EBITDA up 3.8 points, CAC down 41%: turning 61,000 dead followers into a customer data capture funnel with the Restaurant Model Canvas
An owned customer data capture funnel gave this operation 3.8 EBITDA points back in seven months, and not because the posts got better: because the restaurant stopped renting its guest relationship from an algorithm. Case profile: Italian casual dining, 26 tables, 31 employees, a mid-size city of 900,000, average check of 34 USD, eleven years open, Instagram and Reels as the dominant channel with 61,000 followers. Annual revenue of 1.4 million dollars —the above-1-million band— with 4,100 dollars a month in paid media just to fill Tuesdays. Baseline: zero owned records, 92% of traffic riding on organic reach, acquisition cost of 21.40 USD per new guest, 18% repeat rate at 90 days. By month seven: 14,700 permissioned contacts, CAC of 12.60 USD, 34% repeat rate, and paid media down to 2,300 dollars with no loss of covers. The lever was never creative. It was structural.
The owner arrived with a sentence that sums up half the industry: «I have 61,000 followers and the dining room is empty on Tuesdays». His diagnosis was right, his causal theory was wrong. Content was not missing —nine Reels a month, decent editing, occasional spikes at 200,000 views—, OWNERSHIP of that audience was. Every hungry viewer stayed filed on a platform's server that would show them the restaurant again only if the restaurant paid.
Revenue band matters here more than it looks. An operation above 1 million a year can carry a serious customer data capture funnel with proper tooling; one below 500,000 USD would run it on a spreadsheet and a printed QR, and it would still pay off. The sector's mistake is treating this as a marketing budget question. It is an asset architecture question: the database is CapEx dressed as OpEx, paid once and yielding for years.
Before touching a single creative asset we measured the only thing that mattered: how many June guests came back by September. Eighteen out of a hundred. With a 34-dollar check and a theoretical frequency of 2.1 annual visits, guest lifetime value barely reached 71 dollars while acquisition cost 21.40. That spread does not finance a business, it finances a hamster wheel with tablecloths.
Customer data capture funnel: side-by-side comparison
| BEFORE (baseline, March 2026) | AFTER (month 7, October 2026) | |
|---|---|---|
| Owned permissioned contacts (email or WhatsApp) | ✕0 records | ✓14,700 records |
| Customer acquisition cost (CAC) | ✕21.40 USD per new guest | ✓12.60 USD per new guest |
| 90-day repeat rate | ✕18% of guests | ✓34% of guests |
| Average check | ✕34.00 USD | ✓39.80 USD |
| Monthly paid media spend | ✕4,100 USD | ✓2,300 USD |
| 12-month guest lifetime value | ✕71 USD | ✓158 USD |
| Prime Cost (food + labor) | ✕68.4% of sales | ✓62.1% of sales |
| Labor Cost | ✕34.9% of sales | ✓31.2% of sales |
| Tuesday and Wednesday occupancy | ✕41% of seats | ✓73% of seats |
| EBITDA | ✕9.1% of sales | ✓12.9% of sales |
What we found on day one at a 26-table casual dining spot with 61,000 followers?
We found a restaurant with a rented audience and empty Tuesdays: 61,000 followers, nine Reels a month with peaks of 200,000 views, and a return rate of eighteen guests out of every hundred between June and September.
With a 34-dollar average check and a theoretical frequency of 2.1 annual visits, LTV sat near 71 dollars against an acquisition cost of 21.40, which is a hamster wheel with starched linen. The full profile: Italian casual dining, 31 employees, revenue band above 1 million a year. The cause of those empty Tuesdays was never creative. Every person who watched a video and felt hungry ended up filed away on a platform's server, and that platform would only show them the place again if the owner paid a second time for a contact his own kitchen had already earned.
Content stopped chasing views and started chasing sign-ups
Changing what the content was for moved viewer-to-contact conversion from 0.3% to 4.1% in eleven weeks, on the same budget and almost the same number of pieces. Nobody published more. Each Reel closed with a concrete reason to hand over a WhatsApp number: the waitlist for the seasonal menu, the twelve seats at Thursday's fresh pasta workshop, the truffle alert when it landed. That shift matches how people actually search, because 79% of restaurant searches are non-brand (Malou, 2025) and 'food near me' searches grew 99% year over year (Restroworks, 2024): discovery is abundant, what's missing is the bridge between a Thursday craving and a database you control. Without that bridge, the algorithm keeps the relationship and rents it back to you.
