How about SLMs - Small Language Models?

 

Image by BNP Design Studio

I know large language models, LLMs as they are called in AI circles, are all the rage these days. But lately, I’ve been thinking of SLMs — small language models.

Not that I’m the first one to think small in this way. For example, when people are angry with each other, they prefer to speak in SLMs.

Exhibit A: The husband has a mere twitch of the lips when the wifey stops him right there. “Shut up! Don’t say a word!” [I wanted to flip the gender stereotype here and wanted the hubby to be saying the shut-up command, but something told me it wouldn’t fly, so I shut that thought out.]

SLMs are especially popular with other (often better) animals than humans. A coo-oo, clack-clack or screech often does the job better than voluminous human speech.

And then, sometimes, the smallness of some models becomes so infinitesimally small that no words — and hence no language, not to speak of models — are required at all.

A sharp look in someone’s direction, a press of the hand on a shoulder that needs pressing in a certain way, or a shared moment in silence is all that is needed.

Sometimes I wonder what’s going to happen with so many towers of LLM babble being created and business models being built on top (models atop models, huh?) We might even be drowning in a frothy alphabet soup without digesting a single letter.

And just as I was going on further with this post, a voice hit out at me from somewhere: “Time to shut up now!” (Thankfully, it’s my own goddamn brain and not a Neuralink implant freaking out.)

I looked at my analog watch and felt happy to see the second hand moving, regardless of what’s going on in the world.

Will AI help companies deliver better CX in a multi-experience future?

Image: Freshworks

Customer experience (CX) may mean different things to different companies but it means only one thing to customers: whether they liked what a company offered or not in a given context or setting. And whether they are going to have that offering again—and, yes, whether they would grab anything else the company wants to sell them. Perhaps they would also spread the good word on the product or service used.

Alas, in most cases, it is the bad word that gets thrown around—often wildly and out of the company’s control into the ruthless arenas of social media.

For the past several years, most organizations have responded by throwing back more and more technology to fix their CX efforts. According to research firm Gartner, global spending on customer experience and relationship management (CRM) software reached $48.2 billion in 2018, a growth of 15.6% over the previous year.

But, despite the rising expenditure on tech and the best intentions of companies, the struggle to get a handle on customers and delight them with exceptional support and service continues. So, what is going on here? What challenges are companies facing in putting together a complete picture of their customers and serving them better? Whatever happened to the promise that up and coming technologies such as chatbots and artificial intelligence (AI) were supposed to hold in equipping organizations with the wherewithals to delight their customers (and do so at lower costs)?

We spoke to a few industry analysts and experts to dig deeper and see what gives.

One of the fundamental problems, they say, has to do with the ability to use the right data to get a comprehensive view of the customer. “After all these years, having a 360-degree view of the customer is still on the agenda for companies. One of the issues here is, what do you understand by a 360-degree view? Is it an electronic Rolodex? Is it an extended set of data about the customer, something that different vendor tools are now increasingly exchanging? Or is it something else?” says Brian Manusama, a senior director analyst at Gartner.

On his part, he offers a simple, functional definition. “If you ask me, it is the right data in order to serve your customers well. It can have just two or three components or even hundreds of components, including different metrics such as customer sentiment or behavior,” he avers.

There are other aspects to this challenge. According to Ray Wang, principal analyst, founder, and chairman of Constellation Research, “On the one hand, most companies don’t have access to all their internal data. This lives in siloed departmental systems that rarely talk to each other. On the other hand, most companies now rely on more external data which is often seen as not secure, not as safe, and in different data formats. The last part is that data often does not tie back to business processes or journeys—which means it’s hard to determine a recommendation or next best action.”

The ‘recommendation or next best action’ typically refers to suggestive responses provided by the AI engine that is increasingly getting embedded in chatbots, CRM, and other business software. Such recommendations are based on a knowledge repository comprising standard answers mapped to frequently asked questions, previous customer interactions, etc. It is now common industry wisdom that for better recommendations, it is necessary to have a rich data repository and a finely tuned machine learning model.

Wang points to a basic flaw in how most organizations have traditionally dealt with customer experience. “Most [customer] journeys have been designed for internal efficiency, not external efficiency. Customers don’t care what department you are in and this means the design point must revolve around the customer,” he says. To correct this anomaly, a lot of organizations are now “retooling” to support this from an internal process and technology point of view.

