The world’s leading publication for data science, AI, and ML professionals.

Retainer Agreements in AI Consulting

Retainer services are delicate but useful

Google’s definition of a retainer is "a fee paid in advance to someone, especially an attorney, in order to secure or keep their services when required." In this article you will get some insight into how AI consultants can think about adding retainer clients versus engaging in the more standard corporate engagements of defined scope and duration.

There are already some great articles out there on how to select your hourly rate as a consultant, and I’ll try not to replicate that here. Instead, I’m going to hone in on how to maneuver the initial conversations at the lead and opportunity stages into either client retainers, or the more standard hourly/fixed fee engagements. I argue that a retainer is a relationship status, rather than a product.

From this link
From this link

Retainer conversations tend to happen after some initial project, rather than on day 1. The reason is simple. The consultant really needs to invest in the relationship before a retainer is a serious option for the client. The risk of signing off on the retainer based solely on reputation of the contractor is just too high.

It’s useful to consider that the product-market fit for a retainer is different for different potential clients. Given that AI consulting is a services Business, the client needs to fit the service to their needs, and the contractor needs to make sure that they wind up in a good situation as well.

I like to think of this decision between retainer and regular engagement as a generic services industry puzzle, rather than an AI-industry-specific situation. Let’s pretend that AI consulting is Search Engine Optimization (SEO), and a retainer engagement is SEO monthly fee consulting, while hourly or fixed price work for development is standard work (example here). In this analogy, the client pays you a monthly retainer for SEO services because you are delivering a unit of work that they measure by the month, in metrics like traffic, ranking, and sales. Also, note that this is an ongoing effort that doesn’t end. There is no such thing as "we are finished with SEO." This monthly fee is not for professional services by the hour, and instead it is performance based. On the other hand, the hourly or per day rate stuff, be it AI or web development or SEO, has a fixed deliverable that companies tend to scope out and pay for, and then manage internally.

From this link
From this link

This internally versus externally thought process leads me to my next point about retainers. Good managers are always making "build vs buy" calculations. In the proposal/RFP stage, the manager is thinking of meeting some requirement or improving some KPI, or addressing some pain point. It’s at this time when they think about using internal or external resources. The reason we get deals is because the client has opted for the buy rather than the build option (with some exceptions I’ll skip over, like training and strategy Consulting and advisory boards). This same "build vs buy" thought process happens when a company is considering a retainer agreement. The manager thinks about the cost of hiring a resource to fill the role versus the cost of bringing in a consultant to build stuff.

Some clients just don’t like retainers. For example, governments, in my experience, maintain vendor and supplier lists or professional services contracts, but typically not retainers by name. I imagine this is different in other fields like high-end legal work and politics, but in high-end high-tech, it’s just not a thing in government, as far as I have seen. However, that doesn’t mean governments don’t do retainers. Instead, they cast it as fixed-cost consulting engagements. Some clients including government departments may have consistent hourly work that can’t be scoped out, and they also have been allocated a fixed annual or quarterly budget to spend. In this case, they may hire a consultant to do X amount of work for Y amount of hours per year/quarter. Rather than a retainer, this is a fixed-price project. The lingo doesn’t matter though. It is what it is.

From this link
From this link

Some clients prefer a retainer because they are expecting an uneven workload on a project or in a department that they really need to get right. Think drug or medical device development (lots of fits and starts, following the waterfall model), or placing a consultant to add input into an existing Data Science team.

Repeat business is good for the consultant, as it provides more consistent revenue, and for the client, as they don’t have to continuously on-board strangers. There are common security concerns that lead to internal processes for testing consultant reliability, and avoiding doing that over and over again is a plus for the client. A successful retainer provides both parties with some extra positives. However, if the price of the monthly retainer is too high, the company simply staffs the work into a new position (if they can), and if the retainer is too low in hourly rate terms, then the consultant simply backs out into a full-priced hourly engagement. Let’s dig into the numbers to give you a much more tangible sense of how this works.

We charge $250/hr or 2K/day for consulting engagements. More on that here. So, a retainer for 10K/month for 5 days/month sounds smart, but the company can also hire an employee for 10K/month = 120K/year to cover salary and benefits, and get a full-time employee rather than a consultant. That new hire is honestly just as easy to fire as a consultant, and so the consultant needs to make the argument in a few ways that this relationship is worthwhile for the company.

First: Quality. The consultant is awesome. So awesome that a "regular" employee just won’t add the same excellence, output and insight. A retainer only makes sense if the relationship adds high quality. This is a common reason for sitting on an advisory board or sharing in management planning sessions. Sometimes (not always) the company wants to promote that they are working with you. This is basically free marketing for the consultant, and strategic marketing for the company. Other times there is a miles long NDA to keep the relationship a secret.

Second: Value. That 5 days of consulting provides more than 20 days of value you would get from a regular human. This is all about the better faster stronger code and advice.

Third: Flexibility. Within reason, the consultant will be flexible to scale effort up and down as needed, without modifying the pricing of the retainer. This flexibility and planning provides pricing consistency for the client. If suddenly the client needs a huge rush of hours, there is a "volume discount" for the extra effort, and we don’t need to get contracting involved in order to plow ahead and deliver the thing the client needs.

