Site plan

Poppins' item pages were hardly found on Google. People searching to rent a carpet cleaner or a hedge trimmer didn't land there: at most one organic click in months per item.

• We set up AI agents that analyse keyword volumes, gaps and difficulty, drawing on Semrush, Google Search Console and PostHog.
• Copy agents redesigned and rewrote the item pages, in French, English and Dutch.
• YouTube and Instagram content optimised to send traffic to the SEO pages: 48 YouTube videos with SEO descriptions, 26 of them linking to an item page.

10 item pages optimised in the first round. In one month, clicks from Google on the item pages went from about 430 to 980. The giraffe sander went from 10 clicks to 91, the hedge trimmer from 16 to 98.

Ask AI for a website and you get the same page as everyone else: a small label above the title, one word in italics, the typical layout.

We taught our AI agents what taste is. Whenever we come across a nicely designed website, an agent team analyses it (screenshots, code, colours, typography, design and animations) and saves it in our style database with references and prompts.

When we design a new site, landing page or webshop, AI draws inspiration from the saved styles. The result is creative, high-end design instead of the standard AI slop.

Building a product configurator meant a complete development team and a big budget.

With AI we built a configurator where visitors answer a few interactive questions and land on the product that fits, with a live price. They leave their email to get the quote.

First version built in one day and live five days later. It now works as the landing page for paid search.

Almost no traffic came from AI assistants like ChatGPT, Gemini or Claude. The hosting even refused their crawlers: 85% of AI-bot requests were blocked.

• A team of 9 AI agents audited how AI assistants find and cite Poppins (about 270 web lookups) and turned it into a GEO checklist and a 9-step roadmap.
• Structure: every AI crawler unblocked and re-tested, an llms.txt that tells AI what Poppins is, structured data on the category pages and 11 new category pages.
• Content: a copywriting agent that writes for AI answers, with a short answer under every heading, a rent-or-buy table, FAQs from real search demand and hard facts about Poppins.
• Measurement: visits from ChatGPT and Gemini tracked separately in PostHog.

27.6% of visitors from AI assistants click through to the app, against 14.4% from organic search. A month later, visits from AI assistants had grown 3.3 times.

Renders for prospects took the marketing department hours of manual work in fixed Photoshop templates.

We built it in Figma Weave with custom skills, so every sales rep makes renders in the setting and branding of their prospect themselves.

Sales uses far more renders in the sales process, because they make them themselves. That means higher closing rates and more sales.

At a trade show, leads end up as business cards in a pocket. After the show somebody has to type them all in and decide who follows up. Or worse, you hand out your business card and nothing happens.

We built a lead platform with custom forms, made for speed at the stand. New leads get an automatic email with the right follow-up material and land with the right sales rep.

Every lead is already in the CRM under the right rep, so follow-up can start the day after the show.

Sales gets a long list of prospects and no quick way to see which ones to chase first.

We built a multi-factor scoring model per client, based on company data and their ideal customer profile. An agent does the scoring automatically.

Sales has an immediate way to classify leads and spends its time chasing them accordingly.

A good opening means really taking the time to prospect your prospect. Your sales team doesn't have that time for a whole list.

AI reads each person's latest LinkedIn posts and reactions and puts them next to the company context and their role. That's where it finds the common ground to open with.

1 in 10 cold prospects ended in an introduction call with the sales rep.

BiometrIQ needed testimonials from its partners, and chasing them one by one takes a lot of time.

After the partner event, each partner types their name to get their professional headshot. Right after that, they can leave a testimonial on video, by voice or in text.

We delivered 142 headshots to 32 partners, and 20 partners left a testimonial: 3 on video, 1 by voice and the rest in text.

The ads sent people to a slow webshop. In six months that brought 41 leads at EUR 29.48 each.

We made 20 ads in the brand's style with an image model, 3 videos and 17 statics, each with its own hook. The first 17 took two evenings. Clicks went to a short form instead of the webshop.

46 leads in about five weeks at EUR 13.77 each, 53% less per lead than the previous campaign. The best static got clicks at EUR 0.37 and a 1.87% click-through rate.

With 3 products for more than 8 B2B audiences, a new visitor needs time to find what fits. Most stay anonymous and do nothing.

Sam is built on Sundo's data and sharpened on 52 real visitor conversations. It greets people with 'No time to browse?' and connects them to the right person in the team right away.

It's not an annoying customer support chatbot. Sam brings in two to three leads a week, including some with large deal sizes.

People post about what they need and what frustrates them in your market. Your sales team misses most of it.

AI scans social media and press for keywords on the industry, Sundo's products and its competitors. After automated quality checks, the hot leads land in one database for sales.

Sales comes in with a very relevant opening, because they react to a need someone just expressed.

Sales needed new practices to call. Finding them by hand, and checking who is already a customer, is slow work.

