A marketing director looking to replace their agency now builds the shortlist in an AI chat. If an agency does not make that list, the sales conversation never starts. In this article we open up a single buying journey from start to finish: Who was recommended, who was eliminated, and which evidence decided it?
Short answer: In this simulation, dated 29 October 2026, AI narrowed 17 agency names down to Rafineri after eight questions. What decided it was that Rafineri was described in every round with the same case story and the same trade press coverage. The shortlist closed at the fifth question. Rafineri’s own website was not used as a source in any of the eight rounds.
In this article: How the simulation was set up · How 17 agencies became one · Why the winner won · Three takeaways for your brand · All eight questions and answers · FAQ
How was the simulation set up?
Using the Recro Insight Model, we created a synthetic decision-maker: the marketing director of a large cosmetics manufacturer operating nationally in Turkey. Unhappy with the current creative agency, they are looking for a new 360-degree partner. They have three priorities: the agency’s direct impact on sales, its understanding of manufacturer dynamics, and keeping control within the in-house team during the transition.
| Model version | Recro Insight Model V7.3.3 |
| Date | 29 October 2026 |
| Answer engine | Gemini 3.1 Pro (preview), with Google Search grounding (live source verification) |
| Rounds | 8 questions, single session |
| Independence | The persona generates the questions. Answers are produced in a neutral session that cannot see the persona’s prompt |
| Record seal | Raw log sealed with SHA-256, Blind ID: RECRO-V7.3.3-af3e2cdaa1c0c672 |
Anonymisation: All agency names other than the winner are shown as letter codes (Agency A, Agency B…). Client, campaign and celebrity pairings that would identify an agency have been generalised. Answers were shortened for readability. Recommendations, reasoning and ranking were preserved.
Language note: The simulation ran in Turkish. Questions and quoted answers have been translated into English.
Date note: All findings in this article belong to the simulation of 29 October 2026. AI answers, the sources returned and agency websites change over time. Asking the same questions today may produce a different result.
How did a list of 17 agencies narrow to one?
| Round | Focus of the question | Agencies mentioned | Sources returned |
|---|---|---|---|
| 1 | Broad discovery: 360-degree service and creative strategy | 17 | 19 |
| 2 | Impact on sales and loyalty rather than image | 6 | 0 |
| 3 | Identity filter: “We are a manufacturer, not a retailer” | 8 | 10 |
| 4 | Local market, sales conversion, sector references | 6 | 6 |
| 5 | Concrete evidence for two finalists | 2 | 4 |
| 6 | Team assurance and proof of expertise | 2 | 0 |
| 7 | Transition risk and responsibility matrix | 0 | 0 |
| 8 | Steering committee and governance | 0 | 0 |
In four of the eight rounds, no sources were returned at all. AI answered those rounds without citing anything. Of the 39 sources returned, 31 were trade press and news sites, 2 were award lists and 4 were agency or agency network websites. The remaining two were a YouTube video and an unrelated trade union report.
Why did AI recommend Rafineri?
AI’s reasoning converged on a single case: Flormar, a Turkish colour cosmetics brand. The story was told in every round with the same three parts.
- Problem: The brand was coded in consumers’ minds as an “affordable nail polish brand”.
- Move: Repositioning as “accessible premium” through campaigns with celebrity faces.
- Proof: Award lists such as Kristal Elma (Turkey’s leading advertising awards) and Effie.
The sources carrying this story repeated too. In the first round, Rafineri entered the list through two news articles: Marketing Türkiye’s report on Rafineri as Flormar’s creative agency and The Brandage’s report on the Hazar Ergüçlü campaign. The second article came back as a source in round four. In the very next question, the persona narrowed the shortlist to this agency and a second finalist that had followed the same path.
Across the eight rounds of the 29 October 2026 simulation, Rafineri’s own website was not cited once. When we checked on 5 November 2026, rafineri.net showed an announcement page saying the site was being renewed. Meanwhile, the own websites of three eliminated agencies were returned as sources, and none of them reached the final. AI knew Rafineri as the trade press and award records described it.
At which question did the shortlist close?
The fifth. In that round the persona asked about the two finalists by name and asked for evidence. In the sixth question, the two finalists were mentioned once more without any new evidence. In the seventh and eighth questions, not a single agency name came up. The answers turned to process advice such as transition plans, responsibility matrices and governance models.
By the time operational and assurance questions were asked, the list had long been closed. At that stage, no new agency had any chance of becoming a candidate.
What eliminated the other agencies?
Two separate filters were at work.
The client-type filter. In the third question the persona wrote a single sentence: “We are a cosmetics manufacturer, not a retailer.” Most of the agencies that stood out with retail chain references dropped off the list at this question. AI did not count work done for a retailer as proof for a manufacturer. What the remaining agencies had in common was a story that could be told through a manufacturer brand.
