How Novakid turns Instagram into a lead and sales engine across 15 markets

How-to from Maria Gladysheva, Head of SMM at Novakid · Interview by Dmitrii Ive, CEO of Dashly

1.86M followers across 14 accounts

15 regions covered by SMM

6 managers running all of it

17,370 posts published to date

Novakid is an online school that teaches English to kids. It runs in dozens of markets: Turkey, the Arab region, Poland, Germany, Czechia, Spain, Italy, France, Latin America, Israel, Korea, Japan. Instagram for Novakid isn’t a brand channel. It’s a working lead-generation machine with attribution all the way through to a sale.

We talked to Maria Gladysheva, Head of SMM at Novakid. She’s been with the company for almost seven years, long enough to remember when one early-market account was the single biggest growth driver for the whole business.

The conversation covered two things:

1. How content production for 15 regions actually runs with a team of six.
2. How a follower turns into a customer who shows up for a trial lesson.

Part 1 · Growth

Six managers, 15 regions

Dmitrii: Walk me through how the team is set up.

Maria: Our model isn’t classic, because we’re running across different regions at once. When you’re focused on a single market, things are fairly straightforward. We have a lot of regions, so we’re constantly optimizing tasks and moving them between people.

I can’t put one manager on every region. That’s an enormous amount to manage! I have six social media managers, and each one owns 2-3 regions. Each of them is basically a micro Head of SMM for their own patch. They own everything that happens there.

Dmitrii: And who executes the work?

Maria: Each manager runs a team of freelancers. The only Novakid employees are the managers themselves, everyone who helps them get the work done isn’t on Novakid’s payroll.

There’s also a content producer, the person who organizes every shoot with real people, real kids. She’s excellent at it. Her mom is a producer too, and together they can find almost anyone.

Dmitrii: What about in-house production?

Maria: For SMM we need a lot of content. Fast. And in-house production runs on a much longer planning cycle. It’s faster and cheaper for us to hand tasks to freelancers and get our own content back. We can’t wait a few months for a finished video.

The team leads brief freelancers with specifics: what to shoot, where, and how.

The top, middle, and bottom of the funnel

Dmitrii: You mentioned you’re rebuilding the funnel. What’s changing?

Maria: We used to focus mostly on middle and bottom, because SMM was expected to deliver sales directly. Top of funnel barely existed, we never had the bandwidth for it.

Now content splits into three tiers, and each one has its own goal. That’s the core change. If you try to get everything out of one piece of content, you get nothing.

Tier	Job
Top	Novakid’s name never comes up. Aim wide, at parents generally interested in English. The job is to land in the recommendations feed and become recognizable through repetition.
Middle	Someone already follows. Show the product and build trust through content. Get them deeper into how lessons actually work.
Bottom	The target action. Book a trial, grab a lead magnet. Anything that generates leads directly.

Maria: Someone watches once and likes a video. A couple of days later: “Oh, I’ve seen these guys before, this is actually interesting, let me watch more”. That’s the whole mechanic.

The main idea is balance. So we have content to cover all three tiers equally. The logic is simple: someone lands on your feed and needs to see both the recurring content they came for and some kind of offer.

Each tier gets its own metric

One metric drives audience growth at every stage of the funnel: watch-through, if people actually watched a video till the end. Below that, the metrics split by tier:

Tier	What we track
Top	Reach and % of non-followers
Middle	Engagement: saves, comments, likes
Bottom	Requests: comments with the keyword and DM inquiries

Reach and conversion don’t have to agree. If bottom-of-funnel content did its job and generated requests, it might only reach 3,000 people. And that’s fine. It hit the right audience and converted. Marketing’s actual KPI isn’t requests, it’s trial success: someone booked a trial lesson, showed up, didn’t reschedule, and completed it.

Audience size by market

These are 14 accounts we reviewed manually. Data from public profiles as of August 2026.

Account	Posts	Followers
novakidturkiye (verified)	1,812	724K
novakid_mena	1,098	388K
novakidschool	2,075	287K
novakidespana	1,631	134K
novakid_latam	615	57.6K
novakiditalia	1,563	55.1K
novakidpolska	1,441	52.5K
novakid_france	1,178	51.2K
novakid_romania	1,327	35.3K
novakid_israel	911	22.9K
novakid_de	1,236	20.7K
novakid_korea	750	16.4K
novakid_jp	718	9,822
novakid_czech	1,015	7,445
Total	17,370	1.86M

Almost 1.9 million followers and 17,000 posts. That’s six managers and their freelancers!

