How to Automate SASSIE

A mystery shopping company sells fieldwork it does not employ

Mystery shopping providers run on SASSIE, built by SurfMerchants and in use since 1998. It holds the client's programme, the survey behind it, the list of locations, the wave of shops with dates attached, the shoppers who claim them, the reports that come back, and the invoice at the end.

The fact that shapes everything else is that the shoppers are independent contractors. Nobody is rostered. A shop is posted and a shopper somewhere decides whether it is worth the drive, at a fee set before anyone knew the weather. A provider does not assign work so much as make it attractive enough to be taken.

So the business is a promise made on one side and volunteered on the other. A client buys a month of coverage across two hundred locations. The company selling it has no employees at any of them.

A wave is only as good as the shops nobody claimed

Most of a wave assigns itself within hours. It is the remainder that consumes the week.

The store in a town two hours from anyone. The visit that has to happen after nine at night. The location a client added on Thursday for a wave that closes on Sunday. Each is worked by hand: a scheduler going through the shopper pool, offering, waiting, offering again, and eventually raising the fee because the deadline is contractual and a missed location has to be explained.

Then the other end of the same week. Reports land at eleven at night on the last day, all of them at once, and every one needs reading before a client sees it. Shoppers who submitted a fortnight ago want paying, and a shopper paid late stops claiming. The client wants to know where the report is. All three arrive in the same afternoon.

SASSIE

The hard shops worked first, not last Locations still unassigned ranked by how far they sit from any shopper who has ever worked that client, so the difficult ones get attention while there is time.

Claimed shops that went quiet Shoppers who took an assignment and then stopped answering, chased on a schedule and the shop released back to the pool early enough for somebody else.

Failures caught before an editor sees them Submissions checked for the things that fail a client review: a missing receipt photo, a visit logged outside the location's posted hours, a narrative under length.

The AutoScheduler fills the shops that already have takers

The platform is strong on the assignment problem, and a provider not running all of it is paying a scheduler to do arithmetic.

The AutoScheduler and SmartMatch AutoScheduling place shops without anyone touching them. Shopper Rotations show each shopper only the dates they are eligible for and enforce rotation by shopper, client, survey or type, which keeps the same person out of the same store too often. B.E.M.S. and the SASSIE Blaster push postings out over dedicated mail servers so an opening reaches the pool quickly. Fraud Busters covers integrity, with GPSVerifier checking a shopper's location and time against where the shop actually was. On the way out, Dragon gives each user a personal dashboard and Report Delivery pushes results to clients who will not come and fetch them.

All of that works the pool it has and the surveys already submitted. The shopper who claimed and went quiet, the payment account, the client's own vendor portal, the recruiting sites where the next hundred shoppers come from, and the location's actual opening hours are outside it.

Coverage gaps are visible days before they become misses

A wave that reports on itself, checked against the world it depends on, means a scheduler stops finding out on the last morning.

Every wave read against its own end date, with unassigned locations ranked by how far they sit from any shopper who has ever worked that client, so the hard ones are worked on day three instead of day twenty-eight. Shoppers who claimed and then stopped answering chased on a schedule, and the shop released back to the pool while somebody else could still take it.

Submissions checked before an editor opens them for what reliably fails a client review: a missing receipt photograph, a visit time outside the location's posted hours, a narrative shorter than the client contracted for. GPS and timestamp anomalies set beside that shopper's history, so a pattern is visible rather than one isolated flag.

Shopper payments read from the payment account and matched against the shops actually approved, because the fastest way to lose a pool is to pay it late. And the client's own portal read for locations added or cancelled after the wave was built, which is usually how a wave goes wrong in the first place.

Editing is judgement, and chasing is not

The scheduler who knows which shopper will actually drive to that town is the scarcest thing in the business, and most of her day goes on finding out who has not replied.

WebRun is an agent that works a real Chrome browser, signed in as you. It opens SASSIE, reads the waves, shops and submissions you point it at, then opens a client's vendor portal, the payment account or a recruiting site, and puts what it found on the list the scheduler works from.

Reading and compiling run unattended on a schedule you set, in your own environment, and a run can be watched and stopped. Anything that reaches a shopper or a client is drafted and waits for a person, because a message to an independent contractor about their pay is not a place to be approximately right.

The workflows below are built already, and each one names what it opens.