The table QR stopped being a menu and became the funnel's front door
We turned the table code into the cheapest capture point in the operation: scan, menu, and one single question in exchange for the birthday dessert, name and mobile. The habit was already installed and measurable, because QR scan volume grew 433% in two years (QR Code, 2025), ordering by code lifts check size 9% versus traditional dine-in (Sunday, 2025), and a complete digital offer covering menu, ordering and payment moves the check between 20% and 30% (Sunday, 2025). One warning belongs here: asking for email, birthday, preferences and pet size on the same screen collapses sign-ups. We asked for two fields, one of them optional, and capture per table closed at 27% of September's checks (internal measurement of the case).
Segment by check composition, not by age or postal code
The segmentation that worked was behavioral rather than demographic: we split the base into four groups by what they bought, never by who they were. Table for two with wine by the glass, family of four with a kids' menu, long six-person lingering dinner, and the midday executive who leaves in 38 minutes. That fourth group never received a lingering-dinner promotion, and that discipline is what makes a database profitable: loyalty program members buy more often, according to Paytronix (2024). The most repeated mistake is treating a database as a mailing list, when it is really a map of purchase intentions with first names attached.
Seven months, line by line: 3.8 points of EBITDA
The result was 3.8 points of EBITDA in seven months, and it came from four moves you can audit separately in the case's P&L. Tuesdays went from 41 average covers to 78, nearly all of them summoned by direct message to a segment, at a summoning cost of 0.31 dollars per seated guest against the 21.40 it cost to acquire a new one through paid media. Ninety-day retention climbed from 18% to 34%, which pushed LTV from 71 to 129 dollars without touching the price of a single dish. Ad spend fell 44%, because discounts stopped being a public lure and became a privilege of the base. The asset gets paid for once and yields for years: it is CapEx dressed as OpEx, and the industry still books it as a marketing line.
The Masterestaurant tool that held the funnel up: the frequency-and-cash dashboard
What held this funnel up was not a messaging platform but the frequency-and-cash dashboard of the Masterestaurant method, which Diego F. Parra uses to cross three columns almost nobody crosses: average check, days since last visit, and contribution margin of that check's anchor dish. We loaded eighteen months of history, marked the leak point at day 47 without a visit —past that threshold, the probability of return fell below 12%— and automated one single contact on day 40. Nothing else. One message, one real reason, zero discount in the first two waves. Masterestaurant does not sell loyalty systems: it measures whether the frequency you buy leaves margin after the cost of the incentive, and here the coupon's break-even sat at 11% off, not the 25% the owner had been giving away out of habit.
Transferable lessons by annual revenue band
Every band needs a different first step, and it starts this week. Under 500,000 USD: print a QR pointing to a two-field form and ask for name and mobile in exchange for the birthday dessert, with a shared spreadsheet and nothing more; it pays off just the same. Between 500,000 and 1 million: export the last twelve months from the POS and calculate your leak day, that threshold where return collapses, before buying any software. Above 1 million, like this case: segment by check composition and automate a single contact before the leak day. Above 5 million, the media-chef archetype with a strong personal brand must keep the restaurant's base separate from the celebrity's base, because those are two assets with different economics. Above 10 million, multi-site groups: one guest identifier across every location, or you will own five databases competing against each other.
Limits of this case
I would not expect these 3.8 points in three contexts, and it's worth saying so before someone copies the playbook. First, a high-turnover business with structurally low frequency —weddings, roadside hotels, tourist-destination dining rooms— where the guest fails to return because of geography rather than a missing message: there the base serves reputation and referrals, not full Tuesdays. Second, an operation without a stable kitchen: if Thursday's dish swings in quality, calling the base in accelerates customer loss instead of holding it, and the message turns into an amplifier of a mise en place problem. Third, a check below 12 dollars, where the incentive's cost eats the margin before frequency ever compensates. This case started from an already-built audience of 61,000 people; without that starting point, the same work takes twelve to eighteen months.
Where the difference actually came from?
The shift was not posting more, it was changing what the content was for. Reels stopped chasing views and started chasing REGISTRATIONS:
each piece closed with a concrete reason to hand over a WhatsApp number —the seasonal menu list, the Thursday fresh pasta workshop seats— and that single decision moved viewer-to-contact conversion from 0.3% to 4.1%. The table QR stopped being a menu and became the door of the customer data capture funnel. Sunday (2025) reports that QR ordering lifts check size by 9% against traditional dine-in service, and scan volume grew 433% in two years (QR Code, 2025); we rode a habit guests already had and asked for exactly one thing in exchange for the birthday dessert: name and mobile number. Segment by behavior, never by demographics.
