Copilotization of CRM

If there was a Word of the Year jury for the Microsoft ecosystem, 2023 would have been a really easy job for them. Satya Nadella stated that MS was now The Copilot Company. Even if the name wasn’t officially changed at this point, it’s not entirely impossible to imagine them “pulling a Meta” and trying to underscore the profound nature of this new strategic direction.

At least AI would seem like a better card to go all-in on the branding front, rather than what Facebook did with the Metaverse. This is now the new battlefield for all tech giants and each will aim to define the market in their own way. Zuckerberg has chosen to compete with the open source LLaMA that could capture market share also outside Meta AI. Microsoft, as we know, has picked the path of “copilotize all the things!!1!”

Upon reading the new posts on the Power Apps product team blog, one of the feature announcements got me thinking about what generative AI has done to the classic dilemma of CRM systems. Meaning: no one wants to use them, unless they have to. As a result, not nearly enough data is captured into them to make "customer 360" a reality.

The feature I’m referring to is called Copilot assistance for filling forms. With this new capability, users can now generate text into the forms of model-driven Power Apps - and any Dynamics 365 CRM apps. (Also premium canvas apps are covered.)

Sounds amazing, right? AI does more work for us and the CRM systems everywhere will just magically become better! Users will be able to fill forms faster, so that they can… fill even more forms during the day. Hmm?

The real challenge with CRM systems

It’s been over two decades since I started working with CRM systems. The technology we have available today is nothing short of magical compared to what the first Siebel based system was that I had the pleasure to be the end user of. Or the first time I was responsible for implementing a new CRM system for an organization, which was Microsoft CRM version 3.0 launched in 2005. Who could have ever envisioned the wealth of tools that are now magically available from the cloud at the click of a button?

How about the implementation projects themselves? We’ve also seen some progress here via the more agile minded methodologies becoming broadly adopted (on paper, at least). The way us humans operate as part of such project organization has therefore evolved somewhat. But the human evolution of an individual isn’t quite that fast. We the people are pretty much the same beings, no matter if we’re faced with a Siebel CRM or the latest & greatest from the Microsoft cloud. New technology only uncovers new truths and nuances about what us humans are like.

Often when doing the traditional CRM project initiation dance of starting with requirements analysis and detailed specification of what record types and fields are needed, you ended up with waaaaay too long forms. Business process owners and other involved stakeholders saw this as a “now or never” chance to get the exact data they always hoped to have available for their own perfectly valid needs.

What was often forgotten was that the CRM system itself was not gonna create that data. It would largely come via the keystrokes and mouse clicks of the CRM users. Who of course didn’t enjoy the idea of doing more data entry work to serve the needs of those who would consume the summaries and reports created from that data.

So the users would do the bare minimum. Either through not even understanding which forms and fields they should be updating as part of their new CRM powered routines. Or by actively trying to get around this new digital obstacle that had been put on their way, stopping them from achieving the outcomes most critical in their own daily tasks.

When noticing that data is missing and the assumptions made around its existence in the CRM system and related processes that depend on it, one stick to apply as a poor remedy is making fields required. “Yes, now we have rules and enforcement!” What’s the end result from this? Even lower system adoption, for starters. But the big problem is a decrease in the validity of that data. You’ll get “something” in the fields, sure. But collecting garbage in the system will only make it push garbage out into related processes.

Activate the cloud garbage generators

Let’s imagine the CRM user facing yet another daunting form, like an account plan or any other complex business process flattened into a 2-dimensional sheet of digital paper. The natural reaction from a human being is not “I want to really focus all my time and energy on entering values into this form”. Rather, it is “how could I get this thing completed as fast as possible, so that I can move on to things that actually help me achieve more meaningful goals.”

Imagine the delight when the CRM user will see the magic wand button on the UI. Or better yet, when the form is already prefilled with credible looking data that Copilot suggests for each field. OMG! They will totally use it to get to their immediate goal - i.e. moving away from the form.

Generative AI is the ultimate BS generator, trained on the vast piles of digital manure found online. There is no way that we the humanity can fight these superpowers of Large Language Models when it comes to producing data that looks valid. We will expect everyone around us to be using their Copilots to get around text generation tasks that feel too daunting for humans to complete alone.

There are plenty of scenarios where GenAI is gonna be amazingly useful. Yet I don’t think that the act of filling forms in CRM systems is necessarily such a scenario. Yes, it will most certainly help the users to do what is asked from them. But you should pause for a moment and think: is what we are asking from the user really what we need from the system?

The challenge that AI should learn to solve is not in generating piles and piles of unstructured text. It's the opposite: turn unstructured data captured from real life interactions between people (and machines) into something that can be stored in a structured relational database format.

The endless path of least resistance

During the era when I was working exclusively around CRM systems, I was intimately aware of the low user adoption challenge. I tried to overcome this through various means.

Improving the usability and user experience of the applications was one technical way. Microsoft offered pretty nice no-code tools for designing and implementing CRM system improvements on both the UI side as well as back-end automation.

Making the system easier to approach was not just about technology. Providing clear, even contextual user instructions that were found in a logical place was a key area, especially when working on the customer's side of the table. Arranging training sessions for both initial system go-live as well as new user onboarding was another big effort.

