Direct answer: choose an AI marketing consultant who measures results up to the money in your bank, shows their own data and failures, asks for the minimum access, knows consent law, and has a rollback plan. Ask the 12 questions below, then fill in the 100 point scorecard. Reject any candidate who promises ROASROASRevenue from ads divided by ad spend. A ROAS of 4 means each Rp 1,000 of ad spend returns Rp 4,000 of gross revenue.How to calculate ROAS and break-even ROAS, sales, or rankings.
Main condition: this guide is for businesses that pay for ads, get click-to-WhatsAppCTWAA Click-to-WhatsApp ad. It sends people from Facebook or Instagram straight into your WhatsApp conversation.How to track CTWA ads ad (CTWA) chats, or want AI in the sales funnelFunnelThe order of stages a buyer passes, from meeting you to paying you.Open the glossary. Limit: the scorecard filters risk, but it does not predict results. For AI in general operations, such as documents or warehouse automation, read the guide to choosing an AI consultant for business.
Rama Aditya wrote this guide from internal Rama Digital audits: 227 Wulan chat sessions (6 to 20 September 2026), 273 WhatsApp chats (28 to 30 September 2026), and daily Meta ad audits. All numbers are aggregate, with no client names. Data updated 3 October 2026.
How is choosing an AI marketing consultant different from choosing a general AI consultant?
A general AI consultant looks at workflows, data, and automation. An AI marketing consultant touches 3 more sensitive items: daily ad money, the personal data of prospects, and the buying signals you send to Meta or Google. A mistake in those 3 items burns budget or breaks the law at once.
So the 12 questions here focus on measurement, ad account access, consent, stop rules, and rollback. For the role, the scope, and when a business needs a consultant, read the complete guide to AI marketing consultants first. If you still compare work models, read AI marketing consultant vs digital agency vs in-house team.
12 questions for a candidate AI marketing consultant
Ask these questions in the first session. Write down the answers in the words of the candidate. Written answers are easier to score.
1. Which data do you read in the first week?
Good answer: the candidate names specific sources: Ads Manager, GA4GA4The Google analytics tool that records visits and actions on your site as events.What is Google Analytics 4 (GA4)? or the PixelPixelA piece of code on a web page that tells the ad platform somebody opened the page or took an action.What is the Meta Pixel?, chat transcripts, and the list of paid invoices. The candidate also names a period, for example the last 30 days.
Red flag: the candidate offers an AI tool or a content package before they look at your data.
2. How do you measure results up to the money received?
Good answer: the candidate explains how to join the click, the chat, and the payment. Example: a visitor code, the Conversions APICAPIA server path that sends conversion data from your system to the ad platform, without depending on the buyer browser.What is the Conversions API (CAPI)? (CAPI), and paid invoices. The mechanism is in AI marketing attribution from ad to bank transfer.
Red flag: the report stops at CTRCTRClicks divided by impressions, as a percentage. CTR shows how often people press the ad or the search result they saw.Open the glossary, CPMCPMThe cost of 1,000 ad impressions. CPM prices reach, not results.Open the glossary, or chat count. Those numbers are cheap and can rise with no buyers.
3. Do you guarantee ROAS, sales, or rankings?
Good answer: the candidate says no. The candidate guarantees the process: a report schedule, decision rules, and response time. Google Search Central states that no one can guarantee a number 1 ranking on Google (page updated 5 June 2026, read 3 October 2026).
Red flag: the proposal states a ROAS figure, a page 1 position, or "no loss" as a promise.
4. Which access do you need, and why?
Good answer: the candidate asks for partial access per asset, for example the ad account and the Pixel only, and writes a reason for each one. Meta separates full control and partial access in a business portfolio. Meta also has partner access, so you do not share a login.
Red flag: the candidate asks for full control of the business portfolio, a password, or admin rights with no written work reason.
5. How do you handle consent and customer personal data?
Good answer: the candidate names Indonesia Law Number 27 of 2022 on Personal Data Protection (UU PDP). Article 20 makes explicit consent one legal basis for processing. Article 46 requires notice of a protection failure within 3 times 24 hours. The candidate also explains that ad events go out only when the visitor gives marketing consent.
Red flag: the candidate does not know UU PDP, sends raw WhatsApp numbers to an ad platform, or thinks consent is only a legal team task.
6. Can you show your own data, including failures?
Good answer: the candidate shows numbers with dates, sample sizes, and 1 failed campaign with the lesson from it.
Red flag: the candidate has only ROAS screenshots with no date, or refuses to talk about a losing campaign.
