Adan Tran Data and BI analyst, Nashville area LinkedIn

Then I build the fix with data and AI.

monthly sales against the first full month: 4.7 times

One dot per ticket: 26,036 tickets at VSL Nail Spa
VSL Nail Spa is the salon I co-own. Dark dots are the first full month's tickets, repeated every month. Red dots are the tickets above that. The line is monthly sales. See how I measured it

I spent over two years in data consulting for Fortune 500 clients. Then I became co-owner of a new nail salon, an industry I had never worked in.

These are the things I built there. Each one is written the same way: the gap I saw, what I used, what I built and what changed.

I work in Power BI, SQL and Python.

Analytics

Sales analytics for a 16-technician salon

VSL Nail Spa opened to clients in Hendersonville, TN, at the end of January 2025. Every ticket went into the point-of-sale system, but its built-in reports could not answer an owner's questions.

The gap
Which services earn the most? Which days need more staff? How is each technician doing? The POS could not say.
What I used
The POS sales export (one row per service on each ticket), QuickBooks cost data, SQL, Python and AI coding tools.
What I built
A sales dashboard: average ticket, sales per technician per day, service mix, tips, demand by weekday and supply cost. It refreshes monthly from a new export.
What changed
I can answer those questions in minutes, from the data, like the four below.

The charts use the salon's real data: 71,283 sales lines from 26,036 tickets, Jan 2025 to Oct 2026. Dollar amounts are replaced by an index and technician names are removed.

Sales grew 4.7x in 20 months, and the growth came from more visits

Monthly salesClient visitsIndex, Feb 2025 = 100
By the fourth full month, sales were 3.5x the first. They peaked at 5.6x in May 2026. Visits rose almost as fast, while the average ticket stayed 7 to 20 percent above the first month. More visits drove the growth; a higher average ticket added about 12%. Sep 2026 is not in the export, so both lines break there. The chart at the top of the page stacks the 26,036 tickets by month using this visits index.
View as table
MonthSales indexVisits indexAverage ticket index
Feb 2025100100100
Mar 2025190166114
Apr 2025227207110
May 2025345311111
Jun 2025342318107
Jul 2025353311113
Aug 2025380322118
Sep 2025332291114
Oct 2025342303113
Nov 2025328296111
Dec 2025396331120
Jan 2026319278115
Feb 2026379330115
Mar 2026441385115
Apr 2026454409111
May 2026561488115
Jun 2026528453117
Jul 2026488444110
Aug 2026526477110
Sep 2026not in the export
Oct 2026471420112

Friday and Saturday hold 37% of all tickets

  • Mon10.9%77
  • Tue11.4%79
  • Wed12.8%90
  • Thu14.5%102
  • Fri18.2%125
  • Sat18.5%129
  • Sun13.6%97
The same 26,036 tickets as the top of the page, re-stacked. Each column is that weekday's share of all tickets. The small number under it is visits per open day, with an average day at 100: Friday and Saturday run 25% and 29% above it, and Monday and Tuesday sit near 78, so that is where the schedule has room.

Pedicures bring in 32% of service sales

  • Pedicures32.1%
  • Gel polish and gel manicures15.6%
  • Acrylic and extensions12.3%
  • Builder gel11.5%
  • Dip powder10.9%
  • Nail art and add-ons9.3%
  • Seasonal specials4.2%
  • Classic manicures2.2%
  • Head spa0.8%
  • Other0.6%
  • Waxing0.5%
Share of service sales by category. Acrylic, dip powder and builder gel together add 35%.

The top technician earns about 35% more per working day than the median

  • Tech A137
  • Tech B125
  • Tech C118
  • Tech D111
  • Tech E107
  • Tech F107
  • Tech G106
  • Tech H106
  • Tech I100
  • Tech J99
  • Tech K93
  • Tech L88
  • Tech M78
  • Tech N77
  • Tech O75
  • Tech P68
  • Tech Q33
Service sales per working day in 2026, measured from the median technician. Technicians with at least 60 working days; names replaced with letters.

