Notes from the machinery

The most expensive part of a platform migration is usually invisible on the invoice: the organic traffic that quietly disappears when old URLs die without forwarding addresses. Years of accumulated rankings can be spent in a weekend by a launch that treated redirects as a tidy-up task.

Map before you move

Every product, category, collection and content URL on the old site needs a decision: keep the path, change it deliberately, or retire it with a redirect to the closest living equivalent. That mapping is a spreadsheet job done early — export the old sitemap, crawl the live site, and record where each address will point after launch. The pages that earn most of your search traffic deserve individual attention rather than pattern-matching.

Redirect with intent

Permanent redirects should go live at the moment of cut-over, not in the days after. Chains (old page → interim page → new page) leak value and slow crawling, so point each old URL directly at its final destination. Resist the temptation to send everything to the homepage; a redirect that ignores intent is only slightly better than an error page.

Watch the aftermath

After launch, the work is observation: crawl for broken links, watch search-console reports for spikes in "not found" errors, and compare landing-page traffic against the pre-launch baseline for a few weeks. Most post-migration traffic wobbles settle if redirects are right; the ones that don't will tell you exactly which mappings to fix.

Handled this way, replatforming is a controlled move between buildings — not a fire in the archive.

When baskets are abandoned, the instinct is to redesign — new buttons, fewer fields, a fresher look. Worth doing, sometimes. But watch real sessions and a plainer pattern emerges: people leave when the checkout surprises them, or when it withholds something they need to feel safe.

The expensive surprises

Delivery cost revealed at the final step. A delivery date that never appears at all. A forced account before the order total. A discount code field that implies everyone else is paying less. Each is an information failure — the customer asked a reasonable question and the checkout answered late, vaguely or not at all.

Answer earlier

The fixes are mostly editorial rather than structural: show delivery costs and timescales on the product page and in the basket; state returns terms where the decision is being made; keep the order summary visible and honest throughout; let guests buy and offer the account after the thank-you page. Reassurance elements — clear contact routes, recognisable payment marks, plain-English policies — work because they answer the quiet question "what happens if this goes wrong?"

Then measure like you mean it

Instrument each step so you can see where people actually leave, on which devices. Mobile deserves particular suspicion: a form that's mildly annoying on a laptop can be genuinely hostile on a phone. Improve, measure, repeat — and be wary of anyone who promises a guaranteed uplift before they've seen your data.

Churn gets discussed as a persuasion problem — better boxes, better emails, better offers. But split churn into its two species and the priorities change. Voluntary churn is a customer deciding to leave. Involuntary churn is a customer who intended to stay, cancelled by an expired card, a declined transaction or a bank's fraud filter.

Why involuntary churn comes first

It's the cheapest churn to address, because there's no mind to change. The customer wanted the product; the machinery dropped them. Every recovered failure is a full subscription saved at the cost of a retry schedule and a well-timed message.

The machinery that does the saving

Intelligent retries spaced over days, not three attempts in an hour. Card-update prompts before expiry, not after failure. A pre-renewal notice that doubles as a nudge to fix payment details. Account-updater services where the provider supports them. And a grace period that pauses rather than cancels, so one bad transaction doesn't end a two-year relationship.

Then earn the voluntary stays

With the leaks fixed, self-service does the next shift: customers who can pause, swap or skip tend to do that instead of cancelling, because the alternative you offer is easier than leaving. Report the two churn species separately — averaging them hides exactly the distinction that tells you what to fix next.

Businesses account for stock to the penny while their product information — the words, numbers and images that actually sell the stock — lives in duplicated spreadsheets with filenames like products_FINAL_v7_USE_THIS. The result is familiar: pages that contradict the datasheet, filters that miss half the range, and channel listings rebuilt by hand every season.

What "governed" looks like

One agreed home for product truth. Defined attributes with owners — who decides a dimension, who approves copy, who signs off imagery. Variant logic designed rather than accreted. Import workflows that validate before they publish. None of this requires an enterprise PIM on day one; it requires the decision that product data is infrastructure.

The compounding return

Governed catalogues pay out everywhere downstream: search and filtering improve because attributes are consistent; channel feeds become exports rather than projects; support tickets fall because pages answer the questions; and replatforming — when it comes — becomes a data transfer instead of an archaeology dig. The catalogue is the one system every other commerce system depends on. Treat it accordingly.

Trade ordering by email survives because it feels personal and costs nothing to start. What it actually costs shows up elsewhere: office hours spent re-keying attachments, pricing disputes rooted in stale PDFs, orders that miss the warehouse cut-off because they sat unread, and your best customers rationed by your inbox's opening hours.

Why portals fail — and how they don't

Trade buyers reject portals that treat them like consumers: retail checkout flows, no account pricing, no bulk entry, minimums enforced by apology email. A portal earns adoption when it mirrors the trading relationship — each account sees its own prices and terms; bulk and quick-order entry beats browsing; quotes and purchase orders have a proper home; credit limits behave like the agreement, not a surprise.

Make the better way the easier way

Adoption is a migration, not an announcement. Seed each account with its order history so reordering is one click. Keep the phone and email channels open, but let the portal be faster — live stock, instant confirmation, invoices on demand at 6am. When the convenient path and the preferred path are the same path, the inbox retires itself.