Marketing analytics attribution in Birmingham and Liverpool

What to take away

Offline conversions in Birmingham and Liverpool retail campaigns

Birmingham retail runs on a mix of Bullring and Grand Central footfall, suburban high streets and a large logistics and wholesale layer around the M6 corridor. Liverpool is tighter: Liverpool ONE, the retail parks off the dock road, and a city centre catchment that pulls across the Mersey.

Both cities still take a meaningful share of revenue in person or over the phone.

That is the offline conversions problem in one sentence. A customer sees a paid social ad in Erdington, checks a store page, then buys in the Bullring three days later. Nothing in a browser session records it.

The workable approach is to capture an identifier at the point of sale and reconcile it later. That might be a loyalty number, a click and collect reference, a booking email or a phone number on a quote. The identifier is what lets you attach a value to a campaign instead of guessing.

As a hypothetical worked example, not a reported customer case, consider a Birmingham furniture retailer with two showrooms and a Liverpool concession. It tags every quote with a source question at the counter, matches roughly two thirds of showroom sales back to a named channel, and leaves the rest unattributed rather than spreading them evenly.

The unattributed line became its own agenda item each month.

That last point matters. If you force every offline sale into a channel, you are inventing data. A visible unattributed bucket is more honest and more useful than a tidy chart.

For retail volume benchmarks, the ONS retail industry statistics give a national series you can sanity check against before you trust a local spike.

Regional media buying and what it does to attribution windows

Regional media in the West Midlands and the North West is not a single thing. It includes local radio, regional press, out of home on the ring road and the M62 corridor, sponsored content in city titles, and increasingly local creator placements. Each one has a different decay curve.

A paid search click converts in minutes or hours. A radio spot heard on the school run may convert in a shop the following weekend. If both sit in one model with a single seven day window, search will look brilliant and radio will look useless.

This is where attribution windows stop being a technical setting and become a commercial argument. Set the window per channel, state it in the report, and keep it stable for at least a quarter.

Changing windows mid flight makes month on month comparisons meaningless, which is one of the common attribution measurement mistakes teams make when they tidy a model.

A practical way to compare is to hold one definition of a conversion and vary only the window, then watch which channels move. That approach is set out in more detail in comparing channels on common definitions, and it avoids the trap of treating a correlation as a cause.

For regional performance context, the ONS output and productivity statistics are a reasonable benchmark for how a local market is actually trading, rather than how your dashboard says it is.

Tools, media and local campaign examples

Use Google Analytics 4 for web events and campaign parameters, with consistent naming for QR codes, paid media and event follow-up. For paid Google Ads activity, import eligible offline conversions only where the required identifiers and consent permit; test the data flow and keep unmatched records visible. Adobe Analytics can report channel classifications and conversion events, but it cannot resolve untracked store visits by itself.

In HubSpot CRM or Salesforce Sales Cloud, keep the campaign ID, source, venue, interaction timestamp, consent basis and lead or opportunity association. Join records only with permitted identifiers, check sample records from capture through to sale, and leave unmatched records unattributed.

For regional media, check current local availability with BBC Radio WM, BBC Radio Merseyside, Birmingham Live, Liverpool Echo and JCDecaux UK for out-of-home inventory. Treat these as potential placements to verify, not as evidence that a campaign ran or reached a particular audience.

Proposed campaign labels—not claims of past campaigns—make local tests specific:

For local market context, check published data and statistics from Birmingham City Council, Liverpool City Council, West Midlands Combined Authority and Liverpool City Region Combined Authority, alongside the ONS series already noted. Check each source's geography and reporting period, and use these figures as context rather than as proof that a campaign caused a change.

Events-led demand in two named English cities

Birmingham and Liverpool both carry a heavy events calendar. The NEC, Utilita Arena and the ICC sit alongside the M&S Bank Arena, Exhibition Centre Liverpool and a dense schedule of conferences, trade shows and match days. Events-led demand is what happens when attendance, sponsorship or a stand at one of these creates pipeline that closes later.

Events rarely convert on the day. A visitor scans a badge, takes a card, or has a conversation, and the deal lands four to ten weeks afterwards. If your attribution window closes in seven days, the event will look like a cost centre forever.

There are three workable patterns:

As a hypothetical example, not a reported company case, a Liverpool software firm running a stand at a city centre conference captures 240 conversations, matches 38 to opportunities within a quarter, and finds that the strongest predictor of a closed deal is a follow-up call within five working days, not the size of the stand.

That finding changed next year's budget more than any channel score would have.

Setting up multi-channel attribution to reflect offline and events

A multi-channel attribution setup for these two cities needs to be built in a fixed order. Skipping a step usually means rebuilding the model later.

That sequence is close to an attribution workflow that begins with a decision ticket, and it is worth reading before you buy anything.

On the reporting side, keep the model output separate from the raw counts. A well designed attribution dashboard should show both, so a reader can see the model's view and the untouched numbers side by side. When they diverge sharply, that is the signal to investigate, not to hide the raw line.

Data protection shapes all of this. The ICO's data protection principles guidance sets out the accuracy and minimisation duties that apply when you store identifiers and match offline records. UK GDPR and the Data Protection Act 2018 apply here, and PECR governs much of the electronic marketing that feeds the top of the funnel.

Evidence fields to hold for each offline conversion

Every offline conversion that enters the model should carry the same fields. Without them, a join is just a guess with a timestamp.

Two fields get skipped most often: match confidence and consent basis. Both matter. Match confidence tells you how much weight a record deserves. Consent basis tells you whether you may hold it at all, and for how long.

Store these fields once, in one place, and let the model read from them. If each channel team keeps its own version, the multi-channel attribution setup will drift within a quarter and nobody will trust the output.

Common questions