Another big headache for companies is to make their disparate CX systems talk to each other and work as an integrated solution. Today, there is a dearth of holistic solutions that can manage the entire customer lifecycle—from acquisition to retention to life-time value (LTV) management. “There are different piecemeal offerings from different solution providers. For example, there are a lot of sales analytics companies out there who help sales teams optimize their processes; likewise, there are a lot of marketing attribution and automation software that have AI capabilities to help marketers spend their budgets more optimally and so on. Similarly, on the customer success side, there are tools for churn prediction and other areas, but the overall customer journey stack is broken,” says Swaminathan Padmanabhan, director of data science at Freshworks.

According to him, it will be of fundamental value to customers “if we can tie all these capabilities together.”

A multi-experience world
Customers are now interacting with brands through a complex mesh of interfaces and touchpoints—physical as well as digital. “Do you know how many different ways one can order pizza from Domino’s? Twenty four!” says Manusama by way of an example. Such ordering ease includes the use of phone, text, social media, and voice assistants, besides showing up at physical stores and giving the order over the counter.

“We are moving toward a multi-experience world with three different modalities of customer experience across multiple digital touchpoints—gesture, text, and voice,” he says. At Gartner, analysts now call upon tech leaders to get ready to serve ‘the everything customer’—one who requires conflicting things at the same time: to be treated like everybody else but served on their own unique terms, to be connected yet sometimes left alone.

When it comes to customer experience, companies are compelled to move from a reactive way of working to a more proactive way. And while this complexity is generally good for customers, as it gives them more choice and hands them greater control over how they want to be served, it leaves companies in a constant state of flux.

The growing role of AI
Analytics and AI are playing a more important role than ever in improving customer experience, according to Wang. “We are moving from gut-driven to data-driven decisions and this requires a ton of analytics to quantify and anticipate customer needs and requirements,” he says. The rising capabilities of AI offer hope to organizations. “Over time, machine learning will support precision decisions, which means better personalization, fraud detection, and customer  experience,” says Wang. He doesn’t hesitate to call AI “the biggest shift” in CX.

Padmanabhan refers to a six-layer maturity model of AI to lay out the path ahead for customer engagement. In increasing order of sophistication and capabilities, these layers are Data Representation Layer, Knowledge Layer, Ranking and Relevance Layer, Forecasting Layer, Recommendation Layer, and Autopilot Layer. In his opinion, most companies and systems today are operating at the Ranking and Relevance Layer.

“For example, when a customer query comes up, the bot ranks the different solution artefacts and suggests the best solution artefact. Similarly, when you have a bunch of sales leads, the lead scoring system ranks them according to their probability of conversion,” he explains.

As the AI system matures, one can expect AI-based recommendations such as “increase the ad budget by 15% to 20% for a 10% increase in customer acquisitions” or “use this workflow to optimize customer experience” and other actionable insights like these.

The pinnacle of AI capability, according to Padmanabhan, would be realized in the Autopilot Layer. As the name suggests, at this level, AI can replace some common functions performed by service agents or other team members. Rather than recommend something to be done, an AI can execute it as well.

 Not that AI will possibly replace humans fully—nor is that the direction taken by companies or recommended by experts. “Today, we don’t say that we are going to completely replace human labour but say there are a lot of repetitive tasks that are involved in the support workflow or the sales workflow or the customer success workflow which can be automated. So the agents’ time can be better spent by using AI,” says Padmanabhan.

Keeping the human element in customer engagement while still using AI is “actually a question of service design,” says Manusama. What customers want are four things in how they are served: effortless, quick, convenient, and seamless across different channels. “Many companies are discovering that they can do this through self-service. However, for more complex situations, having the human touch will often be more relevant or appropriate. Basically, companies need to answer this question: Where is the business value getting generated for my customers?” he adds.

Another trend he sees is customer service vendors consolidating their solutions into engagement clouds. “Silos that existed previously are getting broken down,” he observes.

Wang’s bet is on a future built around “ambient experiences”. What we have to ask ourselves, he says, is this: When do we automate, when do we augment with humans, and when is something a pure human interaction.

The role of engagement clouds or customer engagement platforms assumes greater significance in this context.  “We need common data models, great integration, and very good journey orchestration. You can do it in platforms or you can do it with really good tooling. I’m betting that the platforms will do 80% of the work and the tooling will carry the other 20%,” says Wang.

