The Sales Problem That Wasn’t a Sales Problem

Talabat was growing. 23% GMV growth. Record numbers. And somehow, the commercial team was still missing targets quarter after quarter.

The easy answer was already circulating in the building: the sales team wasn't good enough. It's the kind of diagnosis that moves fast in organisations because it's clean, it points at someone specific, and it lets everyone else off the hook.

The COO didn't buy it. Neither did I.

I was brought in to run a company-wide commercial transformation across 8 markets. My first move wasn't to build anything. It was to go find out what was actually happening.

I pulled the data. Then I left my desk and went to where the work actually happened.

I shadowed sales calls. Sat next to reps while they prepared for partner meetings. Watched what they did in the thirty minutes before a call: the spreadsheet tab-switching, the Slack messages asking colleagues if anyone knew this partner's history, the educated guesses standing in for actual intelligence.

That's when it became obvious. These weren't underperforming salespeople. These were capable people operating completely blind. There was no single source of truth for 65,000+ partners across 8 markets. Every rep was manually piecing together context before every conversation, pulling from different databases, different reports, different versions of the same data. The insight that should have taken seconds was taking hours. And it was still wrong half the time.

Two more things fell out of the data. Barely half of a rep's week went on actually selling. The rest disappeared into tracking performance in spreadsheets and chasing operational issues across five different tools. And because the pressure was to hit this month's target, reps leaned hardest on the smallest partners, where a yes came easiest, while the biggest partners, with the most room to grow, were under-served. The setup wasn't just slow. It was pointing effort at the wrong places.

The problem wasn't talent. It was that we'd built a €100M commercial operation on top of infrastructure held together with spreadsheets and institutional memory.

I designed two things that were always meant to be one system.

The foundation was the Account Management Centre, a single intelligence platform consolidating every commercial signal that mattered: partner performance, order volumes, engagement history, revenue trends, market benchmarks. I designed the data schema from the internal databases across all 8 markets. Engineering built the platform to my specification. For the first time, an account manager could open one screen and actually know what was happening with their book of business.

But a dashboard alone wasn't enough. Data doesn't tell you what to do next.

So I built the Recommendation Engine on top of it. This was the part I designed myself, and it starts from the partner, not the product. For every partner it works through four questions in order. Where are they in their lifecycle: new, active, or gone quiet? Are the basics right: menu, operations, pricing? How do they compare with similar partners in the same cuisine, area and price point? Only once the gap is clear does it pick a product to close it. Low traffic gets ads. Traffic that doesn’t convert gets discounts. Too few new customers gets acquisition deals. Customers who don’t come back get targeted deals.

Underneath, it’s a hybrid. Rules for the patterns we already understood. Heuristic scoring for the commercial judgment calls that couldn’t be reduced to a rule. And machine learning finding the patterns in 65,000 partner histories that no human would have spotted.

The output wasn't a report that sat in someone's inbox. It was a live dashboard with real-time alerts pushed directly to account managers. Here is who to call today. Here is what they need. Here is what to offer them and why.

Each recommendation landed as an action item, tagged as a new acquisition, a reactivation or an upsell, so a rep could work one kind of conversation at a time. Reps logged the revenue they expected from each action, so they and their managers could see the gap to target before the month ended, not after.

Then we went one step further. We integrated the system directly into the partner portal, so high-potential partners could self-serve into ad products without ever touching a sales rep. That was the move that turned a sales tool into something closer to commercial infrastructure. The system was now generating revenue while the team slept.

I ran the rollout myself. Piloted in one market, starting with surveys and discovery sessions with the people who would use it, then went live across all 8 within a month, with training for every account manager and team lead. A leaderboard made adoption visible, and early wins turned into internal champions.

After launch it became a constant loop. Reps told us what got in their way, and we shipped fixes: bulk-closing actions for key account managers, quick filters by priority and partner type, a pipeline view for team leads. Change management isn’t glamorous but it’s where most good systems go to die. This one didn’t.

Piloted in March 2024, live in every market by April. Within 6 months:

Adoption held steady at 80%. This wasn’t a tool people tried once and abandoned; it became how account managers ran their day.

The recommendations worked. 56% of recommended actions succeeded, and partners who received an action item grew at twice the rate of those who didn’t. Partner wallet share rose 2 percentage points.

Commercially: €1.2M a month in incremental revenue impact. Sales admin time cut by 40%, hours of manual research collapsed into seconds. The recommendation engine alone drove a 7% revenue uplift.

And Talabat posted its record 23% GMV growth that year. This system was part of how.

The presenting problem is almost never the real problem. The sales team wasn't the issue. The infrastructure was. Getting to that answer required ignoring the obvious diagnosis long enough to go find the actual one.

The best systems don't just help people do their jobs better. They eventually do parts of the job themselves.