Editor’s note: In the following exchange, folk’s co-founder and CEO, Simo Lemhandez, shares (much in his own words) the amazing journey of building folk as a AI-native CRM, designing critical and non-critical workflows, and his hunches and beliefs on monetization and pricing models. We had a ton of fun hosting Simo, hope you enjoy reading this chat.
Here’s an index documenting the top takeaways, to help you find the right sections quickly:
- How the AI paradigm-shift is remaking CRMs
- Building AI features: What has changed and what won’t
- Designing intentionally, and differently for critical and non-critical user workflows
- Thinking through pricing models in tandem with your industry, customers, and product promise
- Why software companies might be over-indexing on chat interface and how to approach it instead
- Hot take: CRMs will be the AI SDRs that actually deliver
How the AI paradigm-shift is remaking CRMs
I think the paradigm for what we expect from software has changed. In the previous paradigm users were doing the things; software relied on the principle that a user was literally sitting in front of their computer and typing on the keyboard in order to, let’s say, move a contact in a database, move a contact in a pipeline, and so on.
That’s the paradigm of the past.
Today we expect software to do more for us, software should do the busy work, and let us focus on what we’re uniquely good at as a human.
Take the case of CRMs. I think there is a lot of needless friction in existing CRMs like HubSpot and Pipedrive because of this old paradigm of operating — they rely on manual input from the user, expecting the user to do a bunch of manual work. It’s not ideal. It never was.
We have been flipping the script for CRMs, and with AI we are taking it even further. We never wanted to be just another CRM, how we are thinking of ourselves now is as a sales assistant.
We recently released a feature called AI follow-ups, which is one showcase of how we are making this shift to the new paradigm.
It enables you to skip the entire, old-world hassle of having to sift through past conversations manually to figure out the basic task of identifying which contacts you have to follow-up with.
Instead, “AI follow-ups” surfaces the right contacts for you after intelligently scanning all the data — and will even suggest a fitting reply, saving you a ton of time.
This is just one example, one glimpse of the vision that’s the future of CRM: more proactive rather than waiting on a user to do the work.
Building AI features: What has changed and what won’t
Building AI features is very similar to building any other feature in the sense that what you’re trying to do is solve a user problem. Otherwise you risk falling into the fallacy of building yet another AI feature that just does a little marketing for you.
So I think it relies on the same principle of deeply understanding your user, their needs, their frustrations, and finding a great way to solve them.
One area where it does change is that AI is a very new tool in our toolkit and needs to be tested out differently. Given the statistical aspect of AI, you need to test the prompts on a large volume of data, and that can make the testing phase and say more extensive.
You also need to consider the non-deterministic nature of AI and make the right workflow that is smart enough to differentiate between critical actions and non-critical actions.
Designing intentionally, and differently for critical and non-critical user workflows
One thing we have seen to be helpful with testing is to design the features so that the non-deterministic nature of AI isn’t critical. For example, I won’t send a communication to a customer’s end-user on their behalf without them in the loop to review and validate.
Basically, you design a product so that it takes into account the fact that AI is not deterministic and that for certain actions it requires to have a human in the loop, it requires a step of validation. This links back to deeply understanding user needs and what they consider critical and what they don’t consider critical.
For example, with AI follow-ups, we detect follow-ups for your conversations that have turned inactive. We flag them to you, we suggest the follow up email based on your past conversations and their context in your tone. But we don’t send that email for you.
Whatever is customer facing for you, we consider it critical and we don’t do it on your behalf. Whatever is not customer facing, and probably less critical, we do it on your behalf. For example, we have some fields that are intelligent and are populated more or less through AI, like “the strongest connection for a contact”.
If it happens to be wrong, it’s not a nightmare. We will try to make it right obviously, but we will be bothering the user if we were to ask them to validate/review each time. So we just update it and let the user change it if they want.
Thinking through pricing models in tandem with your industry, customers, and product promise
I would say that customers vote with their wallet. So monetization is crucial in the way we see the products and in our way to understand our product market fit.
That being said, as we see that the product is advancing very, very fast we are able to increase our price quite frequently for new users.
Per-seat pricing tends to make less sense in a world of usage, but it does have some benefits like the fact that it is very transparent and clear to the user. I do think there is no single rule for pricing. It needs to really be adapted to the specific product we are talking about.
When it comes to CRM, there is no real notion of usage. This is an area that we keep looking at very, very closely, but we will never do something that is not in the interest of the users. So we will never go for a usage-based pricing if it’s at the detriment of clients.
Of course there are various kinds of new pricing models like usage-based and outcome-based, but I believe that customers will tell us and they will be the best judge on where we go. If we do realize that it makes sense to look at outcome-based pricing, we’d consider it.
Today, though, one of the benefits users love about us is that we are super simple to use and go with. I feel bringing multiple dimensions to our pricing will go against the simplicity we strive for, so we are sticking to per-seat for now.
Why software companies might be over-indexing on chat interface and how to approach it instead
One of the things that I do believe is that we are over-indexing on the chat interface because of the success of Chat GPT and other AI chat bots. But when you think about it, the AI chat interface is one of the least specialized interfaces.
A lot of software are building chat embedded into their software.
But I don’t want to chat with my CRM, I don’t want to chat with my ERP, I don’t want to chat with tools that I use every day. I just want them to feel that they are making me smarter.
A lot of people tend to over invest in chat interface. What I think is that you should deeply embed AI features in the way your system works and make the interface more specialized, not simply look like a chat.
This realization came from observing how customers interact with ChatGPT. When they want to find a conversation from a while ago, they keep scrolling infinitely. It feels built for instant gratification and those “aha moments,” but it’s not built for an organization that needs to maintain structure over time or for users who need to retrieve specific data at certain times.
Hot take: CRMs will be the AI SDRs that actually deliver
I think the promise of AI SDRs is amazing: having someone do the work on your behalf resonates strongly with the market.
The challenge I’m seeing — and I’m happy to be wrong — is that I’ve yet to meet a customer who was truly happy with their AI SDR.
It just feels spammy when you receive an email from an AI SDR. They aren’t connected enough to your world to make each touchpoint feel as if it was sent by a human.
That’s why I believe the AI SDRs of the future are actually the CRMs because they have the data. I’m obviously biased here, but that’s what I think.