Whichever way organizations tilt, AI is likely to play a greater role in a multi-touch, multi-experience world. Now, depending on how they are able to lend a helping hand—through automation with a smile or by being pesky or ‘unintelligent’—customers will choose to give them a thumbs up or thumbs down.

(This blog post, which I wrote as a lead editor in the corporate marketing team at Freshworks, first appeared on www.freshworks.com.)

How Paytm Uses Tech to Manage 200 Million Users

Key points:

- Paytm processed 1 billion transactions in the quarter ended March 2018
- The firm employs 200 product managers and over 700 engineers
- Its data science lab in Toronto, Canada, develops key tech tools
- App analytics and machine learning are used to retain users and for up-selling

Mobile wallets--mobile apps used to pay for recharges, groceries and other daily items--may have come of age in an increasingly digital India, but much goes behind-the-scenes to keep them working well and users hooked.

Paytm, which has 200 million monthly active users and processed close to 1 billion transactions in the quarter ended March 2018, is a case in point. It competes with MobiKwik, FreeCharge, PhonePe and several others in this space.

Discussing the tech strategy at the company in a recent interview, Deepak Abbot (pictured), senior vice president of One97 Communications Ltd, which owns and operates Paytm, said, “Though a payments firm, we are a technology company at the core and everyone here, including Vijay, is a hardcore techie--he even calls mid-level engineers sometimes to discuss architecture design.” (Vijay Shekhar Sharma is the chief executive of Paytm.)

Abbot said that most in top positions at the company either have technology background or are “quite comfortable” with tech. “Culturally, we have a tech mindset. That is another reason we have been able to build a very complex product in a flexible way.”

Sharing insights into what goes on ‘under the hood’ as they say in tech, Abbot said that quick decision-making and a product-centric approach drive software development. “In our meetings, once an idea is crystallized, Vijay is very clear about what product to build. As a result, the product managers are also clear how to get it done. And when the engineers are given very specific details, they are able to quickly build it,” he revealed.

The simplicity of the Paytm app belies its complex architecture and the number of people that work on it. For instance, Abbot said that there are as many as 200 product managers and 700-800 engineers working on different aspects of the app.

But how does Paytm define a product? “At Paytm, a product is defined as anything a consumer—be it is an end consumer, a merchant or a marketplace seller--interacts with,” said Abbot. For example, recharge is a product in itself. Paytm’s implementation of Unified Payments Interface (UPI), again, is a product (UPI is an easy, instantaneous payment system built by the National Payments Corporation of India or the NPCI). “And then you build use-cases on top of UPI such as P2P, P2M and B2B payments. Wallet--the most used product of Paytm--is another,” said Abbot. (P2P, P2M and B2B stand for person-to-person, person-to-merchant and business-to-business respectively.)

The idea of keeping all these products within the same Paytm app, according to him, is that users should move from one product to another seamlessly—something that requires “a highly scalable product” to be built.

Integration of multiple products within the same app also helps Paytm cross-sell more easily to customers, who may first use one product before being “nudged towards” others, said Abbot.

Talking about stickiness of the app and up-selling to users, he said, “We have observed that if a customer has only used Paytm for recharge, then the retention rate for such a user is 40% after three months. But if we can upgrade him to send money to others, they become power users of Paytm and the retention improves dramatically to 70%.”

Industry experts forecast bright days ahead for mobile wallets. The number of mobile wallet users is expected to grow from the current 200-250 million to around 500 million in the next couple of years, according to Probir Roy, co-founder of Paymate and an independent director at Nazara Technologies. While he believes that “the next big thing” will be “interoperability” among different wallets, he noted that it is a tough space to operate in and some consolidation is “bound to happen” in the coming years. “My guess is that the top two or three companies will have 80% of the market,” he concluded.

------Paytm Labs: Managing customer lifecycles-----

To make the most of app analytics that capture user behaviour, Paytm’s data science lab, Paytm Labs, in Toronto, Canada, works on developing multiple software tools. One such key tool is CLM or Customer Lifecycle Management.

