AI Use Cases for SMEs: GTA AI Quarterly Report (2026)

The GTA AI Quarterly Report documents AI use cases for SMEs that are already in production, told by the four practitioners who built them. AI Central produced it with Global Tech Advocates (GTA), and the first issue came out in October 2026.

Every use case answers one question: where does the machine stop and a person take over? Every number in the report was given by the person who built the system.

It is the first issue of a recurring series from Global Tech Advocates and AI Central. GTA is a volunteer-led, not-for-profit tech community of over 40,000 business leaders, experts and investors in more than 60 locations. The report comes from the GTA AI Quarterly session, where members compared AI systems they have put into production and what they learned from them.

The people behind it:

In the foreword, Russ writes that what he hears from tech leaders, founders and working groups is that "AI has firmly moved past the speculation and into real-world commercial deployment."

The report opens on the distance between what leaders expect from AI and what they can show for it. The McKinsey and Google Global Marketer Survey (July 2026, about 500 marketers) puts numbers on it:

Ganapathy sees the gap in the pricing conversation. A client once told him: "You have quoted X on the project. I have read a report which said AI is able to automate 75 to 80%. Can you do this project for 15 to 20% of the quoted price?"

The market already tested the fully automated version. In 2024 a wave of AI SDR tools promised fully automated outbound, and Tsvetelin's team tested two or three of them. "They were sold as a finished product and they were still very much in MVP stage," he said. Within 12 months revenue leaders went back to the copilot model, with a person signing off every draft before it ships.

Power Belt is a European wholesaler with around 200 employees. Between 700 and 800 supplier documents arrive every month, mostly as PDFs by email, and a complex one took around 150 minutes by hand. 90% or more of that time went into matching supplier product codes to the internal product ID list.

The agent reads the PDF, extracts the product codes, quantities and line details, and checks them against the product ID list. When an ID is ambiguous or duplicated, it does not guess. It flags the line with a reason, and a person resolves it and approves the output.

On a 253-line document the agent took 58 seconds, matched 250 lines and flagged 3. Total handling time, human review included, was six minutes. For 750 documents a month, around £5,850 of human matching time becomes around £90 of agent compute.

Zoltan's rule: "Automation should stop where confidence ends." He measures the exception rate as well as the time saved, because it shows whether the reference data is getting better.

The client is a UK law firm. Its target market sits on a government licensing portal that publishes a list of over 140,000 companies every day. Spotting new registrations, suspensions and revocations in that list by hand was not realistic.

XITE Create built a pipeline that downloads the list every day, compares it with the previous day's file, and buckets each company as new, suspended or revoked. AI adds the contacts, ranks each company, and drafts an email sequence and a call script. A salesperson reviews and sends.

Contacting companies on the day their status changed raised the response rate. The firm signed new clients it would never have found by chance, adding 5-10% revenue growth.

A real estate broker with 20+ realtors asked for five AI agents. Think RevOps shipped two: build the foundations, prove the model works, and build the other three six months later.

A researcher agent takes an investor's name, checks net worth signals, countries of presence and property held elsewhere, and writes an enriched profile back to the CRM. A second agent drafts the outreach. Every draft goes to a person for sign-off through WhatsApp rather than a dashboard, so approvals are faster.

Each realtor gave 15 hours a month to pipeline work that needed 39. With the agents, realtor time drops to about 4.5 hours, mostly review of outputs. Pipeline coverage rises from roughly 38% to near complete, and first contact goes from days to minutes. Tsvetelin's lesson: lead with the coverage gap, not the cost savings.

AI Central grew to 300,000+ readers across beehiiv, LinkedIn and Substack, plus several domains and apps. Every platform held its own version of the reader, and the CRM said "email sent" and not much else.

We unified the data before we touched any AI. Every reader record is matched to behavior across sites, apps and inboxes, then put into three cohorts that recalculate on their own: superfans (lifetime open rate above 70%), engaged readers (click rate above 2%) and ghosts (no opens in 3 months). AI drafts the sequences and picks the offer for each cohort, and a person approves before anything ships.

The agent is only as good as the data underneath it. CRM entropy is the number one bottleneck we meet with clients.

According to the report, what separates the top 10% of adopters from the other 90% is organizational, not technical.

The low-risk way in is a sandbox: keep the safe workflow running while people experiment next to it.

Whichever path a company takes, Ganapathy pointed out, the decision is less a technology choice than a question of how data flows and integrates across all of those products.

All four systems make the same design choice. The agent does the repetitive part, and a person takes over where the system is no longer confident: a flagged line, a draft to sign off, an offer to approve. It is the same pattern we saw at The AI Summit London 2026.

Before you build, do what the contributors did. Time each step, find the one that takes the most time, audit the data, and decide where the agent flags and stops. Ready-made systems for that work are in the AI Central Library.

You can read the full report in the flipbook above, or download it from the Global Tech Advocates reports page.

A free quarterly report by AI Central and Global Tech Advocates on AI systems in production at small and mid-sized companies. Issue 01, published in October 2026, covers four use cases from four practitioners in the GTA network.

Zoltan Vass, Ganapathy Sankarabaaham, Tsvetelin Anastasov and Alex Fiore, with a foreword by Russ Shaw CBE. Marc Duke moderated the session. AI Central Media in London produced it in partnership with Global Tech Advocates.

The report documents four: matching supplier documents to a product list, turning a public licensing list into same-day leads, researching investors and drafting outreach for realtors, and unifying reader data into cohorts. All four keep a person approving the output.

The report lists four paths. The lowest risk is the agentic features of software you already run, and building in-house is the highest. Executives should own the decision and review each SaaS license against what an internal build now costs.

In the flipbook on this page, or as a PDF on the Global Tech Advocates reports page.