TADSummit online Conference, AI in the Middle East, Karim Zaki, Unifonic

Time flies! Preparing for a more Complex Future.

It’s been 2 years since we last talked with Karim, This session is long overdue, and shares an open and honest review on the status of programmable communications in the Middle East, with a focus on AI

Back in 2023 Karim gave a session on Unifonic’s origin story and programmable communications in the Middle East. https://blog.tadsummit.com/2023/10/26/tadsummit-2023-review/

Back then Unifonic had reached 500 employees and 1B+ transactions per month.. Since then they’ve doubled the number of transactions, which given the declines in some markets is a great achievement, and focused on engineering and sales talent while keeping the overall employee numbers the same. We’re going to discuss tackling the current AI engineering challenges later in this session.

Important trends in the region were: automation / AI, conversational channels, and personalization. Unifonic now focuses on three core areas: customer care, marketing automation, and personalization (to the individual thanks to the scale and automation of AI).

You’ll hear through this podcast how Karim has cemented Unifonic’s strategy, and preparing for a more complex future. In Saudi Arabia the popularity of X continues to surprise me. However, TikTok and Snapchat are important conversational channels for marketing that can not be ignored. Look at the popularity of Labubu (an ugly doll), to see their social impact and conversion into sales.

The Shift to Conversation Channels

The shift in the past two years is from telco channels (SMS) to conversation channels. For example, WhatsApp, and RCS depending on who its selling it. SMS was a product, there was no need for enablement, or building out SaaS solutions.

With a conversation channel, enablement software is required. That is the AI agent, its integrations, insights, work flows, deployed use cases, etc. The interesting thing discovered with agents is people expect them to be available anytime. This requires automation and multimedia, to deliver a compelling experience for customers.

I highlighted the shift I’ve seen in India, given the popularity of WhatsApp and the integrated AI agents that deliver an atomic platform for conversations in customer care and marketing automation. It’s an experience that is easy to deploy and for customers return to.

This is lacking for RCS, the AI agents, integrations, monitoring, etc. It’s easy to claim ‘others’ will fill the gaps. Compare the use cases implemented using WhatsApp and RCS, it’s a significant gap. In the US companies like Clerk Chat are attempting to fill the gap. However, in markets outside the US, the maturity of WhatsApp, Line, Wechat, etc, have created a significant barrier to RCS.

Karim highlighted some important gaps during the initial implementation of AI agents. For example ‘Click to WhatsApp’. making it easy to contact an agent from an advert. And the monitoring to see where an individual’s workflow did not complete, so it could be re-established minutes, hours, even days after. Depending on the context.

Conversation AI was a beginning, then the journey expanded to analytics, ROI (Return on Investment) monitoring, marketing, and agent performance. For example, rating re-establishment performance based on workflow completion.

Fraud over Conversation Channels

Earlier in the day I had a conversation on Linkedin about the issues of fraud and spam over SMS, pushing customers away from the PSTN.

The answer is not simple comparing IP and PSTN messaging, as it depends on the platform and country. However, Karim was clear the GCC (Gulf Cooperation Council, (Saudi Arabia, Kuwait, the United Arab Emirates, Qatar, Bahrain, and Oman)) are particularly sensitive to spam and fraud.

Hence the platform providers (outside of SMS) make concerted efforts to protect users. If you’re using the legit WhatsApp APIs, its guarded. I see that in the US, WhatsApp at present has no SPAM, while SMS is extensively polluted. Karim backed up what I’ve see and been told from GCC customers.

RCS in the Middle East

Karim has been strong on RCS having a role in the Middle East, especially when its favored by regulation. Telco adoption limited the rise of RCS, coupled with WhatsApp’s slow rise in adding A2P use cases. Remember, A2P revenues are relatively small for carriers, and PSTN spam / fraud remains a significant problem. Hence the complex decision matrix, that seems to end with indecision.

Karim favors multiple channels for innovation. However, he brought up the rise of Snapchat and Tiktok, even Instagram and Messenger for marketing use cases. While Viber may be #2 in some markets, e.g. Eastern Europe, it is not in GCC, it does not exist. In Saudi Arabia X, previously known as Twitter, is number 2. Regulation has a role, but there appears to be less clarity on direction in sectors such as banking.

All the hyper-scalers are adopting sovereign cloud, so within one year the regulator could endorse a broader decision across sovereign cloud. Enabling banks to use a variety of messaging platforms.

Given history, local hosting from WhatsApp could appear faster than RCS. Remember the gaps in RCS versus WhatsApp, and the need for carrier coordination, and the complex decision criteria carriers face.

Simply, RCS is too little, too late if it does not happen soon. And there are a variety of consumer focused platforms available that have solved the developer, multi country support, market segment focus, and sales. Would Labubu use Tiktok or RCS,? The answer is clearly Tiktok as that’s where it’s target customers reside. With likely Snapchat or Instagram as its number 2 platform.