According to Deepak Abbot, senior vice president of the company who is based at Paytm’s headquarters at Noida near Delhi, what CLM does is “catch every ‘signal’ from the app”. Explaining how it works, he said, “If you use the app for UPI, it segments you as a UPI user; if you do a recharge, it marks you as a recharge user. It also upgrades you automatically based on your behaviour or purchase history. So, for instance, if you make an electricity bill payment or a post-paid bill payment, it upgrades you to a post-paid user.” There is a lot of granularity built into the CLM tool to classify and reward different levels of users at different times.

The tool puts users in different segments and generates actionable triggers accordingly. “For example, if a premium user who earlier made a money transfer of Rs 5,000 has not used the app for a month, he will be shown a cashback offer or an ad on Facebook,” said Abbot. Similarly, alerts are shown for soon-to-expire mobile recharges and other bills. “The CLM tool uses such alerts and offers to get those customers back into the app. And if they are already in the app, it will customise the view for them by showing up frequently used icons upfront and hiding others,” he explained.

The entire user data in the Paytm app goes into a “data lake”, and the team in Canada uses it to formulate the rules of the risk engine and other software. The data lake, Abbot explained further, is a repository of multiple sources of data, including phone usage data, hardware data and address book; then there is transactional data plus the behavioural data (where the users navigate inside the app, how much time they spend shopping, etc). All this data is used through machine learning (ML) algorithms so that the alerts and promotions can be automated and personalized.

The Toronto team comprises 70 data scientists and engineers and, besides the CLM tool, has developed the company’s risk and customer score engines. “We just plug those products here (in India) and start using them,” said Abbot.

---##----

(Note: An edited version of the above post first appeared on www.livemint.com - where I used to work until recently. The interaction with Deepak Abbot took place during my Mint tenure.)

Digital, AI tools easing up legal work for companies

Image: Pixabay.com

When Vaishali Lotlikar joined Wanbury Ltd’s legal department sometime in 2014, little did she know that locating a particular contract or assembling a legal brief would involve sifting through piles of documents, and wastage of precious hours and money in the process. “There was a lot of employee churn in the legal and marketing departments. Nobody really knew where the contracts were kept and what was there in each for the company to keep an eye on,” she recalls.

To streamline operations, Lotlikar and her team gathered all the contracts, digitized the same and put them into a document management system after tagging them for keywords, so that they could be easily searched when accessed from the firm’s servers by authorized personnel. The tool was purchased from PracticeLeague Legaltech Pvt. Ltd, a specialized provider of software and cloud-based solutions for law firms and corporate legal departments.

“I had used their technology at Glenmark and USV,” says Lotlikar, adding that the familiarity helped her get up to speed. “Today, if our international business head wants to know the particulars of a contract, I can get those details within minutes on my laptop—irrespective of the city I’m in.”

This is simply a case in point. As the volume of compliance and other legal requirements increases for companies across industries, technology tools that can ease the legal burden are in great demand. Khaitan and Co., for instance, has been experimenting with technology for quite some time now, according to its chief operating officer Nilanjan Ghose. “We were the first among law firms in India to use software for accounting. Around 2007, we started working with PracticeLeague to develop our own time and billing solution, which is core to our operations.” Over time, he says, the billing solution morphed into what is now known in legal circles as practice management software (PMS). He likens it to an enterprise resource planning software used for operational management by a majority of companies.

For Khaitan and Co., PMS helps in all kinds of processes, including accounting, billing, collections, administration and human resource functions. The company is now integrating new modules into it—an attendance module, for instance.

Digital tools are also helping law firms expand in size or scope. “Many of our clients acknowledge that it is because of technology that they could grow from a 50-60 person law firm to one employing 500-600,” says Parimal Chanchani, founder and director of PracticeLeague. And while there are several technology providers operating in the legal space—LexisNexis, LegalSoft, Thomson Reuters (ProLaw), Jurisnet and dozens of others—Chanchani says “nearly 60% of the corporate law departments and over 40 top law firms in India” use its software.

“You cannot manage a compliance workflow through Excel sheets; everything is now getting automated,” says Chanchani. “What we have is a complete, cloud-based solution sitting on Microsoft servers (Azure cloud). Customers can simply start using any module by just plugging into the platform.”

Role of artificial intelligence (AI)
PracticeLeague has also begun embedding AI into its software. For this, it has opted for Watson—an AI tool developed by International Business Machines Corp. (IBM). Praveen Kulkarni, who heads technology design and delivery at PracticeLeague, says Watson is implemented if a client wants to analyse a contract sent to it by, say, one of its suppliers.