Note: this isn’t the complete picture. Maria talks about 15 regions in active SMM work; we could only find 14 public accounts. Novakid also runs accounts for at least Indonesia and India, plus a separate English-language account. Total audience is meaningfully higher than 1.86 million, so treat the numbers above as a minimum, not the full total.

Here’s where it gets interesting.

Two markets hold 60% of the entire audience. Turkey and the Arab region together account for 1.11 million of the 1.86 million total. These are exactly the markets Maria describes as the most responsive.

• All of Europe combined is smaller than one Arabic-language account. Spain, Italy, Poland, France, Romania, Germany, and Czechia together add up to 356,000. novakid_mena alone has 388,000. Seven markets against one.
• Production volume explains almost nothing. Spain has 1,631 posts and 134,000 followers. Turkey has 1,812 posts and 724,000. Output is roughly the same, audience differs fivefold.
• One breakout video outweighs months of work. The Turkish account’s current feed runs 23,000, 87,000, and 147,000 views per post. Its pinned video pulled 16.1 million.

Markets behave differently

Dmitrii: Do you run the same content approach in every market?

Maria: The structure is the same everywhere. The main difference from market to market is what actually resonates and how well it performs.

Market	What we see
Turkey and the Arab countries	“Whatever you post, they love it, they hold nothing back, they’ll give you everything.” Hearts, likes, and saves come fast, and accounts grow quickly. They respond strongly to anything national, and that noticeably lifts engagement. French audiences react in a similar way.
Germany and Czechia	“We used to joke that they think liking a post means they’ll owe you a couple of euros.” Poor engagement. But they might never like a single post, then book and buy anyway. Conversion is strong. The market is expensive, and the sales cycle is long.
Spain and Latin America	Spanish audiences will like everything and rarely buy. In Latin America the account took off fast, but the price didn’t work: a month of lessons costs a noticeable share of a family’s income.
TikTok	Customers only come from the Arab market. Every other region grows on TikTok, followers are on-target, but they don’t book trials. The platform might just be used for entertainment.

Localization shows up directly in the content. The Korean account posts videos about Korean fried chicken, the German account runs AI banners of a German bakery, the Polish account keeps WhatsApp in its bio. There’s no one-size-fits-all Reel for 15 markets. Every market gets its own local spin.

What actually works: the Italian account didn’t pin three different videos, it pinned one series: two kids side by side, one learning regular English, the other after Novakid. One format, repeated many times, on an account with 55,000 followers. Exactly what Maria means by building recognition through repetition.

Where AI works in content, and where it doesn’t

Dmitrii: Are you making Reels with AI characters?

Maria: We tried, it didn’t really land. Our parents are mostly millennials, and they pick up on it instantly.

If we use AI in video as animation, something clearly drawn, people are fine with it. It’s obviously not real, and that doesn’t matter. But show a person who’s supposed to be talking, and there’s no reaction at all. People just stop watching. So we stuck with live-action shoots.

Meanwhile there’s a huge volume of AI content elsewhere. All the banners, all the static images. Fast and cheap.

Europe is rolling out AI content-labeling rules. Novakid already knew about the chatbot side, disclosing that a conversation partner isn’t human. Now Legal sent over a document covering content too.

Trend-watching and competitors

Competitor tracking used to be manual: someone would sit down with a list of direct competitors per market and see what was landing. On the French market they’re now testing an agent instead. It pulls data on direct competitors, adjacent players, and the category as a whole, plus profiles the target audience.

One shoot produces 4 to 5 videos and costs a few hundred dollars. “I don’t want to get this wrong, but it’s in the low hundreds.” $100 to $300 depending on talent and location. Shooting a single Reel doesn’t pencil out, nobody does that.

Part 2 · Lead gen

How a follower becomes a customer

Content built the audience. Next it has to turn into trial bookings. Novakid runs three entry points here, with an agent layer on top: one agent opens the conversation, a second walks the person through a script to booking, and a manager only steps in where a human is genuinely needed.

Entry point	What happens
01 · Follow	The moment someone follows, a first-touch agent fires and sends a welcome DM. You can only message a user first on Instagram after they’ve taken an action themselves. Miss that window and the lead just stays a follower.
02 · Keyword comment	A post offers something specific: comment “5” and get a set of games. An agent catches the keyword and sends the lead magnet by DM.
03 · Direct message	Someone messages first. On some markets there’s a WhatsApp button right next to it, and the conversation starts in a different app entirely, bypassing DMs.