7. When do you stop an ad or an experiment?
Good answer: the candidate has a written stop rule with numbers. Our rule: pause an ad that spent more than Rp 50,000 with 0 chats. The full audit method is in auditing Meta ads with AI.
Red flag: the candidate only says "wait for the algorithm to learn" with no cost limit or time limit.
8. How do you judge results that arrive late?
Good answer: the candidate knows the gap between chat and payment. In our data, class buyers pay 0 to 2 days after the chat. So we wait 48 hours before we pause a sales campaign.
Red flag: the candidate stops a campaign on the same day based only on Meta purchase counts.
9. What is your rollback plan if an AI change damages results?
Good answer: the candidate records the old version, writes the steps back to it, and tests them before release. Rollback covers chatbot prompts, automated rules, and ad settings.
Red flag: the candidate keeps no version record, or edits the chatbot prompt directly in production.
10. Who checks AI replies before they reach customers?
Good answer: the candidate explains regular audits with real chat samples and manual reading. Our audit results are in AI WhatsApp chatbots for sales closing.
Red flag: the candidate trusts an automatic score fully and never reads transcripts.
11. Whose name is on the accounts, code, and data?
Good answer: the candidate says your name, from day 1. The consultant gets access, not ownership.
Red flag: the ad account, the Pixel, or the WhatsApp Business number is in the name of the consultant or their agency.
12. How do you calculate fees, and what is not included?
Good answer: the candidate separates the service fee from the media budget, writes the price in the quotation, and names third-party costs. Pricing models are in the cost of an AI marketing consultant.
Red flag: a percentage fee on budget with no upper limit, or a renewal fee that appears only after the contract starts.
A 100 point scorecard to grade the answers
Give each criterion 0, half the weight, or the full weight. Ask for the proof in the "how to check" column, not a spoken promise. The weights are the Rama Digital recommendation for businesses that run ads and receive chats.
| Criterion | Weight | How to check |
|---|---|---|
| Measures up to money received (questions 2 and 8) | 20 | Ask for a sample report with paid invoices per campaign. |
| Own data and failures (question 6) | 15 | Ask for 1 dated number with a sample size and 1 failed campaign. |
| Consent and UU PDP (question 5) | 15 | Ask when an ad event does not go out. The answer must name marketing consent. |
| Minimum access (question 4) | 10 | Ask for a written access list with a reason for each item. |
| Stop rule (question 7) | 10 | Ask for a rule with a cost figure and a time limit. |
| Rollback (question 9) | 10 | Ask for a sample version record and the steps back. |
| Human review (question 10) | 10 | Ask for the chat audit format and its sample size. |
| Asset ownership (question 11) | 5 | Check the account and data ownership clause in the contract. |
| Clear fees (questions 1, 3, and 12) | 5 | Check that the fee and the media budget are 2 separate figures. |
Rama Digital recommendation: reject a candidate below 60. For a score of 60 to 79, start with a small project with a time limit. For a score of 80 or more, go to a written quotation.
5 red flags that end the talk
A high score does not save a candidate with one of these red flags. Stop the talk when you find one.
- A ROAS, sales, or ranking guarantee. Ad results depend on the product, the price, the market, and how fast your team answers chats.
- Full control with no reason. Partial access per asset covers almost all daily work.
- No wish to show their own data. A candidate who does not measure their own business seldom measures yours correctly.
- No knowledge of consent and UU PDP. The legal risk falls on you as the data controller, not on the consultant.
- No rollback. One wrong prompt or rule can run in every chat before anyone sees it.
How Rama Digital answers these 12 questions
We use the same standard on ourselves. This table shows the proof for each question, with dates and sample sizes.