For instance, if a firm wants to become the supplier of a pharma company, it will be required to submit several documents. Based on these submissions, the pharma company will send it back several documents to sign such as a non-disclosure agreement or a supplier registration agreement. If done manually, a person from the legal department would need to pick up the relevant content (from the submitted documents) and “draft and redraft the agreement that would take several hours”. With PracticeLeague’s Document Assembly, a Web link is sent to the supplier. “Once the required documents are uploaded through the link, the tool starts asking questions such as the category of supplier, payment terms, etc. After these questions are answered, a ready contract is automatically prepared through the system for the legal department to review and approve,” says Kulkarni.

However, if the pharma company wants an analysis or summary of the multiple documents it receives from suppliers themselves--the documents are exchanged by both parties for signing--then PracticeLeague uses Watson. What Watson does, explains Kulkarni, is “extract certain portions” of the agreement—for instance, contract type, liabilities or jurisdiction, or a termination clause. “So instead of manually picking up these details, they appear on the screen in front of the person reviewing them,” he adds.

But what if the system fails to understand any particular detail? For such situations, PracticeLeague has built an interface through which the reviewer can feed additional information back into the system so that the same can be picked up correctly by Watson the next time. “AI gets better with more and more data fed into it,” says Kulkarni.

Wanbury and Khaitan and Co. are yet to start using the AI tool, but acknowledge the role AI can play in further improving efficiencies for them. “While I have not used the AI tool, I believe it can automate repetitive tasks performed by legal professionals and also suggest the possible options to be taken in a legal case,” says Lotlikar of Wanbury. Nevertheless, she adds that while all of that can be done in the legal field, “strategies thought of by human beings are also important and cannot be fed into a system”.

“AI can help us in faster turnaround times for cases and in due diligence on contracts,” concurs Ghose of Khaitan and Co., but adds that human intervention and checking will also be required. “For example, certain words could be misspelt and thus be unreadable by the machine, or certain clauses could be interpreted differently. So you need somebody to go through the clauses manually,” he adds.

Other law firms using AI include Cyril Amarchand Mangaldas which signed up with Canada-based Kira Systems for the latter’s AI technology in January 2017. On its part, PracticeLeague is now working with Google and Amazon to integrate their AI technology into its solution and, after that, plans to work with Microsoft as well.

(Note: The above article first appeared on Livemint.com.)

Artificial Intelligence has long way to go but it's already creating much value: Neil Jacobstein

Neil Jacobstein chairs the artificial intelligence (AI) and robotics track at Singularity University on the National Aeronautics and Space Administration (Nasa) Research Park campus in Mountain View, California. A former CEO of Teknowledge Corp., an early AI company, Jacobstein was in India recently to speak at the two-day SingularityU India Summit (held recently in association with INK, which hosts events such as INKtalks for the exchange of cutting-edge ideas). In an interview, Jacobstein talks about the confusion around AI, how job losses from AI should be tackled and the possibilities of a brighter future for humanity. Edited excerpts:

There are several definitions of AI. Which one is your favourite?

Artificial intelligence allows us to create pattern-recognition and problem-solving capability in a computer, using software algorithms. AI allows us to tackle practical business and technical problems, and it presents an opportunity for us to allow computers to do things that previously only humans did.

There seems to be a lot of confusion about what AI can or cannot do. What is your reading of the prevailing situation?

I think part of the confusion in the market might be that science-fiction movies have given people very vivid and sometimes incorrect view of what AI is capable of doing. Today, we have AI that is already at human levels of problem-solving in very narrow domains such as chess or go (a Japanese board game) or certain kinds of medical diagnostics. But we don’t have human-level AI that is general across the board. So we don’t have AI with natural language understanding at human level, and we don’t have AI that has humour or empathy at human levels. So it’s a kind of mixed landscape.

When do you think will we achieve “true AI”, so to say? What are the challenges to be overcome?

We have already achieved true AI in the sense of creating problem-solvers that add billions of dollars of value every year to various industries. That’s happening now. But if you are referring to artificial general intelligence that is at human levels, I think that probably won’t happen for several years: it could be as early as mid-2020s or as late as 2030s. The critical thing is not the time frame but the consequences of having AI at a human level and what that means for jobs, for global security, and for opportunity to solve problems.