What this looks like in practice

The German account’s pinned post shows an AI illustration of a German bakery: count the bakers in the picture, comment the right number, get a gift. One post does three jobs at once. Engagement (people count the bakers), a keyword (the number becomes the trigger), and a lead magnet (the gift lands in DMs).

What happens after the first message

From there, a scripted agent picks up the conversation. It walks the person step by step to the target action: answers questions, sends the right material, collects data, and books the trial lesson. A manager only joins where a human is genuinely needed.

Maria: Our bots are trained specifically on how we work. They can sign a customer up and save managers a huge amount of time.

Sometimes a lead gets stuck partway through. The bot keeps messaging, and there’s silence on the other end. A human joins, and the lead comes back to life. Maybe people can sense it’s not a bot anymore. The conversation continues, and conversion happens. It matters to me that we see funnels through to completion.

Warm-up instead of a hard link

Maria: Right now it’s a single message, but we want to test a sequence, especially on European markets. Someone follows, gets a message immediately, and sits there confused: “I just followed, and now I’m getting messages, who even is this?” And we go straight for: come book a trial lesson at Novakid, leave your phone number.

Our sales cycle is long; it runs several months. So the warm-up needs to run a lot longer too, instead of pushing everyone to book immediately.

Attribution: a tag on every content funnel

Maria: Every funnel has its own tag, and there’s a system for how we set them in links. A person gets the link tied to the funnel they came through. If they came from a lead magnet, they get the link tied to that lead magnet.

From the keyword, we can see that a specific person came from that exact post, converted, and made it to a sale. I don’t care much about DM response efficiency; that’s not what interests me. What interests me is which funnels convert well and what audience each one brings in.

A separate warm-up layer. The Italian, Spanish, Czech, Romanian, Polish, French, and Korean accounts all keep a broadcast channel in their bio, something like “VIP channel for parents.” It sits between the follow and the request: a follower who isn’t ready to book yet can be nurtured without the 24-hour limit and without depending on the recommendations feed.

What to take from this

1. One piece of content solves one job. The equal content creation model isn’t about a pretty feed. Reach, engagement, and requests are three different metrics, and content trying to win all three usually wins none.
2. The entry point decides everything downstream. Following, commenting, and messaging are three different states a person is in. The same welcome message for all three burns your hottest entry point.
3. Measure to trial success, not to requests. You can farm any number of requests. The metric that actually moves revenue sits two steps further down the funnel.
4. A market outweighs production volume. The same number of posts produces a fivefold difference in audience between Turkey and Spain. Before you double output, check whether it’s even the right market.

Then the conversation has to become a meeting

Novakid is strong exactly where inbound starts: content across 15 markets, three entry points into a conversation, a tag on every content funnel. They know how to turn a Reel into a conversation with a parent, and that’s a solid reference for any inbound channel.

But a conversation isn’t revenue yet. Next, that conversation needs to be qualified, pushed to a booking, and followed through so the person actually shows up. Novakid calls that metric trial success; in B2B it’s a meeting that actually happens. That’s exactly the stretch where revenue agents work, and it’s what we build at Dashly.

AI SDR Agent is the core orchestrator. Before it says a word, it pulls the lead’s profile, site behavior, and CRM history, so it never opens with “how can I help”. It qualifies through natural conversation instead of a form, decides whether the lead is a fit, and when it is, books the meeting directly into a live calendar and keeps a reminder cascade running until the person actually shows up. The same agent that opens a stalled thread also closes it, so there’s no handoff for the lead to notice. It doesn’t close the deal on that call. A person does.

AI Support Agent is a sub-agent that answers product questions straight from your knowledge base. AI SDR calls it mid-conversation when a question comes up, then takes the conversation back.

Both run in one system on one unified customer profile: Instagram Direct, WhatsApp, Facebook, website chat, email. One conversation, one voice, even when the sub-agent answers a question behind the scenes. On a channel like Instagram, that matters even more: switching who’s “talking” to a lead mid-thread is exactly the point where you lose them, and Maria’s own story about a bot going quiet until a human steps in makes that case better than we could.

70%	~1 sec	~40%	568%
of MQLs book a meeting without a manager touching the thread	first response, down from ~1 hour	of support questions closed with zero human involvement	ROMI on Dashly spend

Find out how agentic AI team can help you convert your inbound leads fast and effectively

Book a demo