P | P | P – feat. Shape

As part of the ‘Product | People | Potential’ series of interviews, we feature and showcase the very best UK start-ups with great potential, truly inspiring businesses that are shaking up their sector. As part of this, we caught up with Victor Palanco, Chief of Staff at Shape, a tool that eliminates the burden of ad-hoc data requests for analytics teams.

The purpose of article series ‘Product | People | Potential’ is to feature and showcase the very best UK start-ups with grand potential, truly inspiring businesses that are shaking up their sector. We capture and share the stories behind the name. We collate authentic peer to peer real talk, while celebrating the growth and success thus far and gather a glimpse of what’s ahead.

Can you please introduce yourself, what your business does, what stage you are at currently and what makes your business and offering unique?

Victor: Shape is a tool that eliminates the burden of ad-hoc data requests for analytics teams. It generates insights and does data retrieval for Product, Operations and Customer Success teams who are otherwise constantly at the mercy of analytics teams (who have better things to do than answer one-off questions) for support.

Shape connects directly to most relational databases. Under the hood, it analyzes the schema and generates various sample queries, stats and insights, which are stored as a special internal “context” to help more accurately answer users’ plain language questions.

We have been beta-testing our product with multiple customers for a few months now , and currently working on the most-requested features. For example, the human-in-the-loop verification system will allow analysts to oversee Shape’s responses (approve or intervene) or to turn it into a ticketing system for analyses, where each ticket would come with a pre-filled query and answer generated by the model.

Can you share the story behind the origin of your business and Product?

Victor: We’ve been working on Shape for just over a year now. However, the company originated in Y-Combinator, where Namit (CEO @ Shape) started working on “Collabkit” during the batch of S21. It was after the launch of GPT-4 that Namit realized that his previous project was soon to be disrupted by AI. Namit saw an opportunity to tackle the problem he and Dominick (Design @ Shape) had previously faced at Meta: Access to data.

Back at Meta they both worked as designers and often found themselves requesting data from the Data Scientists, who were often overwhelmed with similar requests for simple analytics about features. So you would often end up waiting up to a week to get the insights. Even with many self-serve dashboards that were available, there were often custom questions not covered by the dashboard. Initial product discovery research indicated that most Data teams allocated around 40% of their time to answering one-off questions, with some even spending 80% of their time! It is at this point that myself, Nick (Data @ Shape) and other key members joined the project and we’ve been hacking away since.

Moving to Product, what has been your approach to implementing product market fit or sales cycles?

Victor: Since we started working on Shape we knew we wanted to put the customer at the center of the development process. I have daily calls with prospects from all across our addressable market in an attempt to understand how they are solving their data needs today. All members of the team take calls with prospects and use this time to test assumptions they have about the problem we are solving. We have been doing this since way before we had a solution to pitch them.

More important even than discovery and prospecting, is paying attention to your clients’ requests. We construct our product roadmap primarily with our current customers in mind.

Not fully understanding your clients early on can send your product in the wrong direction. Moreover, we are constantly learning about how our clients use Shape. While some are accessing data on behalf of their clients, others are copying the SQL output and using it in various ways outside of the platform. It is our job to properly understand the user journey, and adjust our Roadmap accordingly.

And then Potential, can you share some challenges or barriers you had to overcome to create a Product / Service offering with potential?

Victor: In our space, there have been two key contingencies we have had to manage: accuracy and the expectations of accuracy. We often get potential customers connecting Shape to a very large warehouse with hundreds of table schemas and expecting accurate queries out of the box. However, in these scenarios typically some extra work is required to curate the datasets and input custom instructions.

Sometimes prospective customers try out Shape and see that it can generate data outputs that look correct, but not “quite right”. Conveniently, due to Shape’s transparent assumptions output with every answer, the user can very quickly track the inaccuracies to this or that assumption that they actually did not mean. Still, it is often a challenge to bridge the gap between the implicit assumptions (the internal language of the customer’s org) and Shape’s ability to immediately match those assumptions with no prior context. So the problem often switches from model-based query generation to finding ways of proactively inputting the client’s internal assumptions into the system.

Speaking of “accuracy”, we should realize that it is not often well-defined in the real world of business questions. Whether we are talking about a human analyst or an AI model, their typical stakeholder asks questions that may have multiple interpretations and missing assumptions. Consider the question: “What were the most returned products last year?”. There is no single “most accurate” SQL query for it. Most returned by absolute count or by return rate? Last year being the calendar year of 2023, or the rolling 12 months up to now? How many top-returned products are expected? Real world business questions are inseparable from the assumptions and caveats that need to be clarified. So Shape makes all of this transparent and puts it at the core of the query generation, and users can check and influence the assumptions going into each answer.

We are striving to make our AI produce queries and answers in a way that is as close to what a good human analyst would do – not just the aforementioned transparency in assumptions, but also the clarity of the query (using CTEs for readability, for example) and the succinctness of the summary (the tl;dr) it produces to answer the actual question. Additionally, our Slack integration allows asking data questions in shared Slack channels where everyone in the organization can see the answers. We strongly believe that Shape can empower companies by making the average decision more data driven, putting data at the center of many more conversations.