While there are those who believe in the potential of AI and its applications, a sizeable number— including Stephen Hawking, Bill Gates and Elon Musk—have expressed fears that AI-powered machines could rule over humans. What’s your take on this?

To his credit, Elon has changed his views on this over time. He has invested over $1 billion in an entity called OpenAI to democratize access to AI and to create new AI test beds and capabilities that will allow us to build layers of control into AI software. He has also participated in creating conferences on the future of AI and sponsored Future of Life Institute’s conferences around developing new principles of AI safety, the so-called Asilomar 23 principles (futureoflife.org/ai-principles). So he’s interested in capturing the benefits of AI and wants to help us work systematically to reduce the downside risk.

There may be an alarmist element to job losses resulting from AI, but robots are indeed replacing humans. How do you think should the situation be handled?

I think there is a need to anticipate things and to have some empathy and foresightedness for people who will be affected by job losses. For one, the quality of life for the rich people goes down when there are a lot of angry and alienated and armed people around. So it makes sense to think ahead as to how we can educate people doing routine jobs now and, in anticipation of problems downstream, provide access to free, high-quality education. Not everyone will take advantage of that and not everyone will achieve high levels of skill in some new job. So it makes sense to have some kind of basic minimum income and there are different potential schemes for doing that—but nobody knows the exact answer to this.

While Peter Diamandis talks optimistically about the future in his book Abundance, there’s a widening gap between the rich and the poor? Do you think a technology like AI can bridge this gap?

I think rather than focus on the gap, it would be better to focus on the quality of life metrics: do people have access to high-quality, nutritious food? Do they have access to first-rate education or clean water? If you look at the evidence for abundance on Peter’s website or read Steven Pinker’s book, The Better Nature of Our Angels, what’s clear is that in some respects, we are living in the best times for humanity. The challenge is to create a world where, instead of having a world of haves and have-nots, we have a world of haves and super-haves. Now, the gap between haves and super-haves might still be very big, but the haves will at least have things they never had before.

You have spoken about the huge impact of atomically precise manufacturing in nanotechnology. When will it be achieved?

The kind of nanotech we have today is mostly materials science; it’s not molecular machines or atomically precise manufacturing. But I do think we will eventually have atomically precise manufacturing, as we know it’s possible and researchers have demonstrated in the lab the ability to manipulate atoms and molecules with precision. What’s missing is to do it at industrial scale; that may take years.

(Note: This interview first appeared on www.livemint.com.)

7 Reasons Why the CIO Job Can NEVER be Automated

(Image: Pixabay.com)

Dear CIOs, I know most you must be sick and tired by now of hearing all kinds of stories about automate this or automate that, AI, machine learning, deep learning, etc., etc.

It is possible that someone might come up with the idea of “Hey, why not automate the CIO’s job itself?” After all, haven’t we all seen too many threats to the CIO role already, even without automation?

So, here are a few semi-serious reasons why you don’t have to worry about the CIO role self-driving itself into an auto-pilot system:

- Because no AI system can mouth words like “silos,” “vendor-neutral” and “scalability” with as much elan as a CIO.

- In all likelihood, it was a CIO who coined the very concept of automation; so the conceived cannot possibly turn against its conceiver (and be successful in their machinations).

- Because you need someone with real, rather than artificial, intelligence to control all those bots out there.

- Think about it: if one were to indeed automate what CIOs do, then who will complain about budgets being tight or “doing more with less”!

- Because if CIOs get a hang of the conspiracy to automate them out of the market, they will reset the code execution time for the automation function to forever (or eternity, whichever happens to be programmed into their software :)

- Because before “they” can automate it (whoever this they refers to), the CIOs would have reinvented their roles as CAOs (rhymes with cows but otherwise of a different temper)—which stands for Chief Automation Officers!

- And if none of the above works, CIOs will convince the automatons to outsource the thinking part to them and save precious battery resources for other artificial work!

IMHO, forget CIO, I personally don’t think bots will truly replace human beings. They might take up the tedious or repetitive work being done by an army of workers, but they are highly unlikely to manage or motivate teams, lead people by example, inspire trust or become an emotional wonder-pack that humans are.

Now, tell me bot do you think?