Where the difference actually came from — in practice?
A guest who ordered the osso buco twice does not get the message that goes to someone who only drinks at the bar on Fridays.
Paytronix (2025) reports a 16.5% year-over-year lift in member spend when targeting is one to one, and in this operation the effect showed up in check size before it showed up in frequency. The data file lives inside the business, not inside the agency. That sounds like paperwork until the day you change agencies and the list walks out the door. Here the database stayed exportable, with dated consent and a recorded source for each entry, the condition without which everything else is legal vapor. Marketing stopped being measured by reach and started being measured by margin. Every campaign reports CAC, repeat rate and EBITDA contribution; whatever fails to move those three numbers gets switched off within a month, however nice the comments look.
Before and after, criterion by criterion
What was there: rented audience
- 61,000 Instagram followers, zero owned contacts, no way to message a guest without paying for it.
- 4,100 USD a month buying reach and never permission: every campaign restarted from zero.
- Reservations by direct message, written in a notebook; the guest's name died right there.
- 18% repeat rate at 90 days, guest lifetime value of 71 USD against a CAC of 21.40 USD.
- Tuesdays and Wednesdays at 41% occupancy with the full kitchen brigade and payroll running anyway.
What remains: an owned, measurable asset
- 14,700 explicitly permissioned contacts segmented by frequency, check size and favorite dish.
- Paid media cut to 2,300 USD a month, spent only on capturing new records rather than filling tables.
- A seven-message automation that wakes the dormant guest at day 45, 75 and 110.
- 34% repeat rate and 158 USD lifetime value: same guest, more visits, no structural discounting.
- 73% occupancy on the dead nights and 3.8 EBITDA points that used to leave through paid media.
The case numbers, seven months in
“I thought my problem was the algorithm and it turned out my problem was that after eleven years I did not have a single phone number of my own guests. The first month, asking for a WhatsApp at the table felt like a waste of time, until we sent one message to 3,400 people on a Tuesday morning and that night 62 covers walked in without a dollar of paid media. Today I hold 14,700 contacts, I cut paid media from 4,100 to 2,300 dollars a month and EBITDA climbed 3.8 points.”
Composite case for illustration: the names and figures in it do not describe a real business and are not industry data.
The actual deployment timeline
We put the whole model on one sheet to see where the guest relationship was born and where it died. The obvious surfaced the moment it was drawn: four touchpoints —Reels, direct message, table, payment— and not one of them captured a datum. We measured the raw baseline: CAC of 21.40 USD, 18% repeat rate at 90 days, Labor Cost of 34.9% inflated by full brigades on nights running at 41% occupancy. Root cause was never reach. The sales funnel simply ended at the cash register instead of ending in a record.
We defined what someone actually receives in exchange for their data, because «subscribe to our newsletter» is worth nothing. Two hooks with real kitchen value made the cut: the waiting list for the fresh pasta workshop, 18 seats a month, and a birthday dessert valid for seven days. The table QR was rebuilt so the first screen asked for name and mobile before the menu. Here came the first serious friction: initial conversion sat at 0.9% because the form demanded email, phone, birth date and preferences. We cut it to two fields and it climbed to 4.1% within eleven days.
We rewrote the Reels script with the floor team. Nine pieces a month, same production, different architecture: the first three seconds show the dish cooking, the close names the limited seats and sends people to the link. We added the Demand Radar to pick each week's topic from what the city was actually searching, not from whatever the chef felt like filming. One industry figure framed that call: according to Malou (2025), 79% of restaurant searches are non-branded, so the content had to compete for the craving rather than for the name.
With 6,200 contacts already banked we built the reactivation flow: a message at day 45 naming a specific dish, day 75 with the seasonal menu, day 110 with an invitation to a slow night. Paid media dropped from 4,100 to 2,300 USD and the difference went into pure acquisition. Second friction, an expensive one: for three weeks we messaged everyone identically and the unsubscribe rate hit 6.4%. Once we segmented by frequency and check size, unsubscribes fell to 1.1% with double the open rate.
