Blog

Serverless GPUs improve GenAI TCO by 200X for end-to-end RAG solutions

Aible — Wed, 02 Oct 2024 21:41:04 GMT

Overview:

What’s New In Generative AI with Aible

Arijit Sengupta — Mon, 20 Nov 2023 21:23:36 GMT

I have spent much of my career fighting dashboards – those ubiquitous but inflexible means of looking at the same metrics every day. Augmented analytics also known as “smart data discovery” offered a way to use AI to automatically find what we needed to know, not just what questions we knew to ask in a dashboard. For example, a retail dashboard may include charts of sales by geography or sales by product, but the key insight in the data may be that a specific product is doing especially well in a specific country with a specific demographic. The dashboard would not show that insight unless by chance the analyst had filtered down to that specific pattern and included the corresponding chart in the dashboard. The problem is that there are millions of such possible ‘group-by’ and ‘drill-down’ charts even with a simple dataset and a human analyst simply could not evaluate them all. Thus the need for AI. BeyondCore (my previous startup, now Salesforce Einstein Discovery) was one of the first companies that delivered Augmented Analytics and was supposed to spell the end of dashboards. The problem was that even when the right charts were sent to the user, they needed help with interpreting the charts and asking follow up questions. Thus, while we automated the analysis tasks of the analyst, we failed to address the translation tasks. In essence, we failed to address the last mile problem. Thus, the need for GenAI. And why this Halloween release we launched the Aible GenAI-powered Natural Language Query (NLQ) dashboard. Trick or Treat!

In our user research we discovered that NLQ dashboards were a great way to bridge the gap between what is familiar (the dashboard) and the unfamiliar (GenAI NLQ). Essentially the user starts out with a familiar dashboard with multiple examples of GenAI NLQ questions and responses and then can just copy and edit those questions to ask their own. Aible uses AI to also address the key problem of NLQ solutions – what questions should we ask? Knowing the right question to ask is the biggest part of the problem and the Aible GenAI NLQ dashboard starts the user off with several good questions to riff from. The starting questions can be selected manually by central analytics/IT teams, based off popular questions from peers, or automatically recommended by Aible analysis.

Aible also addressed a key problem with GenAI NLQ – making sure it is answering the right questions accurately, addressing the biggest challenge around GenAI hallucinations. Other solutions use a single model to translate the users’ questions into Structured Query Language (SQL) and then execute that SQL to produce the answer. They provide the SQL to the user so that they can check that the SQL looks correct. The problem is that most business users don’t know how to read SQL and would have no way to know whether it is correct. Given high hallucination rates for the GenAI models (look up exact percent for NLQ) there is a significant chance that the model misunderstands the question and translates it into the wrong answer. As long as the answer is plausible, the end user has no way to check on the answer short of learning SQL. Instead of getting a narrow response to the NLQ, if business users were given additional context to the response, it would make it easier for them to check the validity of the GenAI response. Aible provides the added relevant business context to the response by using two models – one to understand the user’s question and answer it exactly, and the other to generate a SQL query that is slightly broader than the user’s question and includes information that provides additional context to the user to enable them to do a gut-check of the response. In our user testing we found that, for example, if a user asks for revenue from Germany that year and the AI gives a single number answer, it is hard for the user to detect errors. However, if the AI gives the answer but also includes the revenue numbers for Germany for a few more years for context, or includes the revenue numbers from other countries for context, the user is able to detect AI errors much more effectively. In essence, context is key in helping business users detect GenAI hallucinations.

Aible’s GenAI dashboard also addresses the fundamental issues of traditional dashboards and NLQ:
Complex manual setup: NLQ systems have always suffered from the need to do complex metadata setup before they could be useful. It turns out GenAI models are very good at figuring out the data dictionary without any manual setup. Aible thus reverses this process by automatically doing all necessary setup from raw data. If it gets something wrong, it can figure it out quickly based on user feedback. Finally, if organizations have unique ways of defining things – fiscal quarters for example – they can simply define that in the Aible chat template and then the GenAI is provided that information when necessary. Aible thus requires no or very minimal setup to get started.
Dashboard fragmentation: Organizations often end up with hundreds of dashboards that are simple variations on a core dashboard. Essentially different executives insist on one or more additional charts and the analytics team are forced to create a new variant of the dashboard for each such request. With Aible’s GenAI NLQ dashboard, users can simply ask and add a few more questions to their own dashboard without affecting other users. They can also share the modified dashboard with others, or if they have the correct permissions, they can even update the original dashboard for everyone.