We pulled practices from Google Maps and checked every one against the CRM. AI then visited each website to verify the practice's registration.

Two rounds turned 2,782 prospects into 244 verified practices. The CRM check also caught 3 existing customers and 4 open deals.

The sales team had 335 cold email leads and only email to reach them.

We matched each lead to their Instagram profile and an AI agent visited every one. It sent a personal DM to public profiles, followed private ones and skipped the rest.

104 personal DMs went out in less than a week, and nobody sent them by hand. The first 30 led to 8 conversations, 2 discovery calls and 1 new client.

Making a deck or report look like your brand eats hours of layout work, and there's always a page where the text runs over.

AI builds the deck, report or whitepaper in the brand's style and checks every page for text that runs over, weak contrast and pages that are too full. Then it exports the PDF.

Since August it made 21 decks, reports and proposals, from a sales playbook to a trade show lookbook. The first real run, an offsite deck for a client, came out with zero layout errors.

You watch a great talk, nod along, and a week later you've forgotten what you wanted to try.

Every talk or video worth keeping becomes a note, and every useful idea in it becomes a proposal on an action board.

Ideas stop disappearing. Twelve are done and 92 are waiting their turn on the board.

Good campaigns get saved as screenshots and bookmarks, and you never find them again when you need an idea.

Each campaign is filed under the mechanism that made it work. One search box covers it, together with your style library and your resource database.

When a client needs an idea, we look up the mechanism and get every campaign that used it.

Every new AI chat starts from zero, so you explain your clients and your projects all over again.

You keep everything in one notes database: clients, projects, resources and what you learned from performance reviews and feedback. The AI reads a one-page index first and only opens the notes it needs.

Every session starts with the AI already knowing your clients and where each project stands.

Many users showed interest in the app and never rented anything. The question was where they dropped off, and why.

We followed 1,560 borrowers step by step, from opening the app to leaving a review. Then we compared them with users who showed intent but never rented.

Finishing the onboarding missions separates the two groups most. The biggest drop is between chatting and sending a booking request.

AI video tools change every month, and a prompt written from memory gives you something that looks fake.

Three skills cover the video idea, the photo prompts and the video prompts. They work from 101 prompt rules, each checked against its source.

We used them in a client's AI video session the day we finished them.

The ad reports show what works, and the team needs new ads every sprint. A shoot for every idea doesn't keep up.

New ads come out of fixed workflows, as animation or from footage the team films on their iPhone at home. AI swaps the object or character and keeps the real room.

Everybody can now film the exact footage we need with their iPhone, from their own living room.

Finding the hook that works meant hiring influencers, later content creators, or spending thousands of euros on test ads first.

An AI character of your own tests different textual and visual hooks, before any of them becomes an ad.

One test hook on Sundo's feed reached 24,754 people and 50,615 views, 97% of them people who didn't follow Sundo yet. No ad budget needed to learn what works.

Every partner in the certification track needed their own announcement, certificate and badge. Designing a version per partner by hand is a lot of repetitive work.

Five templates cover the three-session track. A script fills in each partner's first name, last name and practice.

Each partner will get their own certificate, a badge that works as a label on their website and stories to share. Pending: the track runs until 24 November.

AI-written copy reads like AI, and your readers notice.

Before the first line, the AI fills in a brief with two fields it can only fill by quoting the guidance. Then it checks the text against 49 English and 12 Dutch patterns that give AI away.

Every text we ship goes through it, this one included.

The yearly partner event needs partners in the room, and a mass invite is easy to ignore.

The two founders filmed the invitation once and said 153 first names on camera. A script cut one video per partner, put the right name in it and checked every video.

202 personal videos were ready about two hours after the final edit was approved. 40 partners have signed up so far, with registration still open.

A testimonial video with a client used to take hours and a full production team.

We record your client wherever they are. AI puts them in a studio and keeps their real mouth movement, expressions and voice.

Now you just ask your client to join a 5-minute recorded call.

Posters, category pages and ads all need photos, and every one of them has to follow the brand's image style guide.

We turned the style guide into an image tool the marketing team uses themselves. It knows the four image types and logs what every image costs.

For the first poster campaign we made 106 images for about USD 16 and delivered 25 options. We also made 423 category images for the website.

Every quarter, the bookkeeping took two working days. Most of that time went into finding invoices: in supplier portals, in the mailbox, behind every line on the card statement.

The bank statement became the checklist. For every expense, an AI agent finds the invoice, names it and files it in the right month. Then it books it in the accounting software the same way as the last booking from that supplier, and runs five checks before every save. The owner only answers what an expense was for.

230 bank transactions for the third quarter of 2026, every expense matched to an invoice or a reason, and 142 of the 161 documents booked. It took about four hours instead of two working days.