What’s working in AI?

We then focused in on what’s happen in AI since we last talked. Unifonic brings invaluable learning. The 3 areas are customer care, marketing automation, and personalization (to the individual).

AI can not be sold as a technology. It must be packaged as a solution, that delivers a clear business benefit. From which a subscription with fair use can be charged.

We covered the silliness of the current Microsoft co-pilot adverts. Which in 2025 shows something similar to FaceApp in 2017. Karim highlighted, Microsoft is a master of bundling. A fee could be claimed for co-pilot, but customers are not buying copilot.

This is the challenge the industry faces with AI appearing “free” from the model providers. However, it’s a significant cost, and both parties (Unifonic and its customers) must work in defining the fair value for specific use cases in specific verticals.

This reflects the current situation of discovering enough value to agree a fee. The tertiary use cases are then icing on the cake. Think of it like charging for a database used in your platform. Customers would not expect it to be broken out, they may want to see a BOM (Build of Materials), but that is for security not pricing.

The focus is on the 3 core use cases: customer care, marketing automation, and personalization (to the individual).

A challenge for many companies in Unifonic’s position is the foundational models are presented as commoditized, almost free. That is most definitely not the case, speech to speech models are more than the cost of some platforms at the moment.

AI by itself is generally not saleable, the integration into the platforms, and data, and guard rails, and workflows is required. BEFORE its value becomes clear. This is why we see the slow adoption, lots of hype with a copilot that makes you look like an anime character. However, delivering an experience that keeps customer coming back again and again takes time, and the experience in creating that is invaluable.

The value is in engineering teams implementing customer care enhanced with AI, not in AI engineers earning $100M pay packaged. Remember Meta’s claims on the metaverse. And that was before any value had been demonstrated. At least this time around we are seeing value, it’s just working on the use cases with customers to build confidence based on experience, not wishful thinking, Though the Microsoft co-pilot adverts leave me worried, as I’ve used Midjourney for years to create hero images for my thought leadership. If that’s the extent of AI, we have a problem.

How to sell AI?

Subscription with fair usage was the starting point. This is a joint discovery process with the customer. The target is outcome based pricing, across use cases and verticals. But the market is not there yet.

Branded Calling

I hopped over to branded calling. The US, has a problem of cost, 12c USD per call. Most of my calls are labeled spam-likely, about 50%.

Karim highlighted in the Middle East, businesses are required to state the purpose of the call. For the benefit of the customer. And it certainly is not 12c, it’s a regulatory requirement for the benefit of the customer.

This shows a problem the US faces. A regime focused on trying to make money on the PSTN at every turn, yet making the PSTN ever worse. I think there needs to be a fundamental shift in US PSTN regulation and business model as its killing the business.

The WhatsApp Balance

The pollution of the PSTN is not the case for WhatsApp in the GCC. WhatsApp governance protects users from excessive A2P messaging. Because customers have a choice, they adopt WhatsApp because of the value of its communities. As soon as the value degrades, people move on.

The US WhatsApp is quiet, it’s only conversations with people and businesses I know. While I hear for some in the UK its noisy. My sister and her family find UK WhatsApp similar to my US experience. It’s the platform we use for our family chats.

Shifts in Customer Engagement

From transactions to contextual customer engagement, That is, using the insights gained from my last workflow that did not complete, so I’m asked if I’d like to complete that.

Proactive engagement, using agentic AI to get to the point faster with a relevant offer. For example, I’m completing a purchase, and I’m asked if I’d like to add sticky tape to the order as the last transaction that did not complete and missed that item.

It’s about getting to relevant results faster for an individual customer

How AI Impacts Engineering.

This is driven by the 3 domains, of customer care, marketing automation, and personalization (to the individual). For example, new AI tech, new case studies, new experiences are evaluated across these domains, and whether they can have an impact for the customer.

Arabic LLMs are a focus, this is part of Unifonic’s moat in being the best provider in the Middle East. This also includes being culturally appropriate, especially in customer care. Things can be said or offered by an agent that works in the West, maybe less appropriate in the Middle East,

A critical skill is balancing build, versus buy, versus partner. Partnering with a foundational AI model provider is obvious, or with infrastructure heavy tools. Build where the customer data or business logic gives you an edge.

We were having so much fun that we ran out of time. Karim shows the importance of TADSummit in sharing experiences and insights that help everyone in the industry. Its an essential detox to the rampant BS that is wasting billions on pipe dreams like network APIs, and RCS outside a few markets like US, Germany, and the UK. One of the many things Karim taught me is the emerging importance of messaging platforms like Tiktok, Instagram, and Snapchat.

Leave a Reply

Your email address will not be published. Required fields are marked *