| Question | Our proof | Date and sample |
|---|---|---|
| 6. Own data and failures | Wulan, our chatbot, asked for the order in only 2 of 43 replies. 149 sessions (66%) stopped after the contact was recorded. | Audit of 20 September 2026, 227 sessions |
| 6. A measured fix | After the fix, Wulan asked for the buying step in 215 of 273 chats (79%). 182 people (67%) sent only the prefilled ad text. | Audit of 30 September 2026, 273 chats |
| 6. A failed campaign | Maps, Chatbot, and MCP ads spent Rp 1,467,160 and brought 135 chats with 0 orders. | 28 to 30 September 2026 |
| 7. Stop rule | A class urgency ad set that we changed to a broad audience spent Rp 155,034 with 0 chats. We moved it back to warm audiences. | 26 September 2026 |
| 4. Placement and access | WhatsApp Status took 98% of the Jasa MCP ad impressions and 58% of its spend, with no buyers. Since then, every ad set runs only on Facebook and Instagram. | 27 to 29 September 2026 |
| 5. Consent | Events go to Meta only when the visitor gives marketing consent. 11,507 events in 7 days did not go out because that consent was missing. | Record of 27 September 2026 |
| 10. Human review | An automatic grading model flagged 33 class chats as "wrong fact". A manual read of 63 chats proved those flags wrong. | Audit of 30 September 2026 |
| 11. Asset ownership | Accounts, code, and data stay in the client name from day 1. No renewal fee to Rama Digital. | Promise on every service page |
Our most expensive lesson: a cheap chat is not a buyer. Ads on automatic placements gave chats at a low cost, but no buyer came from those placements. We now judge ads by paid invoices, not by cost per chat.
For rollback, every Wulan release records the old Worker version. Our operations document writes the steps back to that version, with the checks before a rollback.
Worked example: grading 2 candidate consultants
This is a simulation with dummy data, not a real client. Input: an online shop with a Meta ad budget of Rp 300,000 per day and sales through WhatsApp. The owner receives 2 proposals.
| Step | Candidate A | Candidate B |
|---|---|---|
| 1. Ask the 12 questions | Promises ROAS 5 in 30 days | Refuses to promise ROAS, writes a stop rule |
| 2. Ask for their own data | 3 ROAS screenshots with no date | 1 dated audit with 1 failed campaign |
| 3. Check the access request | Full control of the business portfolio | Partial access to the ad account and Pixel |
| 4. Fill in the scorecard | 35 of 100 | 82 of 100 |
| 5. Check red flags | 2 red flags: ROAS promise and full control | 0 red flags |
| Decision | Reject | Go to a written quotation |
Output: 1 written quotation from Candidate B with an access list, a stop rule, and a report schedule. The owner keeps the scorecard as the basis for a review after 30 days.
Checklist before you sign
- Send the 12 questions in writing. Owner: you. Proof: written answers from the candidate.
- Fill in the scorecard for each candidate. Owner: you. Proof: the score table.
- Ask for 1 dated number and 1 failure from the data of the candidate. Owner: you. Proof: a document or a link.
- Give partial access through Meta partner access, not a password. Owner: account admin. Proof: the access list in the business portfolio.
- Write a stop rule with a cost figure and a time limit into the contract. Owner: you and the candidate. Proof: a contract appendix.
- Ask for a rollback plan for the chatbot and automated rules. Owner: the candidate. Proof: a version record.
- Check the clause that puts accounts, code, and data in your name. Owner: you. Proof: the contract.
- Stop criterion: cancel the process when you find 1 of the 5 red flags above.
Frequently asked questions
How is this guide different from the guide to choosing a general AI consultant? This guide is for marketing only: ad accounts, chats from ads, buying signals, and consent. The general guide covers AI for business workflows and automation.
Is a consultant who refuses to guarantee ROAS less confident? No. Ad results depend on the product, the price, the market, and the speed of your team. An honest consultant guarantees the process and the reports, not a sales figure.
Which access is safe to give in the first week? Give partial access to the ad account, the Pixel, and GA4 to read data. Add edit rights only after you approve a written work plan.
Does UU PDP apply to a small business? Yes. Law 27 of 2022 applies to every party that processes personal data, including the WhatsApp numbers and emails of prospects. The 2 year transition period ended in October 2024.
What is the minimum score to start working together? Our recommendation: 60 for a small project with a time limit, and 80 for a longer contract. One red flag still rejects a candidate at any score.
Next step
The limit that still applies: this scorecard does not guarantee ad results. It only lowers the risk that you pay a consultant who does not measure, is not transparent, or is not safe with data.
If you want to test us with these 12 questions, start with a free 30 minute first consultation. For a written problem map in 1 session of 60 to 90 minutes, choose AI Diagnostic (Rp 1.5 million). For a longer work path, read AI Consulting for business.
Sources
- Law Number 27 of 2022 on Personal Data Protection, JDIH BPK, read 3 October 2026.
- Google Search Central: Do you need an SEO?, updated 5 June 2026, read 3 October 2026.
- Meta: About business portfolio and business asset permissions, read 3 October 2026.
- Meta: Give a partner access to business assets, read 3 October 2026.
- Internal Rama Digital audits: Wulan chats on 20 and 30 September 2026, Meta ad audit on 26 September 2026, ad rules on 30 September 2026.