High cost and long response time: In our analysis of user questions at larger organizations, we found that there was significant overlap. Once we looked at the underlying themes of the questions – because there are many ways to ask essentially the same question – there was even more overlap. In traditional approaches, when such redundant questions are asked, you incur the GenAI model response cost as well as the cost of querying the underlying data. Aible however detects that essentially the same question has been asked recently and responds from a cache without incurring the unnecessary cost. The average response time is also much faster as a result. Of course, users can always turn off the caching response if they want more up to date answers. Others then benefit from the more up to date answers when they ask similar questions.

Finding the right dataset or dashboard: In a world with many datasets and dashboards, users often can’t find the right one to ask questions of. Aible solves this by enabling users to ask their questions outside the context of a dataset, then searching for available datasets/dashboards that may help answer the question. Users can then choose from a shortlist of dataset/dashboard choices to get the specific answer to their question.

In enterprise software we often end up with these ‘past is prologue’ moments. The boring old dashboard may turn out to be the perfect bridge to the GenAI powered analytics revolution that will empower every business user to unlock the value of their enterprise data.

See this 30-minute session with Bo Rankin from Baptist Health to see how they are systematically taking their existing Tableau dashboards and systematically unlocking the value of that data via Augmented Analytics and Generative AI:

5:20: “Its finding some really cool surprising patterns that … would have taken us weeks or months or we would have never even thought to analyze – and it just pops off the page.”

9:20: good examples of business KPIs to start from

20:20: “with other vendors, they said you can do generative AI in your enterprise but you have to do a whole bunch of work in your data warehouse, the data catalog, and that was not our experience at all with Aible. It was literally just, here is your dataset, run generative AI on it, and it was immediately relevant… It was a completely different model from other solutions we have been talking to.”

22:55: “I would put the natural language query in front of [the business user]… And now I have put myself out of the dashboard business. Where they can get whatever question answered that they want right on the data itself and they don’t have to go ‘where was the dashboard again?’”

See this nine minute video from the Gartner Summit that explains the Aible approach and includes customer commentary from Cisco, UnitedHealth Group, and NYC Health & Hospitals
Best Regards,

Aible Partners with Google Cloud to Lower Cost of Analysis By 1,000x and Time to Results from Months to Days.

Aible — Tue, 14 Mar 2023 12:36:42 GMT

Benchmark from a Fortune 500 company shows 1000X improvement in cost ($80 vs an estimated $200K with traditional approaches) and 10X improvement in analytics/ data science project time (14 days vs. 4-6 Months) on Google Cloud.
Customers can now implement Aible on their Google Cloud account in 15 minutes via the Google Cloud Marketplace

San Francisco, Calif. – March 13, 2023 – Aible, the only enterprise AI solution that guarantees impact in 30 days, is now available on the Google Cloud Marketplace. With this launch, enterprises will be able to securely deploy the end-to-end serverless solution for augmented data engineering, analytics, data science and machine learning in their Google Cloud accounts within 15 minutes.

Benchmark for Speed and Cost of Analysis From Fortune 500 Customer

“For too long we have analyzed only a small percentage of the available data. Why? Because analysis today is human-first and expensive. Aible enables true AI-first analysis with dozens of case studies we have published with Google, Intel, and others demonstrating users were able to analyze data completely automatically in minutes. But we were always asked - what if the automated analysis revealed the data was not worth analyzing? Shouldn’t we get our data in perfect shape before we analyze it? This benchmark study proves that the cost of automated analysis is now so low, that you are better off just analyzing data first and seeing what the data quality issues are, than waiting to analyze the data. This is the exact same dynamic that played out in the shift from Yahoo to Google. For too long we thought only expert humans could categorize the web - and as a result only a small portion of the web was categorized. When Google fully automated that same process, and cut the cost of evaluating a page by 1000X, the market shifted to evaluating every webpage. We still use humans in the loop at the end of the process, where most of us click on just a few of the search results instead of choosing ‘I am feeling lucky’ but the automated approach always dominates when the cost gets low enough. This is that moment in the world of analytics.” said Arijit Sengupta, Founder and CEO of Aible.

