DeductSafe

1. One-liner

Reconciles a Kenyan SME’s expenses against KRA’s eTIMS data and flags every deduction about to be disallowed before filing.

2. Trend signal — why now?

On January 10, 2026 KRA switched on its Income and Expense Validation Engine. It now cross-checks every figure in a tax return against real-time eTIMS invoice data at the moment of filing. The rule that bites: any expense not backed by a valid eTIMS invoice linked to your KRA PIN is automatically added back to taxable income. Cash receipts, handwritten acknowledgements, and “M-Pesa + WhatsApp screenshot” — the way most Kenyan SMEs have always documented purchases — no longer count.

The kicker is that your supplier’s non-compliance becomes your tax bill. If you bought Sh500,000 of goods from a supplier who never transmitted the invoice on eTIMS with your PIN, that Sh500,000 is disallowed even though you paid and recorded it. KRA’s first enforcement run flagged 392,162 individuals and firms and identified Ksh 759.7 billion in estimated unpaid liabilities. The 2025-income filing deadline is June 30, 2026 — so the first mass collision between messy SME ledgers and the validation engine is happening right now.

Every accounting firm in Nairobi is giving the same advice: “Download your eTIMS Purchase Report from iTax monthly and reconcile it line-by-line against your internal expense ledger.” That reconciliation is manual, brutal, and recurring — and nobody has tooled it for the small end of the market.

Provenance:

• Signal 1 (Demand): KRA Validation Engine live 10 Jan 2026; expenses without matching eTIMS auto-disallowed; accountants tell SMEs to reconcile the eTIMS Purchase Report against their expense ledger monthly — https://adamjeeauditors.com/e-tims-2026-kra-expense-validation-business/ — 2026-06-25
• Signal 2 (Feasibility): Cash/M-Pesa/WhatsApp documentation “no longer satisfies eTIMS”; messy informal records must now be matched to structured eTIMS data — AI parsing + fuzzy reconciliation makes this cheap — https://www.techinkenya.com/articles/kra-is-now-checking-every-figure-you-file-in-real-time-here-is-what-that-me — 2026-06-25
• Signal 3 (Economic): ERP/accounting vendors racing into eTIMS compliance (Veira, ClearTax KE, Cute Profit POS, Zynobooks); penalty regime now automated (min Ksh 100,000); 392,162 entities flagged, Ksh 759.7B in liabilities identified — https://www.pna.co.ke/kra-validation-income-expenses-effective-1-jan-2026/ — 2026-06-25 Category: Regulatory arbitrage

3. The opportunity

KRA gives away the invoice-creation side for free (eTIMS Lite app, USSD *222#). So don’t rebuild that. The unsolved, recurring, painful job is the buyer side: proving every shilling of claimed expense has a matching eTIMS record before you file — and fixing the gaps in time.

Incumbents are wrong-shaped for this. Full ERPs (the ones marketing “eTIMS-compliant” badges) cost real money, assume a clean accounting system, and target businesses that already have a bookkeeper. The bottom of the market — the duka, the contractor, the small distributor, the salon, the M-Pesa-till business — keeps records in a notebook, a chat, and an M-Pesa statement. For them the validation engine is a trap: legitimate expenses silently disappear at filing and the tax bill jumps. A focused tool that ingests their actual mess (M-Pesa statement export, photographed receipts, WhatsApp supplier chats), pulls the eTIMS Purchase Report, and produces a gap list with a one-tap supplier-chase action does the one job they can’t do by hand and an ERP won’t do for them.

4. Target market

• Primary customer: Kenyan SMEs filing income tax with non-trivial purchases — small distributors, contractors, retailers, salons, eateries, hardware/agrovet shops — typically Ksh 5M–50M turnover, with 0–1 bookkeeper. Secondary: the bookkeepers and small accounting firms who file on their behalf and now own the reconciliation headache.
• Why they buy: “If I claim Ksh 500,000 in expenses but eTIMS only shows Ksh 320,000 linked to my PIN, KRA adds back Ksh 180,000 and recalculates my tax with interest.” They will pay to not get that surprise. The pain is annual at minimum, monthly if they reconcile properly.
• Rough TAM reasoning: Kenya has well over a million registered businesses and KRA flagged 392,162 entities in a single enforcement run. Even 50,000 paying SMEs at a modest annual price is a large business by bootstrap standards.
• Why now for them: The engine went live Jan 2026; the first filing under the new regime is due June 30, 2026. The pain just became real and unavoidable, not theoretical.