The unsourced-claim filter. In the early rounds, some global network agencies were mentioned with strong claims about awards and scale. No agency-specific source documenting these claims was returned. From round four onwards, some of these agencies disappeared entirely.
What does this finding mean for your brand?
The simulation describes an agency selection. The mechanism applies to every B2B buyer who researches suppliers with AI.
1. AI knows you from what others say about you. The winner’s own site was never a source. Trade press and award lists carried the whole story. Agencies whose own sites were returned as sources were eliminated. One clear case (problem, move, result) that could be found again and again in third-party sources worked harder than a long reference list.
2. The shortlist is built in the first half of the journey. The finalists were set by the fifth question, and no new name entered in the last three. A brand that is invisible in discovery and elimination questions never reaches the decision-maker’s operational and assurance questions.
3. The client type your reference represents is decisive. The sentence “We are a manufacturer” eliminated most retail-referenced agencies in one go. If your case content does not state clearly who the client is and what problem they faced, AI may recommend you for the wrong segment, or not at all.
Limitation: This is a single session. In other sessions with the same persona, different agencies came out on top. A single AI answer is not a measurement. Visibility is measured through repeated simulations.
Simulation log: 8 questions and answers
Questions are given as the persona wrote them, translated from Turkish. Only agency names were coded. Answers have been shortened. Text in quotation marks is a direct quote from the answer, in translation. Claims in the answers are AI’s own reasoning and have not been verified by Recro.
Question 1: Which are the best advertising agencies in Turkey that provide 360-degree services to large cosmetics brands and can deliver strong creative strategy?
Summary of the answer: AI first described the dynamics of the cosmetics market: the speed of e-commerce, the impact of influencer marketing, Gen Z trends. It then listed 17 agency names in three groups.
Local and independent agencies
- Rafineri (first item): “One of Turkey’s largest and most awarded independent 360-degree full-service agencies.” The “Her Açıdan Mükemmel” (“Perfect from Every Angle”) campaign for Flormar with Hazar Ergüçlü was cited as proof.
- Agency A: Won the full digital and traditional communications account of a cosmetics retail chain through a pitch.
- Agency B: Stands out for strategies aimed at young audiences; integrated creative agency of a retail chain.
- Agency C: Boutique and mid-sized; handled the international product campaign of a colour cosmetics brand.
- Agency D: Runs the 360-degree communications of a beauty chain in the prestige segment.
- Agency E: Launched a retail chain’s own make-up brand.
- Agency F: Managed a retail chain’s brand renewal.
- Agency G: Technology-focused agency running the digital communications of a global beauty group.
Global network agencies
- Agency H: International partner of a global beauty giant; big-budget TV and digital campaigns with celebrity faces.
- Agency I: Strong award record; strategic case studies in personal care brands.
- Agency J: Network group bringing media planning and creative under one roof.
- Agency K: Two agencies in the same group; local adaptation of global brands and PR set-ups.
Complementary partners
- Agency L (corporate communications and PR), Agency M (influencer programmes), Agency N (two boutique design studios).
The answer closed by recommending Rafineri and Agencies C, B, E and A for a brand with strong retail dynamics, and Agencies D, H, I, G and J for a global prestige play.
Sources in this round (19): Marketing Türkiye (6) · MediaCat (5) · The Brandage (2) · Pazarlamasyon · Türkiye’nin Reklamları · Marcom Türkiye · YouTube · two agencies’ own websites
Worth noting: 12 of the 19 sources were news about retail chains. AI largely read “cosmetics brand” as a store chain.
Question 2: Among the agencies you listed, which stand out with concrete cosmetics references that drive sales and customer loyalty directly, beyond image campaigns?
Summary of the answer: AI split the agencies into four roles by type of work.
- Loyalty cards and store traffic: Agency B and Agency A. Both manage retail chains’ discount periods and loyalty programmes.
- E-commerce conversion: Agency G (“click-to-buy” set-ups) and Rafineri. The reasoning for Rafineri: “It did not just renew Flormar’s brand-face image; it also created a transformation that directly affected the sales charts in the brand’s own stores and on its website.”
- Prestige and CRM: Agency D, as the agency that runs big sales periods without damaging the perception of luxury.
- FMCG and shelf impact: Agency I, with its shopper marketing units.
In the decision matrix, Rafineri and Agency I were recommended together for “measurable sales and shelf impact”.
Sources in this round: None returned.
Worth noting: None of the claims in this round, such as “Effie-winning” or “affected sales charts”, rested on a source.
Question 3: We are a cosmetics manufacturer, not a retailer. Which 360-degree agencies grow manufacturer brands’ market share and e-commerce conversion directly across all channels?