In my blog, which was called “Surviving CRM” back in those days, I spent quite a lot of time and words to analyze the intricacies of usability, user experience and user interface design within the frames of Microsoft Dynamics CRM. Not just the features in the latest product versions, but also studying the platform evolution over time. Digging up examples of what the design goals for the technical product were, it became obvious that “ease of use” and “simplicity” had been on the agenda across decades:

New features get introduced as the platform grows. The UI elements keep moving around and the interaction patterns change as the industry trends demand something that looks “modern”. Engineering teams keep going around in circles, tyring to solve the challenge of how to make room for new features while still preserving compatibility with all the old functionality already deployed in real customer environments.

There’s an endless amount of things that could be done, both on the development and adoption sides, to try and reach CRM nirvana. And yet it feels like an elusive goal. At some point, I arrived at the conclusion that none of these activities would be enough to solve the fundamental problem. Which is us humans. We are the biggest barrier for CRM success.

The ultimate solution to CRM adoption challenges?

The answer is simply to remove humans from the CRM equation. Well, if not entirely, then at least do away with as many dependencies to getting individuals to follow the common instructions and processes to operate the CRM machine.

Before you put up any forms and fields for users to see, step away from the customization tools of your CRM system and take a deep breath. Ask questions like this that will help you ground yourself in real world instead of computer world:

• Have you explored all options of getting the data into the system via an integration instead? (Yes, you’ll pay more for IT. No, the human labour of CRM users isn’t free either.)

• What bad things will happen to the business if this data doesn’t get entered by a human? (If you can’t think of any, just forget it.)

• Will something bad happen to the individual human if the data entry task is neglected? (If not, why would they bother following your new rules?)

• How reliable is the entered data gonna be anyway & how could you evaluate it? (Remember: filling form fields is super easy for AI. Knowing what’s real and what’s fantasy is harder every single day.)

• Why would the users keep the information up to date after initial save? (If you can’t be sure the data remains valid, determine why you’re better off storing it instead of deleting it?)

Point being: in the year 2024 there’s unlikely to be great new business innovations to be discovered through manual data collection. The various data sources and streams encountered by the same people who should also use the CRM system are overwhelming enough as it is. Every new task of digital interaction you assign to them increases their cognitive load. Process improvement too easily leads to complexity growth, if all you have is good intentions on operational level but no method to evaluate the systematic impact on your employees.

Instead of building a “one place for all customer data” type of an enterprise monolith, maybe technology could instead reduce the amount of information us humans have to even see? Having LLMs generate new data from existing data isn’t really helping here, especially if humans need to verify its reliability. Sure, the act of tireless AI bots summarizing big piles of data for us sounds nicer, but again: pause for a moment to reflect on what will likely happen. AI hallucinated outputs will inevitably find their way into user inputs - like what is already happening on the public web.

“Garbage in, garbage out” becomes “garbage out, garbage in”.

Of course some user data entry is likely to be always required in business processes. Today, a basic CRM for managing accounts, contacts and sales opportunities is a commodity. You can quickly build one with Power Apps. Or if you prefer not to start from scratch, look at apps like RapidStart CRM, as an example. Thanks to the Power Platform capabilities, you’ll get to enjoy similar bogus data suggestions from Copilot as with Microsoft 1st party apps:

Copilots vs CRMs, apples vs oranges

Copilots aren’t inherently bad, though. That’s not the message to take away from my ramblings here and on social media. Just like CRM, they are a tool. Neither the existence of CRM nor Copilot will generate business value on its own. We must strive to develop an understanding of how they can be a part of the solutions that us humans actually need. The Copilots of today are around the phase of Siebel in terms of evolution in technology and business process alignment.

Looking at Microsoft’s business applications product portfolio, the Copilot features (with a capital C) will likely increase the share of built-in product capabilities used in Dynamics 365 apps. The common processes that all organizations doing B2B sales need to handle is what Microsoft is best positioned to cater for. But there’s a catch: the more there are built-in AI features included, the bigger the barrier for customizing such area of the product in a significant way.

The divide between product and platform, Dynamics 365 and Power Platform, may grow wider as a result of this. I don’t think they’ll be detached from one another, though. There’s no technical reason to do this. It’s just the commercial models, both for MS and partners, that are going to keep evolving into different directions. Both will be subject to the unstoppable wave of copilotization - and the shifting role of humans.

Building your own copilots (lower case c) and expanding the AI capabilities via plugins/connectors will surely need some expert resources. Just like the XRM customization tools never were that easy for citizen developers to dive deep into, the world of AI inside the MS ecosystem can also be a bit overwhelming for many. The more we try to shield away the CRM users from getting exposed to too much data, unreliable metrics or irrelevant copilot suggestions, the more work there is for building the automations and processes that happen behind the scenes.

One question that remains is: will the consultants who are used to performing the work required in classic CRM implementation projects be willing and able to jump into these new roles? Copilotization of CRM will not only change the way the end users are expected to interact with business applications. We’re going to need the architects, sense makers and challengers who can make the marvels of new technology fit with the real-life scenarios of CRM users. Either to better support them, or replace them (where applicable).