With Tuesday and Wednesday occupancy already at 68%, we tuned the menu through the Standard Recipe Generator so the dishes selling through messages stayed under 32% food cost. Average check rose to 39.80 USD without touching the price list, purely through order composition. Labor Cost landed at 31.2% because occupancy stopped being random and shift planning could lean on forecast demand. The result consolidated in month 7 and held for three more months before we closed the measurement.
And with AI?
Accelerate content, targeting and repurchase: more reach with less effort. Diego F. Parra is an expert in AI applied to restaurants.
Free tools for customer data capture funnel
The tools that held the funnel up
None of this was custom-built. We used closed products from the Masterestaurant suite, in the order the business could absorb them, and that constraint was deliberate: a system that needs a consultant in the room dies the day the consultant leaves.
The selection criterion was simple and slightly unfriendly. If a tool could not produce a number that landed in the P&L within the first ninety days, it did not enter the deployment however elegant its interface looked.
Diego F. Parra — International consultant, expert in creating and scaling restaurants and in AI applied to restaurants, foodtech and HORECA. Methodology applied in 8.400+ restaurants across 43 countries · Expert in Artificial Intelligence applied to restaurants, hospitality and food businesses · 20+ years in restaurants, catering, large events and business growth · Author of 3 ISBN-registered books: «Triunfar o morir en el intento» (2013) and «De esclavo a dueño» (2023) · International keynote speaker for the HORECA sector.
Questions that always follow this case
How many contacts does a customer data capture funnel need to pay off?
How many contacts does a customer data capture funnel need to pay off?
You feel it at 800 active contacts. There is no magic number, the ratio is what rules: if your database is at least double the guests you serve in a month, one well-segmented message can fill a dead night. In this case the first profitable send went out to 3,400 records and returned 62 covers that same evening.
Does this work for a restaurant below 500,000 USD a year?
Does this work for a restaurant below 500,000 USD a year?
It works, and it pays off proportionally more because it starts from zero. An independent in that band needs no platform: a table QR, a spreadsheet with name, mobile and last visit date, and the discipline to send something every fortnight. Customer acquisition cost usually falls faster in small operations because their market radius is shorter.
How long before EBITDA moves?
How long before EBITDA moves?
Five to seven months in an operation this size. The first ninety days build the base and barely touch the result; margin appears once repeat rates climb and you can cut paid media without losing covers. Any improvement you see before month four is seasonality rather than system.
What if my audience lives on TikTok instead of Instagram?
What if my audience lives on TikTok instead of Instagram?
The hook format changes, the architecture does not. TikTok converts worse to direct registration and better to discovery, so the capture link carries more weight and the reason to click has to be stronger. The rule holds either way: while the relationship lives on the platform and not in your database, you rent guests instead of owning them.
2026 data on customer data capture funnel
Verifiable industry benchmarks from official, non-commercial sources (government, industry associations, market research) - not competitors.
| Metric | Value | Source |
|---|---|---|
| average Instagram engagement rate per post against reach | 0.48%: tasa media de interacción en Instagram, calculada como interacciones (likes+comentarios) sobre seguidor | Socialinsider — 2026 Instagram Organic Engagement Benchmarks |
| of consumers have purchased from one brand over another based on the service they expect to receive | 60% of consumers have purchased something from one brand over another based on the service they expect to receive (2023) | Zendesk — 35 customer experience statistics to know (cita el Zendesk CX Trends Report 2023) |
| Retention increase that lifts profit by 25% to 95%, the economic case for LTV over pure acquisition | aumentar las tasas de retención de clientes en 5% aumenta las utilidades entre 25% y 95% (2014) | Harvard Business Review — The Value of Keeping the Right Customers 2014 |
| Maximum commission delivery aggregators charge per order | 15% a 30% por pedido (2025) | Independent Restaurant Coalition — Delivery Apps 2025 |
| Profit increase (25%-95% range) from a 5-percentage-point increase in customer retention, per Frederick Reichheld/Bain & Company research | 25% a 95% de aumento de utilidad (2014) | Harvard Business Review (citando investigación de Frederick Reichheld, Bain & Company) — The Value of Keeping the Right Customers 2014 |
| profit increase from just 5 additional points of retention | 25% to 95% increase in profits from a 5% increase in customer retention (2014) | Harvard Business Review / Bain & Company (investigación de Frederick Reichheld) — The Value of Keeping the Right Customers 2014 |
Related content
Customer data capture funnel: repeat this case in your restaurant
Applied in +8.400 restaurants across 43 countries.