“Aible’s platform provides important tooling and capabilities that helps business users, who are a whole new type of AI developers, better and more efficiently utilize their data,” said Manvinder Singh, Managing Director, Partnerships at Google Cloud. “We’re excited to make it very easy for AI customers to deploy the Aible platform on Google Cloud, and to support Aible’s architecture, scale, and model training with Google Cloud infrastructure, BigQuery, and Vertex AI.”

This absurdly low cost of analysis was not a fluke. Similar benchmark results from other enterprises including a Fortune 500 technology company, a Fortune 500 telco, and a top 5 airline will be featured at the upcoming Gartner Data & Analytics Summit, in Orlando FL on March 20-22, 2023.

Leading Enterprises Migrating to Google Cloud to take advantage of Aible’s end-to-end Serverless AI and Vertex AI

With its latest product release, Aible combines an end-to-end serverless architecture and in-browser Information Model manipulation with serverless queries in Google BigQuery to cut augmented analytics costs to 1/1,000th compared to other cloud data warehouses and BI tools. Several enterprises across industries who leverage Aible have migrated their analytics backend databases to Google Cloud to take advantage of serverless analysis of any size data which translates into both cost savings and speed to results. Additionally, with the ability to automatically create dynamic Looker dashboards in seconds without writing a single line of LookML code, leveraging BigQuery without coding, and even using Vertex AI to train models without any coding, Aible empowers business users to easily leverage the power of Google technology. Aible also enables scalable collaboration across user types to help organizations build consensus and act faster.

To meet representatives of these enterprises that were part of benchmark evaluations at the Gartner Data and Analytics Summit in Orlando, FL, March 20-22, 2023, schedule a meeting here.

Aible now available on the Google Cloud Marketplace

The following Aible products are now available on the Google Marketplace:

Aible Sense (Augmented Data Engineering)
Aible Explore (Augmented Analytics and Automated Dashboards for Looker)
Aible Optimize (Augmented Data Science and Machine Learning)

Google customers or enterprises looking to improve their speed and cost of analytics can now implement Aible on their Google Cloud account in 15 minutes via the Google Cloud Marketplace. Details available here.

What's New in Aible

Aible — Thu, 14 Jul 2022 20:55:07 GMT

July 4th Product Release

Ever since we launched Aible Sense on Valentine’s Day this year (our “features you will love” release), customers loved the automated features in Sense such as automated data cleansing, automated feature creation, automated data evaluation, etc. They also loved the fact that all of this happened by bringing the code to the data that remained in their own private cloud, via Python code in Jupyter notebooks that they could access and reuse. They loved the security model with built-in k-anonymity that meant Aible users could not access the raw data that they evaluated, analyzed, or created models based on.

But, customers asked: “why can’t we do manual data engineering the same way you are doing automated data engineering?” For our Fourth of July release (“features that set you free”) we decided to give them that ability, while still leveraging Aible’s augmentation, data security, and end-to-end auditability capabilities.

These features are automatically available to Aible Explore and Optimize users today at no additional charge. Existing Sense-only users would need to upgrade to a Sense+ license to get these augmented data engineering capabilities.

Market Guide for Multipersona Data Science and ML Platforms

Aible — Mon, 23 May 2022 18:03:23 GMT

Aible: The Only Vendor in the Gartner CAIDS Magic Quadrant That Guarantees Impact From AI in 30 Days

The latest Gartner Cloud AI Developer Services (CAIDS) Magic Quadrant and Critical Capabilities reports just came out. Aible did very well against some incredibly well funded companies such as Google, Microsoft, Amazon, Dataiku, etc. In fact, our score in Automated Machine Learning (AutoML) was within 2% of the top score in the category (3.62 vs. 3.7).

However, this underlying market is shifting significantly right now. On the data & analytics side of Gartner, they even discontinued the Data Science and Machine Learning MQ as they navigate these shifts. We are extremely excited about Gartner’s focus on the Multipersona Data Science and Machine Learning Platforms space which reflects the needs of the business user / citizen data scientist / analyst personas. We completely agree with Gartner on the importance of delivering business impact and collaboration across personas. Please read the new Market Guide for Multipersona Data Science and Machine Learning Platforms. It is well worth the read. We expect to be part of this market once it turns into a Magic Quadrant in the near future. Some of my favorite recommendations in the document include:

How to Eliminate Bias in AI

Aible — Fri, 25 Mar 2022 21:34:08 GMT

How to Eliminate Bias and Ensure Fairness in AI.

