Who should own pricing? ๐ธ
Deciding whether pricing strategy stays with product/GTM, the costs of pursuing foundational models, and positioning towards JTBDs not personas.
Welcome to the 91st edition of The SaaS Baton.
A fortnightly newsletter that brings you hand-curated pieces of advice drawn from the thoughtful founder-to-founder exchanges and interviews taking place onย Relay (curated with ๐ at Chargebee for Startups)ย and the interwebz. So, stay tuned!ย
In this edition, youโll find the following instructive and inspiring pickings:
#1: Sendbirdโs co-founder and CEO, John S. Kim, tells us where the pricing discipline must ideally be housed โ โI donโt recommend putting it in finance orgsโ โ across the various growth phases of a startup.ย
#2: Tara AIโs co-founder and CTO, Syed Ahmed, breaks down what it takes โ โcan cost anywhere from $10M to a $50Mโ โ to train oneโs own Large Language Model.ย
#3: Paperformโs co-founder, Dean McPherson, writes about โ โone of the biggest strengths and frustrations weโve hadโ โ decisively not picking a niche in the most horizontal of SaaS markets.ย
Finding these discerning founder takes valuable? Please consider sharing this edition with an ever-curious teammate or a much-cherished SaaS friend. ๐ New readers can sign up here.ย
๐ Recently on Relay:
โก๏ธ Heuristics and Hunches (February 9th) โ โSharing Our Experiences, Wins, and Lessons is Also Humbling. It Keeps Us Honest,โ and Other Notes on Building a Customer-First, Bootstrapped Business with Cannyโs Co-Founder, Sarah Hum
โ Documenting Cannyโs path in public
โ Why they didnโt have to revisit PMF
โ Building in a tough market
โ Being self-funded
โ Making bets as a lean team
โ Cannyโs freemium experiment
โ What theyโve gotten right from the start
โ How they look at competition and integrations
#1: Who should own pricing? ๐ธ
(From: Sendbirdโs John S. Kim) (Source: Code to Cash)
If youโre early on, itโs probably the CEO/CTO. Somebody does it, right? But as you go beyond $10m/$50m in ARR, then pricing becomes such a huge factor. Because if you just have a 5%/10% optimization through pricing, thatโs a material difference (could be millions of dollars worth for your company.)
Somewhere along the way, letโs say, series A to series BโฆBased on competitive dynamics, if, generally your product is in a differentiated market, you have a unique value proposition, then I actually recommend having as a product manager or a head of product or the CEO herself/himself own the pricing as long as they possibly can.
But if youโre in a market thatโs highly competitive, very easily commoditised, then have it with, say, the marketing function. Because the pricing is more reflective of whatโs out there in the market. The differentiation then will come from brand and positioning, rather than your whole productโs unique differentiation.
So it really depends on whatโs your core strength and what are the competitive dynamics of the market. Thatโs kind of the big pillar.
After that, once youโve reached a certain scaleโฆ it does make sense to consider other options. Having a pricing specialist. Having revenue operations and other departments form a joint task force to understand pricing and for you to iterate and review gross margins and other pricing factors.
Also the continuous process of getting feedback from the market, win/loss analysis, getting all this feedback back into your pricing organization.
I still donโt recommend pricing put into organisations like finance. It has to be more directly connected to either the product/market/customers. Having some form of oversight or decision-making power ultimately on the go-to-market function or product function by having a specialized pricing organization would make sense if youโve reached that scale.
Also consider bringing on other experts like Simon-Kucher or other pricing experts. Who cost anywhere from six figures to seven figures, but letโs say you spend half a million dollars on the pricing exercise thatโll get you a few million dollars in net new ARR. Itโs worth it, right?
Thankfully, a lot companies that went through this pricing exercise actually share on their blogs. I think Fivetran shared their model, going from a subscriptions based to a consumption based model. Learn from other best practices too. Especially thankfully in tech community, a lot of people do pay it forward. They shared their mistakes and learnings.
#2: Should you pursue your own LLMs? โจ
(From: Tara AIโs Syed Ahmed) (Source: LinkedIn)
I get asked this question a lot.....As a startup should we pursue training our own Large Language Model (LLM)?
My answer to this question is usually along these lines.
How diverse and sizable is your dataset?
Foundational models depend on both a diverse and large dataset so that they can provide utility to end users. Similar to a human, your dataset will need to be a mix of knowledge sources such as Math, Code, English grammar, History, etc. in order to create a foundation that meets the needs of your end user persona.
Do you have the resources to structure and tokenize data?
This is the actual bulk of both time and cost when going out to train your foundational models. It can potentially make or break your model. Having run this process many times in my career. The data pipeline, statistical and heuristic work on identifying what data is useful for my model vs what is not is a tedious process. This can take anywhere between months to years to get right even if you're using acquired data for tokenization.
Does my team have the expertise to parametrize the model configuration?
Once your data is in the right shape to be processed, the next step is to articulate how the model will be configured for training and evaluation. This is also an iterative process, rarely is this figured out in one shot.
Do I have the compute resources to train, retrain and maintain this foundational model?
The compute resources needed to process your data, train, evaluate and publish the model are all dependent on variable compute costs. For example it cost Meta approximately $4M-5M in compute to train LLaMa on 2048 Nvidia A100 GPUs.
If you're willing to go through all of the above, then training your foundational model in totality can cost anywhere from $10M to a $50M, dependent on data, team, compute and expertise.
#3: Not niching down ๐
(From: Paperformโs Dean McPherson) (Source: Relay)
Not niching down is both one of the biggest strengths and biggest frustrations weโve had. From the very beginning, we always thought of Paperform as more than just another form builder.
We wanted it to give people the ability to make their own solutions to their own problems. Over time, the phrase, โno codeโ started being thrown about, and it was like remembering a dream โ we finally had the words to explain what we were trying to do.
Thereโs certainly wisdom in the early startup advice of niching down.
But I believe that itโs just a common solve for the problem of differentiation. We didnโt want to differentiate by being โthe form builder for South African dental practice patient registrations.โ We wanted to empower people to create for themselves.
Over time, our โnicheโ has actually been to be a broadly horizontal tool.
Paperform is a digital swiss-army knife.
Today, customers use us for data capture, surveys, quizzes, landing pages, selling products, scheduling appointments, lead generation, client onboarding, subscriptions, consent management, and hundreds of other things.
As a self-serve product, this means we spend a lot of time thinking about what people are aiming to achieve when they sign up for a trial.
Unlike a heavily verticalized product, we donโt get much value out of thinking of our customers in personas. Instead weโve found it much easier to both articulate and focus by thinking about โJobs to be doneโ.
The frustration that not niching down brings, though, concerns both how you target and communicate with customers. Weโve gotten much better at this over the years, but itโs something that we will always need to contend with.
In some ways, this means weโre actually aiming for two AHA moments in our customer journey. The first AHA moment is, โI can do the thing I came here for.โ
For example if youโre running an event and youโre looking to create an event registration form, then the first AHA moment might be that youโve copied a beautiful event template and can tell that itโs going to suit your needs.
The second, and more powerful AHA moment though is, โlook at all the other problems I can solve with Paperform!โ.
We rely a lot on customers self reporting what their needs are for the first AHA moment (we literally ask what kind of form youโre trying to create in our onboarding flow). But for the second AHA moment, we try to gently expose more and more of the capability and breadth of problems that the product can address, over time.
This means relying a lot on automated messaging, be they onboarding emails, or in-app tips, help documentation, or high quality video guides.
๐ค Founder social:
Thanks for reading! ๐ป
Team Relay (Chargebee for Startups)