AI on the parts of a campaign that actually take time
Choosing the site mix, pricing a face, spotting a recycled proof photo, ordering a crew's route, reading a vendor invoice. Each one is a named feature with an input you can see and an output you can correct.
- Planning
- budget to site mix
- Proof
- image checking
- Ops
- routing & alerts
AI features that decide where the budget goes
The part of the work a client actually pays for.
Site-mix planner
Give it a budget and the localities you want covered and it returns a mix of faces and screens, with the coverage it buys and the reason each site is in the list.
Brief to draft plan
Write the brief the way you'd say it — budget, cities, dates, objective — and get a draft campaign back with every field editable.
Natural-language inventory search
“Lit hoardings under ₹80k near Kukatpally, free in November” becomes a set of filters you can see and adjust — not an answer you have to trust.
Rate recommendation
A suggested rate for a face, with the confidence band and the comparable sites it drew from shown next to it.
Occupancy forecasting
How much of an estate is likely to sit empty next month, with intervals, so a media owner can discount early instead of late.
Modelled audience targeting
Sites ranked by audience modelled from traffic and location data — labelled as modelled, and built without a single camera pointed at a person.
AI features that check the evidence
These run on every capture before anyone opens it.
Location and time anomaly checking
Was this really taken at that face, at that time? A trust score with the specific anomalies named rather than a pass or a fail.
Duplicate and recycled-image detection
Perceptual hashing and image matching across the whole account, so a photo reused from another campaign is caught.
Creative verification
Is the poster up, is it ours, and is it intact? A verdict with reason codes and a confidence attached to it.
Site code reading
Reads the site plate from a framed, cropped shot and fills the field in. A full street scene is not attempted, because that does not work reliably.
SLA breach prediction
A ranked list of the campaigns about to miss their proof deadline, and who to nudge first.
Dispute triage
Reads a complaint, pulls the evidence already sitting on the booking, and proposes a resolution for someone to approve.
The AI features nobody demos, and everybody misses
They save hours a week rather than winning a pitch.
The order a crew should ride today across the sites assigned to them, accounting for capacity and the material they are carrying.
A shortlist of who should get a job, weighted by proximity, current load and past quality — not whoever answers first.
A finished campaign becomes a client-ready summary in the language the client reads, edited by a person before it is sent.
A selected site list becomes a branded draft proposal. Specified and coming — it is not shipped yet, and this page will say so until it is.
A supplier invoice pre-fills the payable, with anything uncertain queued as an exception rather than guessed.
Which invoices are about to go bad, and the order to chase them in.
Identity documents are read to save typing during onboarding. Approval is always a person's decision, and the record keeps whose.
Every one of them proposes. You dispose.
Four rules that apply to every AI feature on this page.
Named, not sprinkled
Each feature does one job you can describe in a sentence. There is no general 'AI mode' doing something unspecified in the background.
Shows its confidence
An output the model is unsure about is presented as unsure, on the screen where the decision gets made.
A person decides
Nothing is approved, rejected, paid or billed by a model. Suggestions arrive with their reasoning; people act on them.
No cameras on people
Audiences are modelled from traffic and location data. There is no camera-based demographic measurement and no facial recognition.
- One job per feature
- Confidence shown
- Human decides
- Multilingual output
The limits, before you ask
Every line below is a real ceiling. Designing around them beats apologising for them later.
Automated image checking is nowhere near perfect. Routine captures clear themselves; flagged ones go to a person. That is the design, not a gap we are closing.
Benchmark scores belong to the benchmark. When we publish an accuracy figure it will be ours, measured on our own held-out data, carrying the date it was measured.
Natural language maps to filters you can see and correct. A question becomes a search you can check, never an answer you have to believe.
Every uplift percentage published in this category came without a sample, a period or a control. We would rather show you the plan and the proof.
Time-of-day delivery follows the configured schedule. Weather and air-quality-triggered creative changes are not currently available.
AI on the slow parts, people on the decisions
Try the planner on a real brief
Tell us how you plan today and we'll run one of your real briefs through the site-mix planner in your own workspace.