Eliminating bias in AI has emerged as one of the most important challenges in technology. With AI increasingly being used to make important decisions in areas such as hiring, college admissions, medicine, and bank lending, it’s critical that these decisions don’t discriminate based on race, gender, age, or other factors.

The stakes are high. A Gartner report predicts that through 2022, 85 percent of AI projects will deliver erroneous outcomes due to bias in data, algorithms, or the teams responsible for managing them. Bias in AI leads to unfair outcomes, the perpetuation of systemic racism, and serious reputational damage and costly failures for businesses.

Unfortunately, many of the approaches that aim to mitigate AI bias don’t fix the problem. On the contrary, they often perpetuate bias and make it even harder to root out. Reactively solving these problems via best practices and governance is unlikely to scale. We need to solve them by design.

Consider this example in bank lending. If historically, a bank has not lent money to certain racial groups or genders as often as others, and the AI trains on that historical data, the recommendations of the AI will perpetuate the bias. That’s because AI learns only from data. If there’s bias inherent in your data, that will become further institutionalized when you use it to train your AI.

The conventional “solution” to this problem is to remove the race or gender variable from the AI so that it can’t directly consider those factors. Problem solved, right?

Wrong.

Strategy-First AI: The Key to AI Success

Aible — Fri, 25 Jun 2021 23:59:02 GMT

An MIT-BCG study found that a mere 10% of organizations achieve financial benefits with AI. This is because businesses often unwittingly deploy a set of tactics with their AI that’s in conflict with their overall strategic objectives. That can result in AI that actually destroys value. And maybe even put you out of business.

I saw this first-hand when my co-founder Jonathan and I were working on a project with his previous company. Jonathan used BeyondCore, my previous product, to analyze sales for an industrial products business. Jonathan’s company had introduced a lead scoring system based on a traditional rules-based model. The system performed well initially, making it easier to identify high quality deals. But over a period of three years, the revenue from new deals declined steadily and nobody could identify why.

BeyondCore revealed something very strange about his data. When decision makers were directly involved in the process, BeyondCore found it increased the likelihood that a deal would go through. It was also a good thing when opportunities were for new design wins – cases in which a component is designed to go into a new product, which leads to ongoing sales. But for some reason, the combination of a decision maker being directly involved and a new design win was a bad thing.

Lengow Case Study

Aible — Mon, 21 Jun 2021 22:14:47 GMT

Aible Helps Lengow Identify Customers with 50X value with AI

Lengow is a successful Paris-based e-commerce automation SaaS solution that helps more than 3,600 merchants across 45 countries with their e-commerce strategy, by enabling them to optimize product data and sell their products worldwide on all distribution channels: marketplaces, comparison shopping engines, affiliate platforms, display, and retargeting.

3 Reasons Supply Chains and Manufacturing Should Consider AI

Aible — Fri, 18 Jun 2021 03:41:12 GMT

Supply chains continue to face unprecedented challenges.

According to a recent survey by the Institute of Supply Management, manufacturers face “record-long lead times, wide-scale shortages of critical basic materials, rising commodities prices and difficulties in transporting products across industries.”

The effects of Covid-19 on business supply chains continue to be long and deep, even as economies have opened up. Increased demand is posing unique challenges, as inventories and production have to be increased. Since the pandemic hit, 28% of companies have faced shortages and stockouts, while 56% of companies have had to renegotiate contracts due to supply chain disruptions.

3 High-Impact AI Use Cases for Higher Education

Aible — Tue, 25 May 2021 06:14:09 GMT

Higher Education Is Facing Unprecedented Challenges.

In the last decade, the US has seen an enrollment decline rate of over 2 million students. Now in the wake of Covid, institutions are faced with a fresh set of problems around student enrollment, student and faculty retention, donor engagement, marketing spend optimization and more. Some institutions are reopening all of their classrooms, but with new safety protocols in place. Others are adopting a hybrid model, with a mix of online and in-person instruction. Still others are accommodating students and teachers who prefer a 100% online experience for the foreseeable future. That means colleges and universities increasingly need robust scenario planning so they can adapt quickly to rapid changes in the months ahead.