The board meeting where the engagement started was 40 minutes long. The founder had a spreadsheet on screen and it was, by his own description, not the spreadsheet he had planned to show. Three of the five locations were burning cash. The fourth was flat. The fifth was carrying the group. Payroll for the whole $30M business was being covered week to week by the one healthy unit and by the founder's own reserves. The catering channel that was supposed to open in Q2 had been paused. The plan to expand from three to five locations was on hold.
He asked me one question. If you had eleven months and a scope of work that could touch three locations, what would you do first? I gave him the honest answer, which was that I would spend the first month not touching anything, and if that was not acceptable he should hire someone else.
He hired me anyway. Eleven months later the three units were contributing roughly $3.1M annually against the $1.8M loss they had been running when the engagement started. That is the $4.9M number people ask about. This is how it actually happened, month by month, including the parts that did not go well.
The picture at day zero
The group was a Michelin-recognized Bay Area operation, five locations, roughly $30M in annual revenue, 215 employees across front of house, back of house, and central support. Two of the five units were healthy. Three were not. The healthy units were carrying labor at 26 percent of sales and food at 30 percent. The three underperforming units were running labor between 34 and 39 percent, food cost between 34 and 37 percent, and one of them had a repair-and-maintenance line that had grown 180 percent year over year and nobody knew why.
The general managers of the three underperforming units were not bad operators. Two of them had been high performers at the healthy units and had been promoted to open the newer sites. All three had inherited a P&L they did not fully understand and a scheduling habit that had been built by whoever last covered the shift, not by anyone reading the demand curve.
The founder was exhausted. The chef, who was also a partner and the reason the group had Michelin recognition, was frustrated with the operations side and was not sure what to make of an outside operator being brought in. My first hour on day one was a coffee with the chef, not a meeting with the founder. That coffee, in retrospect, was the most important 60 minutes of the whole engagement.
The chef partner in a Michelin-recognized group is not a stakeholder. The chef partner is a co-founder. Treat the operations work as scaffolding around the culinary program, not as a redesign of it, and the whole engagement changes character.
Month 1: The diagnostic
The first month did what I said it would do, which was not change anything. I walked all three underperforming units on Friday dinner and Sunday brunch services. I read 90 days of daily P&L for every location, line by line, in a plain spreadsheet with no dashboards. I sat with each general manager for two hours, one on one, and asked the same set of questions about the decision rights they actually held.
What I found looked exactly like every turnaround diagnostic I have run. Labor cost variance was volatile day to day inside the same week, which is the signature of a schedule that was not built off the demand curve. Food cost was 4 to 6 points high, and a physical inventory across all three units on the same Tuesday morning surfaced roughly $47K of walk-in inventory that did not exist on paper. Comps and voids were 3 to 5 points of sales at all three units, and about half of them had no code attached, which meant the P&L was quietly understating COGS.
The general managers all told me some version of the same story: they had been asked to run a P&L they had never been trained on, with a scheduling tool that did not talk to the POS, using inventory data they knew was wrong, while the founder called them weekly with new priorities that did not line up with the priorities from the previous week. They were not underperforming. The system around them was.
On day 30 I delivered a written diagnostic to the founder. Nine pages. Three parts. Here is what is actually wrong. Here is the sequence to fix it. Here is what has to be true at the end of month 11 for us both to agree the work is done. He signed it that afternoon.
Fig. 1 · The recovery curve, month by month. The steepest slope was months 4 through 8.
Months 2 and 3: Stabilization
The stabilization phase attacked three cost lines in order: labor variance, food cost, and comps. All of it was done with the general managers, not to them, which took longer in the first two weeks and paid back the time three times over by month six.
The schedule rebuild
Location A got its schedule rebuilt first, because the demand data was cleanest. We pulled 12 weeks of hourly sales by day part, overlaid the schedule as it existed, and identified 14 hours per week per station where the schedule and the demand curve disagreed by more than 25 percent. The rebuild moved the same total labor dollars into different hours. Not a headcount cut. A redeployment. Labor as a percent of sales at that location moved from 37 percent to 32 percent inside six weeks, without a single termination.
Location B took longer because the demand data was noisy. Location C we did not touch until month 3 because I had made a mistake at Location A that I did not want to repeat: I had rebuilt too early and lost two weeks of clean signal. At Location C we waited for full data before moving.
The comp code discipline
At the end of week 5 we installed a written comp code requirement at the point of sale across all three locations on the same night. Every dollar comped required a code and a manager PIN. No policy change. Just tracking. Within two weeks the comp line at all three units dropped 28 percent on average, purely because the act of tracking suppressed casual comping. The remaining comps, now coded, showed us where the actual guest-recovery patterns were, which turned out to be one server at Location B who was single-handedly responsible for 40 percent of the site's comps. She was not skimming. She was over-serving to compensate for a kitchen ticket-time problem she had never been given tools to escalate. That is a different fix.
The inventory reset
We did a full physical inventory across all three walk-ins on the same Tuesday morning in week 8. Reconciled to the system. The $47K of missing inventory was mostly waste that had never been logged, plus about $8K in vendor short deliveries the receiving process was missing. Neither was theft. Both were process. We put in a five-minute closing ritual where the closing manager wrote the day's waste and comps into the system before locking up. Food cost across the three units dropped 2.4 points inside 60 days.
Month 4: The rhythm
By month 4 the stabilization work was producing measurable results, and this was the moment I made my third real mistake of the engagement. I had underinvested in the operating rhythm because the numbers were moving and it felt like the stabilization work was doing all the lifting. It was not. The stabilization work was doing about 55 percent of the lifting. The rhythm was going to have to do the other 45 percent, and I had not yet built it.