5. Product sketch (MVP)

• Pull the truth: one-time connect to KRA iTax / eTIMS to fetch the monthly Purchase Report (the list of invoices suppliers actually transmitted against the buyer’s PIN).
• Ingest the mess: upload an M-Pesa statement (PDF/CSV), photograph paper receipts, forward WhatsApp supplier messages — AI extracts supplier, amount, date, item into a structured expense ledger.
• Reconcile: fuzzy-match each recorded expense to an eTIMS purchase entry (names, amounts, and dates never line up cleanly) and produce a clean three-bucket view: Matched, Mismatch (amount/date off), No eTIMS record = at risk of disallowance.
• Quantify the exposure: a running “deductibility risk” number — total expenses about to be added back, and the extra tax that implies — so the owner sees the cost before filing.
• Chase the supplier: for each at-risk expense, auto-draft a WhatsApp message in plain Kiswahili/English telling the supplier exactly what to transmit (the buyer’s PIN, the invoice, the amount) to fix it.
• File-ready pack: a month-end / year-end reconciliation report the owner or their accountant can use to file with confidence, plus a flagged list of expenses to drop rather than risk.

6. AI angle — what’s load-bearing

The hard, load-bearing work is turning informal records into a structured ledger and matching it against eTIMS — two messy datasets that never agree on names, amounts, or timing. AI does the OCR on photographed receipts, parses free-text M-Pesa and WhatsApp lines (“paid Mama Njeri 12k for stock”), normalises supplier identities, and runs the fuzzy reconciliation. It also drafts the per-supplier chase messages with the exact missing fields. Strip the AI out and you’re left with a manual spreadsheet reconciliation — which is precisely the unpaid, error-prone job the customer is drowning in today. The AI is the product, not a chatbot bolted to a form.

7. Localization angle

This is a localization play, not a global one. The wedge is the geography: KRA’s specific eTIMS rules, the iTax Purchase Report format, M-Pesa as the dominant payment rail, WhatsApp as the dominant supplier-communication channel, Kiswahili/English chase messages, and KES micro-pricing. A generic “expense reconciliation” SaaS built for QuickBooks-land does not touch any of this. The same shape ports later to other African e-invoicing regimes (Nigeria FIRS, Tanzania, Uganda EFRIS) and to Philippines EIS — but Kenya is the beachhead because the validation engine is live and biting now.

8. Business model — path to $1M–$5M ARR

• Pricing: Two tiers. SME self-serve: Ksh 1,500/mo ($11) for monthly reconciliation of one business. Accountant/bookkeeper: Ksh 6,000–12,000/mo ($45–90) to manage 10–40 client books from one dashboard. Annual-filing-only micro plan at Ksh 3,500 ($27) one-time for the June rush.
• ACV: Blended ~$150–250/yr, skewed up by the accountant tier (which is the real engine).
• Rough math to $1M ARR: ~600 accounting practices at ~$70/mo ($840/yr) ≈ $500K, plus ~3,000 self-serve SMEs at ~$130/yr ≈ $400K → ~$900K. Achievable in-country.
• Rough math to $5M ARR: broaden to ~3,000 practices (each managing 20+ books) + 15,000 self-serve SMEs, and add per-filing-season surge revenue. Requires being the default reconciliation layer accountants reach for, plus expansion into one adjacent market (Nigeria or Tanzania).
• Expansion path: seats per accounting firm grow with their client book; add VAT input-credit reconciliation, supplier-compliance scoring (“which of my suppliers keep breaking my deductions”), and a year-round bookkeeping upsell.

9. Go-to-market wedge — first 100 customers

• Accountants first, not SMEs. They feel the pain across dozens of clients and they’re findable. Scrape ICPAK (Institute of Certified Public Accountants of Kenya) member directories and the dozens of Nairobi/Mombasa/Kisumu small-firm listings; send a personalised WhatsApp/email showing a reconciliation of a sample messy ledger against eTIMS. One firm = 10–40 books.
• Ride the eTIMS-panic content wave. Every Kenyan tax blog (Adamjee, ClearTax KE, Veira, Tuko, Pulse) is publishing “your return will be rejected” explainers. Pitch a free “eTIMS deduction-gap check” lead magnet and get featured / guest-posted where the worried owners already are.
• WhatsApp + Facebook groups. Kenyan SME and bookkeeper groups are large and active (NTV ran a segment: “most people are worried about how they’ll file”). Drop a free single-month gap check; convert the ones who see real money at risk.
• Filing-deadline blitz. The June 30 and subsequent quarterly windows are forced buying moments. Run a “reconcile before you file” campaign timed to each deadline.