The unglamorous part

The register let staff type service and staff names freely. Before any chart could be trusted, I grouped the service names into categories and merged the staff spellings into people.

679service names as typed

11categories, after grouping

81staff spellings

36people, after merging

Product

MaiMói, a marketplace for nail technicians and salon owners

In development

Running a salon showed me a gap in the whole industry. Other industries closed it years ago.

The gap
Vietnamese nail technicians and salon owners still find each other the way they did 20 years ago.
What I use
Expo and React Native, Next.js, Supabase (PostgreSQL), Stripe and AI coding tools.
What I am building
One place to post listings, browse them and talk to each other directly. The work is planned across eight milestones.
Built in from the start
Every salon's data is separated at the database level, and a pgTAP test suite checks those row-level security rules.

One listing, from post to tryout

  1. A salon owner posts

    Vị tríThợ bột, full-time

    Khu vựcNashville, TN

    LươngBao lương

    Đăng tinĐã đăng
  2. A technician browses

    Cần thợ bột, full-timeNashville, TN · bao lương

    Cần thợ chân tay nướcFranklin, TN · part-time

    Cần thợ dip, cuối tuầnMurfreesboro, TN

  3. They talk directly

    Cần thợ bột, full-timeNashville, TN

    Chị ơi, tiệm mình còn cần thợ bột không ạ?

    Còn em. Thứ Bảy em ghé tiệm thử việc nhé.

    Dạ, em cảm ơn chị.

A Nashville owner needs an acrylic technician. By the third message, they have arranged a Saturday tryout. This is a sketch of the flow, in the Vietnamese its users write in. The app is in development, so it is not a screenshot.
AI automation

An AI assistant that answers client messages

Clients message the salon on Instagram and Facebook at all hours. Many of the questions are about services and prices.

The gap
Replies waited until someone at the front desk was free. A client who waits may book somewhere else.
What I used
Python and FastAPI, Meta webhooks, and a language model that looks up answers in the salon's own documents (retrieval-augmented generation).
What I built
A service that receives each message, finds the matching part of the menu and price list, writes the answer and replies in the same thread.
What changed
Clients get an answer at any time of day. Staff stay with the client in the chair.

How one message is answered

  1. MessageA client writes on Instagram or Facebook

    Hi! Do you do dip powder? How much is it?

  2. WebhookMeta sends the event to my FastAPI service

    EventHandled

    Same event againAlready handled

  3. Look upThe service finds the right part of the menu and prices

    Gel polish

    Dip powder

    Builder gel

  4. AnswerThe model writes a reply from that text

    Yes, we do dip powder. It is .

    From the price list: Dip powder

  5. ReplyThe answer goes back in the same thread

    Hi! Yes, we do dip powder. It is .

An illustrative exchange and menu, with the prices left out. Every event is logged and checked against the ones already handled, so a webhook that Meta sends twice never produces two replies. The knowledge base loads from the salon's website and its PDF price lists, so a price change is a document update and needs no new code.
Data engineering

A Snowflake warehouse with Power BI on top

Warehouse built, Power BI report in progress

The dashboard above reads a monthly export. I rebuilt it on a cloud data warehouse. All 71,283 sales lines are loaded, and the totals match the point-of-sale system to the cent.

Three layers, from the raw export to reporting

  1. RawThe POS export, loaded exactly as it arrives
  2. CleanedOne row per sale, with 81 staff spellings mapped to 36 technicians
  3. ReportingViews for daily sales, monthly summary, technician performance and service mix
Before the salon

Data consulting, 2021 to 2023

As an associate consultant at Daugherty Business Solutions (now part of CGI), I worked on four client projects. Client names are confidential.

  • Pet-food companyAutomated recurring reports on a legacy SQL Server Reporting Services system, so marketing and finance got their numbers on schedule.
  • Home-improvement retailerAssessed a data warehouse before its move to the cloud: every table, how the tables connect, and which applications use each one.
  • Financial-planning firmTagged personal data fields and wrote the masking rules for 20,000 customer records, then ran the pipeline that applied them.
  • Electronics retailerBuilt pipelines for a central data warehouse and checked its totals against the old reports.