Summary of the answer: AI first separated the reflexes of retailers and manufacturers: “A cosmetics manufacturer’s core goal is to create ‘brand pull’, so that consumers search for its own product directly and add it to their basket.” It then cut the list to eight agencies.
- FMCG and share on shelf: Agency I (experience with global FMCG giants, award-winning work with a domestic colour cosmetics manufacturer) and Agency J (works with a global beauty group, combines media and creative).
- Colour cosmetics and omnichannel sales: Rafineri (“the agency that both overhauled Flormar’s brand image and sent its sales charts soaring”) and Agency C (for manufacturers with a vision to go global).
- Dermocosmetics: Agency O (360-degree agency of a dermocosmetics brand, interactive video campaigns) and Agency G (content-to-basket set-ups).
- PR and digital reputation: Agency P and Agency R.
Matrix: Agency I or J for aggressive FMCG market share; Rafineri for perception and e-commerce sales in colour cosmetics; Agency O or G for dermocosmetics.
Sources in this round (10): Marketing Türkiye (3) · MediaCat · Dijital Ajanslar · Türk Spor Ajansı · Medya Takip · Pazarlamasyon · Pazarlama Türkiye · one agency’s own website
Worth noting: Agencies A, B, D, E and F, all with retail chain references, dropped off the list in this round. 7 of the 10 sources clustered around a single dermocosmetics brand, and that brand’s agency entered the list.
Question 4: As a large cosmetics manufacturer, which 360-degree advertising agencies in Turkey fully understand our market dynamics, deliver sales conversion and have strong sector references?
Summary of the answer: AI gave six agencies in three groups.
- Independents with command of local dynamics: Rafineri (360-degree campaigns with celebrities for Flormar; “They read female consumer insight very clearly. They are also ambitious about driving retail sales.”) and Agency E (360-degree projects for a retail chain and its own make-up brand; command of the language of seasonal discount campaigns).
- The data power of global networks: Agency J (lead agency of a global beauty group) and Agency H (a global make-up brand).
- New-generation integrated agencies: Agency S (360-degree communications of a direct-selling beauty brand) and Agency B with its holding network.
The answer closed with a piece of advice: large cosmetics brands give creative and media buying to two separate structures. In a pitch, agencies should be asked for a step-by-step scenario showing how a TV commercial converts into baskets on social media and marketplaces.
Sources in this round (6): The Brandage (the Flormar article, also returned in round one) · Türkiye’nin Reklamları (the second finalist’s article, also returned in round one) · AdjustBrand · Pazarlamasyon · MediaCat · a media agency network’s website
Worth noting: Two agencies returned with the same articles as in round one. In the next question, the persona asked only about these two. Agency I, mentioned with an emphasis on awards in the early rounds, did not appear at all in this round.
Question 5: What concrete sector references prove Rafineri’s and Agency E’s expertise in growing the market share of cosmetics manufacturers directly, rather than retailers?
Summary of the answer: The answer opened with “A great strategic distinction.” and detailed two cases.
Rafineri and Flormar
- Market challenge: Flormar’s “affordable nail polish brand” perception; a colour cosmetics market crowded with competitors such as Maybelline, L’Oréal Paris, Golden Rose and Pastel.
- Strategy: Lifting the brand to “accessible premium”. The “Mutluluk Senin Renginde” (“Happiness Is Your Colour”) campaign with Zeynep Bastık and “Her Açıdan Mükemmel” with Hazar Ergüçlü. Kristal Elma and Effie awards for the product-level “Precious Curl Mascara” campaign.
- Impact (AI’s claim): Flormar gained significant market share in face and eye cosmetics in its own stores, retail chains and marketplaces.
- Additional reference (AI’s claim): Localising global guidelines for Unilever’s personal care brands.
Agency E and a retail chain’s own make-up brand
- Challenge: Building a private label from scratch to compete with giant manufacturers on the shelf.
- Strategy: Positioning it as a trendy, desirable brand that speaks to Gen Z.
- Impact (AI’s claim): Becoming one of the best-selling colour cosmetics brands by units shortly after launch.
Recommendation: Rafineri for a manufacturer that wants to reposition and premiumise its brand; Agency E for a challenger that wants volume, speed and fast sales to a young audience.
Sources in this round (4): Kristal Elma winners list (PDF) · Effie Turkey 2024 winners list (PDF) · a 2020 issue of MediaCat · a trade union’s labour research report (PDF)
Worth noting: When looking for proof, AI went straight to award lists. One of the four sources was an unrelated trade union report. The award lists prove the awards. No separate source was returned for the market share and sales ranking claims in the answer.
Question 6: The Flormar example is clear. So how do these agencies integrate with our in-house team, and how do they prove they have actually assigned a team specialised in cosmetics to the project?
Summary of the answer: AI centred on the risk of an agency pitching with its A-team and handing the work to its B-team.