10. Build complexity — justification

Medium. Receipt OCR, statement parsing, and fuzzy matching are off-the-shelf AI plus standard web stack — a pair can ship a credible v1 in 10–14 weeks. The non-trivial work is the KRA iTax / eTIMS integration to fetch the Purchase Report reliably (portal scraping or whatever official API access exists, plus handling auth and format drift) and getting the reconciliation accurate enough that accountants trust it. M-Pesa statement parsing and WhatsApp messaging are well-trodden in the Kenyan dev ecosystem.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Reads the SME’s own data with consent; helps them comply. No grey area.
Ethical — no harm / dark patterns	✅	Reduces accidental over- or under-payment of tax; helps small suppliers get formalised.
Market exists (evidence above)	✅	Live regulation, automated penalties, 392K entities flagged, accountants already doing this by hand.
1–5 person team can build this	✅	2 builders, ~3 months to v1.
Launchable with <$50K / ₹40L	✅	API/inference + WhatsApp costs only; no capital intensity.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Hair-on-fire: lost deductions = a direct, automatic, dated tax hit. Felt at every filing.
Demand evidence	15	13/15	Multiple independent signals: live engine, automated penalties, accountant guidance, public worry, vendor land-grab.
Build feasibility	15	11/15	Core is off-the-shelf AI; the iTax/eTIMS data pull is the gnarly, fragile part.
Distribution clarity	15	11/15	Accountant directories + deadline-timed campaigns are concrete; conversion math is estimated, not proven.
Revenue mechanics	15	11/15	Pricing fits KES wallets; the accountant tier carries it, but ARPU is low so volume must be real.
Time to first revenue	10	7/10	Pain is acute and timed to filing windows; expect paying accountants within 4–8 weeks of a working demo.
Defensibility	10	5/10	Moat is the eTIMS-integration reliability + accountant workflow lock-in + supplier-compliance data that compounds. Copyable, but a focused head start wins.
Total	100	75/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (fragile government-portal integration + reconciliation accuracy) · domain-expertise-required (Kenyan tax/eTIMS rules; ideally a co-founder or close advisor who is a Kenyan CPA).

Key assumptions to validate (3–5)

1. Assumption: The eTIMS Purchase Report can be pulled programmatically (or via reliable portal automation) per taxpayer with consent. How to test: Build the fetch against 5 real consenting SME accounts and confirm it returns complete, current data.
2. Assumption: Accountants will pay ~$45–90/mo to manage client reconciliations here rather than in Excel. How to test: 25 discovery calls with ICPAK small-firm members; show a sample reconciliation; ask for a paid pilot commitment.
3. Assumption: Fuzzy matching of M-Pesa/receipt records to eTIMS entries hits accuracy high enough to be trusted (low false “matched”). How to test: Reconcile 10 real months of SME data by hand vs. the tool; measure precision/recall.
4. Assumption: SMEs perceive the disallowance risk in money terms strongly enough to pay before (not after) a painful filing. How to test: Run the free single-month gap check on 50 SMEs and measure conversion to paid.

Risk flags

1. Platform dependency: The whole product hinges on access to iTax/eTIMS data. If KRA blocks scraping or changes formats without an API, the data pull breaks. Mitigate by pursuing any official integration path and designing for format drift.
2. Regulatory churn: KRA has already extended/changed eTIMS rules repeatedly; thresholds (e.g. the Ksh 5M exemption) and validation behaviour could shift, changing who needs this.
3. Low ARPU / collections: KES micro-pricing means volume and retention must be real, and Kenyan SME churn/payment-collection is harder than US SaaS. The accountant tier is the de-risking lever — anchor there.
4. KRA builds it themselves: KRA could add a buyer-side reconciliation view to iTax. Possible, but government UX shipping speed is the moat; the supplier-chase + messy-record ingestion is the part they won’t do.

14. Structured verdict

Score:                  75/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder + Kenyan CPA advisor/co-founder
Time to revenue:        6–10 weeks (paid accountant pilots, timed to filing windows)
Capital to launch:      $5–10K (KES 0.7–1.3M) — inference, WhatsApp, hosting
Top 3 assumptions to validate first:
  1. eTIMS Purchase Report can be fetched reliably per consenting taxpayer — build against 5 real accounts
  2. Accountants pay $45–90/mo for this — 25 ICPAK discovery calls, ask for paid pilot
  3. Reconciliation accuracy is trustworthy — hand-check 10 real months, measure precision/recall
Kill criteria:
  - Abandon if the eTIMS Purchase Report cannot be obtained programmatically or via reliable automation for everyday SMEs
  - Abandon if <5 of 25 accountant calls commit to a paid pilot
  - Abandon if fuzzy-match precision stays below the level accountants will trust (high false-matched rate) after 6 weeks of tuning

15. Next step — 1-week validation sprint

• Day 1–2: Confirm the data spine. Manually pull the eTIMS Purchase Report for 3 friendly SME accounts and a real M-Pesa statement; reconcile by hand to prove the matching logic and quantify a real “at-risk” number.
• Day 3–4: Build a one-screen demo (upload statement → see gap list + tax-at-risk) using a sample dataset. Run 15–20 WhatsApp/calls to ICPAK small-firm accountants showing it.
• Day 5: Go/no-go. Go only if ≥5 accountants say “yes, I’d pay for a pilot this filing season” and the hand reconciliation shows the tool would have caught real disallowed expenses. Falsifiable: no paid-pilot intent, or no real gaps found = no-go.