In month 4 I finally installed the three things I should have installed by day 40. First, the weekly P&L review with each general manager, 45 minutes every Monday, same agenda every week: last week's labor as a percent of sales, food cost, top three variances, one thing to fix this week. Second, the monthly regional operating review with all general managers in the same room, each presenting their P&L and next month's plan in five minutes. Third, and most important, the dashboard the general manager opens every morning on their phone. Labor variance yesterday, food cost yesterday, top three items sold, comps and voids yesterday.
The dashboard was the single highest-impact tool in the whole turnaround, and I built it 40 days later than I should have. Every future engagement I run, the dashboard is a day 40 deliverable, not a month 4 one.
What I underestimated was how much the general managers wanted the rhythm. I framed the weekly review as accountability. They told me later they experienced it as relief. For the first time, someone was going to look at the numbers with them every week.
Months 5 through 8: Compounding
Months 5 through 8 were the phase where the fixes started reinforcing each other, and the P&L started moving faster than the sum of the individual interventions. Better schedules meant better service, which meant fewer comps, which meant cleaner comp data, which meant the coding pattern started to reveal specific service problems, which meant we could fix root causes instead of symptoms.
In parallel, we launched the enterprise catering channel that had been paused when the engagement started. This was the founder's project originally, and my job was to build the operational chassis under it so it could actually run at scale without cannibalizing dine-in labor. We built a dedicated catering prep line at Location B during the slow morning hours, hired one dedicated catering ops manager, and started opening accounts one at a time. By month 8 the channel was doing recurring corporate drops into Stanford, Google, Apple, Meta, LinkedIn, Salesforce, Cisco, Adobe, and Nvidia. Contribution margin on the catering channel was 32 percent, materially better than dine-in, and it added a revenue line that was not competing with any existing capacity.
The two healthy locations, which were not part of the turnaround scope, also improved during this period because the tools we built were shared. The dashboard, the weekly review, the closing ritual. All of it got adopted at the healthy units within a quarter because the general managers there saw their peers using it and asked for the same setup. That was not part of the scope. It was a byproduct of the rhythm.
Fig. 2 · The gap between the before and after did not come from any single cost line. Four small moves.
Months 9 through 11: The handoff
By month 9 the engagement was in handoff mode. Every dashboard, every standing meeting, every SOP had a named internal owner from the day it was built. In the last three months, my job was to sit in the rooms and not run them. Two of the three general managers ran their Monday reviews with me silent from the corner. One needed extra coaching and we pushed his handoff two weeks. That was a healthy conversation, not a failed one.
In month 10 the group opened its fourth location, and in month 11 the fifth. Both openings ran on the same operating rhythm we had installed at the turnaround units. The dashboard, the Monday review, the monthly regional review. The expansion happened not despite the turnaround but because of it. The recovered $4.9M is the reason there was capital to open the two new sites.
On the last day I did a written closeout with the founder. Numbers versus the day 30 diagnostic. What we did. What we did not do. What I would do differently. The founder wrote back a two-sentence email that I have kept: "You changed less than I expected. I understand now why that mattered."
What actually did the work
People ask what the single biggest move was. The honest answer is that no single move accounted for more than 30 percent of the swing. Three moves together accounted for about 70 percent, and the other 30 percent was small, boring, unglamorous fixes that added up over 11 months.
The three big moves: the schedule rebuild at each location, worth about 4 to 6 points of labor as a percent of sales. The comp code requirement at the POS, worth about 2 points of margin and roughly $220K per year in recovered comps. And the launch of the enterprise catering channel into the tech campuses, worth an incremental revenue line at 32 percent contribution.
The other 30 percent: the physical inventory reset. The petty cash cap. The overtime approval requirement. The vendor autopay audit that killed $2,400 a month of dead subscriptions. The Monday P&L review that turned three general managers who had been running blind into three operators who could tell you their prior-week labor variance from memory. Each of those was small. All of them together were most of the difference between a stabilization and a real turnaround.
The temptation is to tell the story as three heroic decisions. It was not three heroic decisions. It was forty small ones, made in the right order, with the general managers holding the pen, over eleven months.
What I would do differently next time
Three things, in order of how much they cost me.
One: build the dashboard on day 40, not month 4. The dashboard is the single most important tool for making the rhythm real. Every day it does not exist is a day the general manager is running on memory instead of data. If I could redo one thing about the Zareen's engagement, this is the one.
Two: rebuild the first schedule two weeks later than I did. I moved on Location A in week 2, and I lost two weeks of clean diagnostic signal because I could not tell what was my change and what was the underlying pattern. Even under pressure to show early motion, the schedule rebuild belongs in week 4 or 5, not week 2.
Three: frame the weekly review as support, not accountability. I framed it as accountability in the first month, and it took the general managers three or four weeks to understand that the meeting was for them, not for me. If I had framed it correctly from day one, we would have gotten to the trusting version of that meeting a month faster, and the whole rhythm would have gone in a month earlier.
The point
The $4.9M number is real. It is also, on its own, misleading. The recovery did not come from a decision. It came from a sequence, run patiently, with the operating team holding the pen and an outside operator making sure the sequence stayed intact. The stabilization work produced roughly 55 percent of the recovery. The operating rhythm produced the other 45 percent, and without it the first 55 percent would have decayed inside two quarters.
If there is a single lesson from the eleven months, it is not any of the individual moves. It is that a multi-unit turnaround is a cadence problem, not a personality problem. The standing meeting, the shared dashboard, the number the general manager can name from memory. That is what holds. That is the whole game.
Cadence beats charisma. Every time.