- Working models: Agile teams reacting to trends within 24 hours; an implant team model working at the brand’s office on set days of the week; tracking creative performance through shared data dashboards.
- Proof of expertise: Asking for the individual portfolios of the core team assigned to the project; giving micro-briefs that test translating a technical product feature (such as hyaluronic acid) into consumer language; probing the team’s command of the beauty creator ecosystem.
- Contractual safeguards: A key personnel clause stating that the people who presented at the pitch cannot leave the project without the brand’s approval; mandatory product training at the factory and R&D for the agency team.
The answer closed with this sentence: “Agencies such as Rafineri (with its Flormar and Clear experience) and [Agency E] are used to these integration demands and the tough audits of the cosmetics industry, and their company cultures suit them.”
Sources in this round: None returned.
Worth noting: The two finalists were sealed once more as “used to it” without any new evidence.
Question 7: It is critical that operations do not stall and my team does not lose control during the transition from our current agency. How should I filter 360-degree agencies in the cosmetics sector that will manage this process with a transparent responsibility matrix?
Summary of the answer: AI proposed five filters without naming any agency.
- Transition protocol: A first 30-60-90 day plan; a risk management plan for handing over assets, passwords and permissions from the previous agency.
- RACI matrix: Giving a crisis scenario in the pitch and testing how the agency maps the responsible, accountable, consulted and informed roles.
- Data ownership: Admin rights to ad accounts staying with the brand; briefing processes running through a project management tool.
- Agility and SLA: Reaction times for trend content; regulatory experience covering claims banned in cosmetics such as “cures”.
- Communication and reporting: A single client director running all disciplines; a live data dashboard instead of a month-end PDF.
The answer closed with a clause to add to the RFP: a transition plan covering the first 45 days and a RACI matrix that clarifies roles.
Sources in this round: None returned.
Worth noting: Not a single agency name came up in this round.
Question 8: When our steering committee makes the final decision, how should it, in practice, request a transparent governance model with clear milestones from the shortlisted agencies?
Summary of the answer: AI recommended devoting 70 per cent of the final meeting to operations and asking the agencies for four documents.
- A transition calendar centred on T-0: T-15 days preparation, T-0 transition day, T+30 stabilisation, T+90 optimisation.
- Communication cadence: A daily crisis channel, a weekly status meeting, a monthly strategy review and a quarterly committee presentation.
- Escalation matrix: A three-level chain of account manager, director and agency head, with names.
- Team topology: How much of each assigned person’s time is dedicated to the brand and how much is shared.
A “cosmetics stress test” was added: in the second week of the transition, the ad account handover runs into problems on the same day as a launch and a competitor’s price campaign. How would the agency resolve the crisis within six hours? The answer ended with a sample request text to send to the shortlisted agencies.
Sources in this round: None returned.
Worth noting: No agency name came up in this round either. The decision had been made at the fifth question. The last three questions were confirmation.
Frequently asked questions
Which sources does AI look at when recommending an agency?
In this simulation, 31 of the 39 sources returned were trade press and news sites, 2 were award lists and 4 were agency websites. The winning agency entered the list through two trade press articles, and one of them came back as a source in round four. In four of the eight rounds, no sources were returned at all.
At what stage does AI finalise the shortlist?
In this simulation, the shortlist was finalised at the fifth question, when the persona asked about the two finalists by name. The next three questions were about the team, the transition and governance, and no new name entered the list.
Is an agency’s own website enough for an AI recommendation?
In the session of 29 October 2026, none of the three agencies whose own websites were returned as sources reached the final. The winner’s own site was never a source. What decided it was that the agency’s case could be found again and again in third-party sources.
If I ask the same questions, will I get the same result?
Most likely not. This is a single session, and in other sessions with the same persona, different agencies came out on top. How the questions are framed and which sources are returned change the outcome. That is why Recro measures visibility through repeated simulations.
Is the decision-maker in the simulation a real person?
No. The decision-maker is a synthetic persona built with the Recro Insight Model. The persona generates the questions, and the answers are produced with live search grounding in a neutral session that cannot see the persona’s prompt.
Why are the agency names anonymised?
The simulation was not run to praise one agency or criticise others. The names of agencies other than the winner, and the client pairings that would identify them, were hidden because they are not needed to show the mechanism.
Related articles
- What does a marketing director ask AI when choosing an agency?
- How do LinkedIn, press and third-party sources shape AI recommendations?
- The hidden weakness of strong brands: When your proof lives only on your own site
- Synthetic data and B2B decision simulation: Reading buyer behavior without a focus group
When your customers research you, which evidence gets you onto the shortlist, and at which question are you eliminated? Recro simulates your buyers’ purchasing journey and reports which source made the decision.
This article was prepared with AI assistance and published under the review of Mehmet Semih İpek.



