Finding new advertisers with AI-assisted research

Discovering opportunities after competitors

The agency's head of sales approached me for help finding new advertisers. There was plenty of market information, but the team could not always identify promising partners and contact them in time. Often, by the time it discovered a suitable company, competing agencies had already signed contracts with it.

There was no central source where companies announced their readiness to partner. Existing directories did not solve the problem either: they listed familiar advertisers with whom competitors had already signed contracts. The agency needed to find new partners proactively rather than wait for responses to its own advertising.

The head of sales wanted to discover advertisers with promising partnership terms earlier and focus the team's efforts on them. A recently launched affiliate program, for example, could be a useful signal. The team had already tried direct, repeated searches through ChatGPT and other language models, but these had not worked well for the task.

Building a scheduled lead-discovery system

I interviewed the head of sales in detail about the existing research process and the types of advertisers the agency needed. From those discussions, I defined the requirements, built a prototype, and then independently developed and launched the complete system.

The solution is a sequence of data collection and processing steps, not a single prompt to a language model. It gathers specialist industry analytics through source APIs and interfaces. Custom algorithms process and structure the data, after which the system identifies the legal entities behind the businesses it has found.

Web search and language models then help find contact details, partnership terms, and affiliate program information. If a company meets the selection criteria, the system adds the lead to a Google Sheets spreadsheet, together with contacts and context for outreach.

The process runs at a set frequency. It automatically prepares leads for the selected advertiser category, but it does not replace all of the sales team's prospecting work.

Using language models within a broader workflow

Direct requests to ChatGPT and other language models had not delivered the results the sales team needed. I therefore integrated language models into a system that first collects fresh data and prepares it for further processing. Unlike an employee asking a chatbot questions, this connects research, processing, and delivery into a recurring workflow.

Language models help interpret information about companies, contacts, and partnership terms. Data collection, structuring, and other steps are handled by custom code. Language models alone could not solve the problem: the value comes from combining methods, not simply adding AI.

I do not disclose the analytics sources, exact metrics, or internal algorithms here. I explain how to divide work between language models and conventional code in my article on choosing between AI and rules.

Giving sales a regular source of relevant leads

The agency's internal system has been running for almost a year, regularly finding potential advertisers with promising partnership terms. For this category of company, sales receives more than a list of names: each lead comes with contacts and supporting context.

Information gathering and preliminary selection run on a schedule. Salespeople no longer need to repeat the entire research process for this category each time. They can start by reviewing the prepared information and contacting a potential partner. Meeting the selection criteria does not replace negotiations or guarantee a deal.

Where this approach can help

This approach is useful in markets where information about companies ready to partner is scattered, and spotting an opportunity and making contact earlier creates an advantage. Instead of relying on inbound inquiries, a business can organize regular research across available sources.

Adapting the approach to another market starts with understanding the work, not choosing a model. I first discuss the process with the people currently finding prospects: which companies they need, where useful information is available, and which signals help assess partnership terms. I then determine which tasks belong in conventional algorithms and where AI can help. Read more about this approach on the AI integration and automation page.