Fourth: Availability. The retainer client gets first in line if something needs to get done. They won’t be bumped down the schedule by other one-time projects, and in fact, their needs will bump other projects out of the way to maximize the quality of service experienced by the client.

Fifth: Scale. The consultant can pull in their team in a crunch. It helps to have the ability to scale up without hiring people.

Sixth: Small stuff. Uneven small work is really hard to parcel out to outsiders. The new guy/gal doesn’t know the context of the small task, or the technical setup within the organization, and so pushing these small tasks to a consultant that is in the loop really helps to iron out the hard edges of the company’s internal team. Many lawyers engage in 6-minute billing time increments. For small items like a phone call, this micro-billing gets rounded up. It is a very real problem when consultants act like this, and so I don’t do it. Neither should you. A retainer brings the relationship past the nickel and dime tracking, and into the more productive zone of working to get what needs to be done, done.

Seventh: Knowledge transfer. Post deployment support can melt into a retainer agreement if the deployed system keeps being updated and maintained by the client. They may want to keep the consultant around on a very part-time basis to help out. This is often not the case, and the transition ends the engagement, but sometimes the situation does call for such a move. Institutional memory should, by design, never be in the hands of consultants. The main reason for keeping the engagement going are likely around small improvements over time. This iterative process can present a risk to the client that they want to mitigate by keeping the client around for some work on odds and ends between big pushes in product development.

Eighth: Science! A lot of stuff in ML/AI R&D is more R than D, and the R is really not possible to capture with a fixed sized scope. Instead a retainer agreement can help to push a project along at a fixed pace (burn rate), with the consultant committing to push the project along for no more than $X/month. This is typical of a research unit in a large corporation or a start-up.

Ninth: Experience. It makes a difference to bring in outside voices who interact with a wide array of solutions and client requirements. This is the idea behind having an outside advisory board. That experience can be transmitted into the thinking of the corporation, avoiding pitfalls and identifying opportunities from information outside the corporate bubble.

Tenth: Reason. Be really clear to offer to do what makes the most sense. You don’t upsell a retainer agreement. Instead, you offer it as an alternative to continuing the per-project pricing relationship you are already in. The client should feel that a retainer represents a win-win, rather than something you want to sell them.

But wait…

There are many reasons why you should not do a retainer with certain clients. Until this point we talked about how a retainer can possibly be a good thing. Let’s now consider why retainer situations can go off the rails. Our CEO Mathieu Lemay likes to call warning signs "red flags". Some clients wave a huge red flag when it comes to retainer agreements. These clients engage in a pattern of feature creep, scope creep, and renegotiation tactics that you can handle on a per-project basis with very tight scope, but that you don’t want to navigate in a more unstructured retainer relationship. Think of this like a dinner partner. Going dutch is totally OK. Getting into a retainer is like taking turns paying for dinner. If you feel, emotionally, like the client is going to order a 72 oz steak when it’s your turn to pay, then don’t get into that situation in order to find out if that’s really what will happen. The consultant needs to feel confident that the relationship is on a good footing before committing to this kind of deeper relationship. From the perspective of the consultant, if you are going to make a bit less money a bit more consistently, it’s a fair deal. If the assumptions don’t hold, then stick to the standard pricing model.

Summary of main points:

  • How to think about offering a retainer
  • Sometimes a retainer is the right fit for a client, but not always
  • A retainer needs to be a win-win situation
  • Overcome build vs buy to get to a retainer
  • Factors that help promote the need for an AI consulting retainer agreement are: Quality, value, flexibility, availability, lots of small stuff, knowledge transfer, science, experience, and reason
  • Don’t get into a retainer agreement lightly

The Context

I decided to write this article amid some much appreciated attention around my previous articles, and also some drama. The Paris ML study group selected my articles "How to Hire an AI Consultant" and "How to Price an AI Project" for their discussion group "AI Professionalism: Consulting, Project Pricing, Delivering Results". The drama was about plagiarism of one of those same articles, and is best summarized in this post on LinkedIn and this followup post. In fact, I did a webinar about drama in the AI world that same week ("AI: Truths, Falsehoods, and Outright Lies"). I also reached 3K followers on this platform (THX!), and reached out to conduct a poll. Here is the poll question I asked:

What do you like reading about?

  • Business stuff. I want to make money. Stop showing me code.
  • Applications of AI. Do cool stuff with code. Keep me up to date.
  • Startup stories. Tell your story. Inspire me.
  • Everything. You do you, buddy!

Only 52 people voted in the poll. If you didn’t get the email and want to participate in stuff like this, you can join the newsletter or just email me directly. I really do read my email.

Here are the results of the poll:

And so, it seems a big chunk of the audience is most interested in seeing cool AI applications. I respond to your feedback. My next article will be about exactly that. I am working on an alpha model using some previous work from a 2017 article I did on interpreting the type of a small company based on the name of the company. Here is the code for the library. You will notice a bunch of commits to the code as I continue to do stuff for the article.

While Mathieu Lemay is in Dubai with the stallion.ai team, I’m here in Ottawa running the shop, and office-ing the technology in a chiefly fashion. Samuel is running AuditMap.ai demos, and we are generally just doing our thing.

If you liked this article, then have a look at some of my most read past articles, like "How to Price an AI Project" and "How to Hire an AI Consultant." And hey, join the newsletter!

Until next time!

-Daniel Lemay.ai [email protected]


Related Articles