data science consulting services. They’re the CEO, the CTO, the software engineering manager, and the department head that’s working on a crucial project in their domain (such as a Head of Risk or Head of HR).

These people are your customers. You should get to know them.

If you’ve put up a website describing your services and set up an AdWords campaign hoping that these people will be looking for ‘Freelance Data Scientist’ in Google, you’ve done it wrong. This group of people will never find you that way. What you find if you search for ‘Freelance Data Scientist’ is articles like this one (including another one of mine), articles explaining how to become a freelance data scientist or what the day-to-day life of a freelance data scientist is like.

The people in the position to purchase your services don’t start with a desire to purchase, they start with a problem.

Here’s a case-by-case explanation of what I mean:

CEOs are under pressure to stay competitive, to stay profitable. Their problem is finding the best ways to gain an edge in a competitive marketplace. They (usually) don’t care if you know your Adaboost from your Adagrad. They simply need help improving the one or two numbers their job performance is judged by.

CTOs and Software Managers are the people CEOs turn to when they think a business problem needs to be solved through the smart use of technology. CTOs aren’t looking to hire freelance data scientists for the hell of it. No CTO I’ve ever met has had the ‘my budget’s too big’ problem. These are the people who will try to figure things out themselves (with help from their teams). This means that they’ll be searching for specific answers to specific questions. What is the best algorithm for Churn Prediction? Can I use Azure ML with AWS Redshift? That kind of thing.

Finally, the departmental managers who run vertical-specific projects are not looking for freelance data scientists because they have no idea what a data scientist is. They need to know how to calculate employee turnover on a pro-rata basis over a group of 10,000+ employees. They need to know how to automate certain sections of their underwriting procedures to reduce the man-hours spent on menial tasks. They need honest advice from an expert in words they can understand.

Running ads against the words ‘Freelance Data Scientist’ might be a lot easier, but it’s a lot less effective than helping any of these people with their actual problem.

Pick a niche

Data Science is a broad field. Nearly every data scientist worth their salt could put together a rudimentary data warehouse, create a dashboard, implement linear regression from scratch, and train a convolutional neural net. This is what makes the field fun.

Unfortunately, diverse skillsets make for indistinguishable businesses.

In order to reach your customers (listed above) you’ll have to create marketing plans dedicated specifically to their problems.

Some people think that data science is already a niche. I’d argue otherwise but that semantic distinction isn’t even important here - the more focussed you can be, the better your marketing will work.

The reaction you want from someone visiting your website is ‘I can’t believe this exists’. You want to come up with an offering so unique and tailored to a specific group of people that their decision to purchase (or contact you at the very least) becomes automatic.

This doesn’t mean that you can’t do anything else for your entire career and that you’ll always be known for just one thing. It means that the people you care about finding you will find you. Once you’re in the door, and once you’ve built up trust, you can work with them on any projects you wish. In fact, if you do the first project well, they’ll probably ask you to do the second, regardless of whether or not it’s in your niche.

To give you some examples of niches, here are the ones that I’ve had throughout my career:

  • Building data warehouses for data science projects in FMCG businesses
  • Building recommendation engines for content websites
  • Predicting rare events in retail and e-commerce (purchase, churn, fraud, theft)
  • Using computer vision in SaaS apps

Each of these relies on a specific area of data science and a specific industry. For your first niche, you should pick the area of data science you’re most comfortable with and an industry that you care about helping.

Marketing by sharing

Once you’ve chosen a niche you need to get to work marketing inside of it. Most people like to learn and most people will try to solve their problems by themselves first. You will go a long way teaching others.

I read so many data science blogs that analyse the same Kaggle datasets over and over again. Or worse, they use np.linspace to generate a dataset to explain some concept or phenomena. This is not how data science works in business. Articles like this start too late (they don’t describe the data retrieval and cleaning efforts) and stop too early (they don’t describe the business impact of the solution).

If you’re going to write a blog post to explain a concept in the hope that a potential client will find it, you need to explain in the article how the concept solves their problem.

People seeking data science clients should never write a general exposition on a particular algorithm. Go the extra distance and find a problem area in your niche that you can apply it to. And do this a lot. The more you write about your niche, the more people will equate your name with that specific area, you’ll become the no-brainer hire.

If you build up a large enough body of this kind of work, your SEO game will be on top form and customers will start finding you. Not everybody will convert but some will and they’ll be the people whose problems you know how to solve.

Build a process

Now that you’re getting clients, you need to optimise the process of working with them. This is important for two reasons:

  1. It helps the engagement go smoothly and prevents scope-creep
  2. It allows you to extract and reuse key deliverables

Another benefit of having a niche is that you end up seeing the same problems time and time again. You can reuse the same algorithms, explain your solution in the same way, and run the same discovery meetings over and over again. This saves you an incredible amount of time and allows you to focus on the important work in the project.

The process you end up developing will become a selling point. It will be the thing you use to explain how you work in case studies and pitch presentations. So as soon as you can you should start documenting the stages of every project, getting all of your deliverables in order, and learning how to write case studies based on your process.

Double your rate

Most freelancers charge too little. They’ll do some algebra and work out how much they made per hour at their old job and use that. Despite lots and lots of advice to the contrary, that still seems to be the way people go.

So, what I’ll say is this - if nobody pushes back on your rate when you submit a proposal, double it next time. And keep doubling it until someone says it’s too much.

It might be easy to write this off as greedy or self-destructive but the fact is, if you are doing very niche work, you’re going to very quickly become an expert at it. Experts get to charge a premium. You should be adequately rewarded for developing processes and deliverables that allow you to do the work quicker than anyone else. Why should you be penalised for being efficient?

Have an exit strategy

Without a specific plan, relationships with clients go the way of entropy. They say they’ll get back to you when they have a new project, you’ll say thanks for the opportunity.

The moment that you’ve finished doing great work for a client is the exact right time to ask for referrals and testimonials. When you’re finished, sending an email with a short questionnaire about how the process went and asking for the names and contact details of people with similar issues makes you look like a professional who cares about helping the community.

The only times this would be a problem was if the work was substandard. So do good work and ask to be referred.

Glowing quotes help with social proof for your marketing materials. And the exit questionnaire allows you to determine what the client would be happy to see in a case study about them.

Asking a client to write you a recommendation (without being pushy) gives them the chance to reflect on the good things that happened throughout the project and to end on a high note instead of the pressured, rushed feelings that large-scale deployments can usually leave you with.

Thanks for reading

If you have any questions about being a freelance data scientist, feel free to email me.