AI Repair Assistant

You have the broken clip in one hand and your phone in the other. You search for “small white drawer thing”. Helpful.

Repair involves several separate jobs: identifying the item, understanding the fault, finding a compatible part and installing it. A useful answer to the first question can still leave you stuck on the third.

This isn’t a claim that spare parts don’t exist. IKEA already offers small spare parts free of charge in the UK. iFixit provides repair guides and sells parts. The gap worth testing is the awkward case: an unidentified fitting, an older model or a part whose name bears no resemblance to what a normal person would type.

Start with owners and furniture refurbishers trying to save a worthwhile item by replacing a non-safety-critical knob, trim cover or cosmetic fitting. A £20 search service makes more sense beside a piece worth keeping than a £15 flat-pack table.

Build a repair assistant that follows the job through to a verified fit.

1. Identify the item. Collect photos of the furniture, broken part, model label and attachment point. Ask what happened and which function stopped working.
2. Resolve uncertainty. Request another angle, measurements and a ruler in the photograph. Suggest a likely match with supporting evidence. A picture alone is not proof of compatibility.
3. Find the best route. Check manufacturer stock first, then compatible and second-hand parts. Show dimensions, supplier, total delivered cost and the evidence for fit. Recommend a free manufacturer route when available.
4. Explore fabrication where appropriate. If no suitable part exists, ask a maker to review a simple replacement. The maker checks the design, material and intended use before quoting. AI producing a plausible shape is not a completed engineering check.
5. Complete the repair. Provide item-specific fitting steps where suitable, ask whether the part fitted and save the confirmed outcome. Refund or correct a failed match under clearly agreed terms.

The concept demo illustrates that journey using sample results. It does not perform live diagnosis, check supplier inventory or process orders.

Keep the first offer narrow. No electrical, gas, structural, load-bearing or other safety-critical repairs. A chair clip can carry a load even if it looks trivial, so assess function rather than appearance.

These are proposed test assumptions, not traction, market estimates or a forecast. Use one currency consistently in the pilot; the examples below use GBP.

Test £19 per parts-finding job, with parts and delivery paid separately. Define what the buyer receives and what triggers a refund before taking payment.

That is roughly 53% contribution, not net profit. If research takes 30 minutes, labour rises to £12 and contribution falls to £4. An £8 acquisition cost would then turn the first job negative. Time spent searching is the business model hiding in plain sight.

At 100 completed jobs a month, fees would generate £1,900 revenue and £1,000 contribution under the first scenario. At 1,000 jobs, those figures become £19,000 and £10,000, but manual delivery alone requires 250 hours. That volume would need staff, better matching or both. It is not software margin by default.

For a small initial service, a bottom-up target is more useful than a giant repair-market statistic: 20 repeat buyers sending five jobs a month produces 100 monthly jobs. Whether refurbishers actually have that workload, and will pay to outsource it, is what interviews must establish.

The workflow can be tested cheaply. A form, a shared case tracker and manual searches can establish whether people pay for a correct part before a founder builds image recognition.

Repair also has a policy tailwind. The EU’s repair directive promotes repair for products within its specified scope. That is context, not a claim that every furniture manufacturer must supply every discontinued fitting, or that regulation proves demand for Fixsmith.

Our hypothesis is that visual assistance can make the intake less painful, while a growing library of confirmed matches makes each subsequent search faster. Both need testing. An impressive diagnosis screen means very little if the parcel contains the wrong part.

Start with a paid concierge. Charge per clearly scoped search rather than asking a household to subscribe to breaking things. Offer a free eligibility check so unsuitable jobs never enter the paid queue.

Keep parts costs separate. Let customers buy directly from the supplier during the pilot. Any referral commission should be disclosed and excluded from the core economics until agreed and measured. Rank by suitability and total cost.

Quote custom work separately. Design review, fabrication, shipping and revision time can overwhelm a £19 fee. A later test could add a fixed coordination fee to an approved maker quote. Establish responsibilities, design rights and refund terms first.

Explore repeat buyers next. Furniture refurbishers and second-hand dealers could buy bundles if their repeated search work justifies them. A proposed £99 five-job pack is a pricing experiment, not a validated offer. Track jobs actually used and time spent before adding unlimited plans.

Manufacturer support should often win. IKEA’s spare-parts service demonstrates why Fixsmith must check the original route first. Charging someone to obscure a free solution is a poor foundation for trust.

iFixit and repair guides help people understand repairs and obtain parts. Fixsmith would need to earn its fee through identifying an obscure component and establishing compatibility, rather than rewriting an available guide.

Search engines, resale marketplaces and a patient friend are the DIY alternative. The customer is buying time and confidence. If a normal search solves the job in two minutes, there is little reason to pay.

Local repair shops and makers bring practical judgement and fabrication skills. They may be partners as well as alternatives. Test whether better intake and sourcing saves them work.

The possible advantage is a database of confirmed part-to-product matches: model, dimensions, attachment details, supplier and real fit outcome. Collect it with permission and respect source and design rights. A folder of AI guesses is not that database.

Find the first ten buyers where the broken parts already are. Interview five refurbishers and ten owners with an unresolved repair. Ask to see the object, their search history, quotes and what they would do if they cannot fix it. Avoid asking only whether they like the idea.

Recruit through local refurbishment businesses, repair communities and relevant groups where promotion is permitted. Offer to assess one specific current problem. Useful before-and-after examples, with permission, are stronger than broad “AI for your home” ads.

Run a 30-day experiment:

• Week 1: Select one category of non-critical furniture fittings. Write eligibility and refund rules, set up the intake form and identify supplier and maker routes.
• Week 2: Recruit ten paid pilot jobs at £19. Track every enquiry, including rejected and unpaid cases, so conversion has a denominator.
• Week 3: Deliver manually. Record search minutes, confidence, supplier response, delivered cost, delivery time and corrections. Follow up after the part arrives.
• Week 4: Verify installation outcomes. Ask repeat buyers for another paid job and document why unsuccessful customers dropped out.

Suggested learning gates: at least seven of ten paid cases end in a customer-confirmed fit; median research and support time stays at or below 15 minutes; at least three repeat-capable buyers place a second paid job. These are founder-set thresholds, not industry benchmarks. Report pending deliveries separately; a link found is not a repair completed.

If customers pay but searches take too long, narrow the category further. If they praise it but will not pay, change the customer or offer before building more screens. If fit quality fails, fix verification before acquisition.

A confident wrong match. Require dimensions, model evidence and human review before promising fit. Record corrections and repeat failures.

Small orders, expensive support. Measure all research, messages, returns and rework. A ten-minute search followed by an hour resolving a return is a seventy-minute job.

Fabrication creates new responsibilities. Material, tolerances, wear and loads matter. Refer unsuitable work to qualified professionals; exclude categories the pilot cannot assess properly.

Supplier availability changes. Date checks, show provenance and avoid guaranteeing stock you do not control. Do not base the service on copying catalogues without permission.

Infrequent consumer demand. Validate contribution on the first purchase. Repeat business from trade buyers would be valuable, but cannot be assumed.

The service is useful but hard to scale. Automate intake, duplicate searches and known matches first. Keep human judgement where a mistake causes costly rework.

We’d test Fixsmith as a parts-finding service for the frustrating repairs people already want to finish. The initial promise is concrete: send us the broken fitting and the details, and we’ll help establish a replacement route.

The first milestone is ten paying customers and a record of parts that actually fitted. If those jobs leave enough margin, turn the repeated work into software. Save the furniture first. Build the clever bit around what worked.